Copyright Permission Note: Permission of the Publisher has been given to the International Food Policy Research Institute (IFPRI) to post this book to its web site as allowed under the copyright law of the United States (Title 17, United States Code) which governs "fair use" or the making of photocopies or other reproductions of copyrighted material. Photocopy or reproduction is not to be “used for any purpose other than private study, scholarship, or research.” The correct citation for this article is: Alston, Julian M.; Norton, George, W.; and Pardey, Philip G. 1995. Science under scarcity: Principles and practice for agricultural research evaluation and priority setting. London UK and Ithaca, NY: Cornell University Press for the International Service for National Agricultural Research (ISNAR). Reprinted with the permission of Cornell University Press and CAB International. The book was reissued in paperback in 1998 by CAB International. Thank you in advance for respecting these conditions. For additional information, contact IFPRI- library@cgiar.org . Science under Scarcity Food Systems and Agrarian Change Edited by Frederick H. Buttel, Billie R. DeWalt, and Per Pinstrup-Andersen A complete list of titles in the series appears at the end of this book. SCIENCE UNDER SCARCITY Principles and Practice for Agricultural Research Evaluation and Priority Setting Julian M. Alston George W. Norton Philip G. Pardey Published in cooperation with the International Service for National Agricultural Research Cornell University Press ITHACA AND LONDON This book was printed from camera-ready pages provided by the authors. Copyright © 1995 by the International Service for National Agricultural Research All rights reserved. Except for brief quotations in a review, this book, or parts thereof, must not be reproduced in any form without permission in writing from the publisher. For information, address Cornell University Press, Sage House, 512 East State Street, Ithaca, New York 14850. First published 1995 by Cornell University Press. Printed in the United States of America e The paper in this book meets the minimum requirements of the American National Standard for Information Sciences- Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984. Library of Congress Cataloging-in-Publication Data Alston, Julian M. Science under scarcity: principles and practice for agricultural research evaluation and priority setting / Julian M. Alston, George W. Norton, Philip G. Pardey. p. cm. - (Food systems and agrarian change) "Published in cooperation with the International Service for National Agricultural Research." Includes bibliographical references (p. ) and indexes. ISBN 0-8014-2937-4 1. Agriculture-Research-Evaluation. 2. Agriculture-Research-Cost effectiveness. I. Norton, George W. II. Pardey, Philip G. III. International Service for National Agricultural Research. IV. Title. V. Series. S540.E92A58 1995 338.4'363072-dc20 94-40637 To Cameron, Deborah, John, Louisa, Marj, Michelle, Sydney, and Toby Contents List of Illustrations . xiii List of Tables. . . xvii Foreword Vernon W. Ruttan xix Preface ..... . xxi Introduction . . . . xxix Intended Audience xxx Topical Outline . xxxi Part I Institutional and Conceptual Framework The Institutional, Scientific, and Policy Contexts 3 1.1 Institutional Setting . . . . . . 4 1.1.1 Executing Agencies. . . 4 1.1.2 Structure within Agencies 5 1.1.3 Types of Decisions . . 6 1.1.4 Funding Arrangements. . 7 1.2 Scientific Context ..... . 8 1.2.1 Basic, Applied, and Adaptive Research 8 1.2.2 Research and Technology-Transfer Linkages . 9 1.2.3 Dynamics . . . . . . . . . . . . . . . 10 1.3 The Policy Context . . . . . . . . . . . . . . 11 1.3.1 The Economic Justification for Government Intervention 12 1.3.2 Research as an Instrument of Social Policy. 14 1.3.3 Political Economy Perspective on the Demand for Research . . . . . . . . . 16 vii viii Contents 1.3.4 Roles for Social Science Research . 17 2 Research Evaluation and Priority-Setting Principles 19 2.1 Investing in Research and Technical Change 21 2.1.1 Relating Research, Knowledge, and Production . 22 2.1.2 Economic Consequences of Agricultural Research . 27 2.1.3 Other Issues - Uncertainty and Adjustment Costs 34 2.2 Measuring Benefits and Costs Using Economic Surplus Concepts . . . . . . . . . . . . .. 40 2.2.1 Basics of Economic Surplus Measures . . . .. 41 2.2.2 Criticisms of Economic Surplus as a Welfare Measure 43 2.2.3 "Alternatives" to Economic Surplus Analysis. 54 2.3 Determinants of the Size and Distribution of Benefits and Costs . . . . . . 57 2.3.1 Critical Assumptions in the Model . 58 2.3.2 Extensions to the Basic Model. . . 65 2.4 Economy-Wide (General-Equilibrium) Implications of Research . . . . . . . . . . . . . 78 2.4.1 Distinguishing between Partial- and General-Equilibrium Models . . . 78 2.4.2 Practical Approaches for Research Evaluation 79 2.5 Reconciling Multiple Objectives of Research. . . . 80 2.5.1 Economic Efficiency . . . . . . . . . . 81 2.5.2 Equity (Income Distribution) and Security Objectives 82 2.5.3 Trading off Multiple Objectives . . . . . 87 2.6 Conclusions and Discussion 92 Part II Measuring the Effects of Agricultural Research 3 Econometric Measurement of the Effects of Research 97 3.1 Conceptual Models of Production, Productivity, and Technical Change . 99 3.l.l Parametric Approaches 102 3.1.2 Nonparametric Approaches 116 3.1.3 Index-Number Approaches 120 3.2 Specification and Measurement Issues 142 3.2.1 Primal Models 143 3.2.2 Dual Models . 146 3.2.3 Single-Equation Supply Models 150 3.2.4 Output and Input Data . 153 3.2.5 Research and Extension Variables 167 3.2.6 Statistical and Econometric Issues 188 3.3 Calculating the Effects of Research 191 Contents ix 3.3.1 Growth Accounting . . . 191 3.3.2 Research-Benefit Streams 193 4 Economic Surplus Methods ..... 207 4.1 The Basic Model . . . . . . . 208 4.1.1 Surplus Distribution in the Basic Model 208 4.1.2 Disaggregating Benefits and Costs . . 210 4.2 Horizontal Market Relationships ..... 212 4.2.1 Multiple Markets for a Single Product. 212 4.2.2 Disaggregating Consumer and Producer Surplus 228 4.2.3 Multiple Products: Some General Issues. . 230 4.2.4 Multiple Products Related in Consumption 237 4.2.5 Multiple Products Related in Production. 240 4.2.6 Demand Shifts . . . . . . . . . . . . 243 4.3 Vertical Market Relationships ........ 246 4.3.1 Two Factors with Fixed Factor Proportions 246 4.3.2 Two Factors with Variable Factor Proportions 251 4.3.3 Research Benefits with Input Substitution . . 256 4.3.4 Models with More Than Two Factors of Production 264 4.4 Market-Distorting Policies and Research Benefits. 266 4.4.1 Closed-Economy Examples. . 270 4.4.2 The Small-Country Trader Case . . . . . 275 4.4.3 Large-Country Trader Models. . . . . . 284 4.4.4 Overvalued or Undervalued Exchange Rates 291 4.5 Sustainability Issues and Other Externalities 293 4.5.1 Research Benefits in the Presence of Environmental Externalities. . . . 294 4.5.2 Resource Depletion, Intergenerational Equity, and Agricultural Research . . . . . .. . 297 4.6 Conclusion.............. 298 Part III Evaluation and Priority Setting in Practice 5 Economic Surplus Measurement and Application . . . . . .. 303 5.1 Defining the Problem . . . . . . . . . . . . . .. 305 5.1.1 Clients for the Analysis - Decisions to Be Served 305 5.1.2 The Objectives of the Analysis - Terms of Reference 306 5.1.3 The Scope of the Analysis - Research Programs and Program Alternatives. . . . . . . . . . 307 5.1.4 Objectives for the Research System - Measures of Benefits .............. 309 5.1.5 Strategy for the Analysis - Degree of Detail 311 5.2 Market-Related Data . . . . 314 5.2.1 Price and Quantity Data . . . . . . . . . 316 x Contents 5.2.2 Elasticities. . . . . . . . . . . . . . . . 319 5.2.3 Discount Rate and "Exogenous" Growth Factors 324 5.3 Measuring the Research-Induced Supply Shift 326 5.3.1 Conceptual Issues. . . . . . . . . . . . . 328 5.3.2 Practical Measurement. . . . . . . . . . . 332 5.3.3 Research Risk and Lags in Research, Development, and Adoption . . . . . . . . . . . . . .. 349 5.4 Application - Analyzing and Using the Data and the Results. . . . . . . . . . . . 361 5.4.1 Calculating the Streams of Research Benefits and Costs . . . . . . . 361 5.4.2 Capital Budgeting. . . . . . . . 362 S.4.3 Calculating Other Consequences of Research. 364 5.4.4 Variance of the Research Portfolio and Sensitivity Analysis . . . . . . . . 36S S.4.5 Using the Results in Decision Making. 369 5.S Conclusion....... 377 5.5.1 Reality Check . . . . . . . . 377 S.5.2 Achieving a Balance ..... 379 Appendix AS.l Computing Research Benefits 380 A5.1.l A Spreadsheet Approach 380 AS.1.2 The Dream© Approach. . . 386 Appendix A5.2 Selected Formulas for Calculating Research Benefits ........... 39S AS.2.1 Simplified, Two-Country Model 395 A5.2.2 Alternative Formulas for a Small, Open, Distorted Economy. . . . . . 401 Appendix A5.3 Estimating K Using Industry and Experiment Data . . . . . . . . . . . . . . . " 411 AS.3.1 A Two-Factor Equilibrium-Displacement Model 411 AS.3.2 Summary of Algebraic Results . . . . 416 Appendix AS.4 Data for Estimating the Supply-Shifting Effects of Research . . . . . . . . . 419 A5.4.1 Elicitation Form Cover Sheet. . . . 424 AS.4.2 Research Resources . . . . . . . 42S AS.4.3 Research Impact - Estimating kMAX . 429 AS.4.4 Research Dynamics 433 AS.4.S Research Risk. 438 AS.4.6 Reconciliation. 439 6 Mathematical Programming 441 6.1 Mathematical-Programming Principles 443 Contents xi 6.1.1 Basics of Mathematical-Programming Models 443 6.1.2 Formulations for Research Resource Allocation . 450 6.2 Mathematical Programming in Practice . . . . . . 456 6.2.1 Model Design . . . . . . . . . . . . . 457 6.2.2 Compiling Data and Calculating Coefficients . 460 6.2.3 Running the Model . . . 460 6.3 Conclusion......... 462 7 Scoring and Other Shortcut Approaches. 463 7.1 Scoring ......... . 465 7.1.1 Common Practice versus Basic Principles 465 7.1.2 Defining a Simple Scoring Model 472 7.1.3 Implementing a Scoring Model 482 7.2 Other Shortcut Procedures . . . . . 487 7.2.1 Rules of Thumb and Guidelines 487 7.3 Conclusion........... 492 Appendix A7.1: The Problem of Units in Eliciting Weights 494 Part IV Overview and Assessment 8 Assessment and Conclusion 501 8.1 Conceptual Framework Revisited . . . . . 502 8.2 Deciding on the Method and Degree of Detail 503 8.2.1 Methods for Ex Post Evaluation of Research Programs 504 8.2.2 Methods for Ex Ante Research Evaluation 506 8.2.3 Setting Priorities . . . . . . . . . . . . 507 8.2.4 Selecting Projects or Experiments . . . . . 508 8.3 Areas for Future Model Development and Application 509 8.4 Conclusion 511 References. . 515 Author Index . 553 Subject Index . 561 Illustrations 2.1 Gross annual research benefits . . . . . . . 28 2.2 Research, development, and adoption lags 30 2.3 Net research benefits over time ..... 30 2.4 Producer and consumer surplus measures . 41 2.5 Accuracy of consumer surplus . . . . . . . 46 2.6 Implicit assumptions in cost-benefit analysis 55 2.7 Effects of elasticities on distribution of benefits . 60 2.8 A proportional supply shift in a constant-elasticity model 62 2.9 A parallel shift down of a "constant-elasticity" supply function . . . . . . . . . . . . . . . . . . . 63 2.10 Schematics of multimarket effects of R&D . . . . . 70 2.11 A trade-off of equity and efficiency using research policy alone . . . . . . . . . . . . . . . . . . . . . 89 2.12 A trade-off of equity and efficiency using the least-cost policy combination . . . . . . . . . . . . . . . . . 91 3.1 Economic relationships between supply functions 103 3.2 Technical change and input substitution effects. . 122 3.3 Technical change and output substitution effects 124 3.4 Biased technical change, homothetic technologies 138 3.5 Biased technical change, nonhomothetic technologies 140 3.6 Finite research lag structures. . . . . . . . . . . . 180 4.1 Surplus distribution in the basic model of research benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 xiii xiv Illustrations 4.2 Size and distribution of research benefits for a traded good (exporter innovates, no technology spillovers, large country) . . . . . . . . . . . . . . . . . . . . . . . . . . 215 4.3 Size and distribution of research benefits for a traded good (importer innovates, no technology spillovers, large country) . . . . . . . . . . . . . . . . . . . . . . . . . . 218 4.4 Size and distribution of research benefits for a traded good (exporter innovates, with technology spillovers, large country) . . . . . . . . . . . . . . . . . . 220 4.5 Research benefits in a small open economy . . . . . 227 4.6 Welfare effects with feedback in consumption ... 240 4.7 Effects of exogenous demand shifts on the size and distribution of research benefits . . . . . . . . . . . 246 4.8 Research benefits with two factors in fixed proportions . 248 4.9 Biased technical change with fixed proportions . . . . . 252 4.10 The Muth equilibrium-displacement model . . . . . . . 257 4.11 Research benefits in a closed economy with a target price and deficiency payments . . . . . . . . . . . . . . . . . 271 4.12 Research benefits in a closed economy with a maximum price ceiling. . . . . . . . . . . . . . . . . . . . . 272 4.13 Research benefits in a closed economy with a per unit subsidy. . . . . . . . . . . . . . . . . . . . . . . . .. 274 4.14 Research benefits in a closed economy with an output quota 276 4.15 Research benefits in a small open economy with price supports (or output subsidies) ................ 277 4.16 Research benefits in a small open economy with maximum domestic prices . . . . . . . . . . . . . . . . . . . . 279 4.17 Research benefits in a small importing country with a tariff. . . . . . . . . . . . . . . . . . . . . . . . . 281 4.18 Research benefits in a small country with an import quota. 282 4.19 Research benefits in a small country with a revenue-pooling price-discrimination scheme . . . . . . . . . . . . . . . .. 284 4.20 Research benefits in a large country with a target price and deficiency payment . . . . . . . . . . . . . . . . . . . 287 4.21 Research benefits for a large-country exporter with an export tax . . . . . . . . . . . . . . . . . . . . . . . . 288 4.22 Research benefits for a large-country importer with an import tariff . . . . . . . . . . . . . . . . . . . . . . 290 4.23 Research benefits in a small country with a negative production externality ..................... 295 Illustrations xv 5.1 Components of research-induced supply shifts . . . . . . . . 330 5.2 Presumed probability distribution of experimental yield gains ........................... 352 5.3 Triangular probability distribution of experimental yield gains ..................... 354 5.4 The triangular probability distribution . . . . . . . 367 5.5 Assessing the trade-off between equity and efficiency 374 A5.4.1 Benchmarking elicited yield effects for a single research option. . . . . . . . . . . . . 430 A5.4.2 Trapezoidal adoption profile . . . . . . . . . . . . . 435 A5.4.3 Logistic adoption profile . . . . . . . . . . . . . . . 437 A5.4.4 Example of a triangular probability density function 438 6.1 Illustration of a set of noninferior solutions generated by parameterizing weights on the objective functions ... 445 6.2 Trade-off between objectives and illustration of the ideal point and compromise set ................. 448 6.3 Single-objective, linear-programming model for allocating resources to agricultural research .............. 451 6.4 Multiple-objective, linear-programming model for allocating resources to agricultural research ............... 454 Tables 4.1 Notation used in the Muth model ......... 255 4.2 Solutions to the Muth model . . . . . . . . . . . . 261 4.3 Incidence of benefits from technical change in the Muth model ..................... ..... 263 5.1 An illustrative list of commodity and noncommodity research programs .................. 308 5.2 An illustrative list of agricultural research objectives 310 A5.1.1 Sample spreadsheet to evaluate research in a small open economy .................... . . . . . 381 A5.1.2 Sample spreadsheet to evaluate research in a closed economy ....................... 384 A5.2.1 Selected formulas for computing changes in economic surpluses . . . . . . . . . . . . . . . . . . . . . 407 A5.3.1 Consequences of new technology for yields and supply shifts ......................... 417 A5.3.2 Sufficient conditions under which yield changes accurately reflect supply shifts. . . . . . . . . . . . . . . . . 418 A5.4.1 Example of an elicitation form cover sheet . . . . 424 A5.4.2 Research resources: Human resource information by commodity ......................... 425 A5.4.3 Research resources: Human resource information by research program area . . . . . . . . . . . . . . . . . . . . . . . .. 426 A5.4.4 Total resources and disposition ............... 428 A5.4.5 Sample data sheet for recording the yield effects of various research options ....................... 430 xvii xviii Tables A5.4.6 Sample data sheet for recording changes in input costs 431 A5.4.7 Sample data sheet for eliciting R&D lags for research program areas. . . . . . . . . . . . . . . . . . . . . . 434 A5.4.8 Sample data sheet for adoption parameters for the trapezoidal adoption profile . . . . . . . . . . . . . . . . . . . . . . . 436 A5.4.9 Sample data sheet for adoption parameters for the logistic adoption profile. . . . . . . . . . . . . . . . . . . . . . . 437 7.1 Mapping stated objectives into measurable objectives .. 473 7.2 Sample questionnaire for eliciting weights on agricultural research objectives . . . . . . . . . . . . . . . . . . . . . 475 A7 .1 The effects of units in scoring . . . . . . . . . . . . . . . 495 A7.2 Use of the scoring model to determine agricultural research priorities by commodity . . . . . . . . . . . . . . . . . .. 497 A7.3 Disaggregating the scoring model to include both commodities and research program areas . . . . . . . . . . . 498 Foreword This book represents the culmination of a research agenda that extends back to the late 1940s. The agenda was initiated by a group of scholars at the U.S. Department of Agriculture and the University of Chicago who began to explore quantitatively the sources of growth in agricultural production. This initial research revealed that the growth of agricultural production in the United States could only be partially explained by growth in conventional inputs - in land, labor, capital, and operating inputs - as traditionally measured. But what was the source of the newly discovered growth in the productivity of the factors used in agricultural production? An initial step was to associate productivity growth with technical change and to attribute technical change to agricultural research. This set off an extended debate that lasted well into the late 1960s about the relative significance of technical change embodied in physical inputs as well as new knowledge arising out of education and experience, embodied in the human agent. By the early 1970s the methodology that had been developed to measure rates of return to agricultural research and to human capital was being extended to the process of research planning and to allocating research resources. Alternatives to the more intuitive approaches, such as scoring methods or the congruence between research expenditures and the value of production, were being ex­ plored. The growth of international support for agricultural research in develop­ ing countries added impetus to this effort in the 1970s. And the decline in research support in the more constrained economic environment of the 1980s represented one additional source of demand for more effective research plan­ ning and management. This volume by Alston, Norton and Pardey is the culmination of both the rate xix xx Foreword of return and the research resource allocation studies. It is impressive both in its comprehensiveness and in its depth. It will hardly be possible in the future to contribute to either research or practice in the fields of research evaluation and research resource allocation without referring to this book. VERNON W. RUTIAN Regents Professor University of Minnesota Preface Resources for agricultural science are scarce. Worldwide, public agricul­ tural research systems are being asked to do more with less. As government budgets tighten generally, agricultural research administrators face ever­ sharper pressures to justify their budgets and to economize. Increasingly, research administrators are being asked to provide evidence that the costs of their operations are justified by the benefits. At the same time, the demands placed on agricultural science are also changing. Agricultural science is being asked to contribute to knowledge and technology and to satisfy demands for information on many new issues - environmental problems, food safety and quality, and rural development - without foresaking traditional work. Closer scrutiny and tighter resources imply a greater need for evaluation of public-sector research and for economically effective allocation of re­ sources to it. All research is planned and evaluated to some degree. The relevant questions are how much planning and evaluating to do, who should do it, and what form it should take. Some say there should be little formal planning and evaluating - that too much planning and evaluation can stifle the generation of new ideas, the heart of any research program. Others say planning and evaluation are necessary for accountability in the use of scarce public funds. The former group would argue that relatively unstructured planning and evaluation in the past have produced results with high rates of return: "If it ain't broke don't fix it." The latter group point to the slowness of research systems to adjust to the changing needs of society and to the realities of scarce public resources. It is important to distinguish the economic problem of research priority setting from the related scientific, technical, and management issues that arise in implementing priorities, getting the research done, and getting the xxi xxii Preface results adopted. It is tempting for many, in the pursuit of accountability, to take the process of evaluation and priority setting too far. Our view is that processes for planning and evaluating research can be helpful at every stage in the research system but that structured quantitative methods are most beneficial for making strategic decisions when research priorities are being set across broad commodity programs, disciplinary (and multidisciplinary) programs, and research problems. The formal analytical apparatus and priority-setting approaches that are developed and described in this book are most useful when they are applied at an aggregative, program level. They are less useful at a detailed, disaggre­ gated, project level for at least three reasons. First, the costs of fine-tuning might not justify the benefits in terms of improved allocation of resources to research. Second, measurement problems become increasingly important as the degree of disaggregation increases. I Third, micromanaging creative endeavors such as research can be counterproductive; more detailed alloca­ tion decisions are probably best guided by well-structured incentive systems instead of interventionist, "hands-on" allocative mechanisms. The last is perhaps the most important consideration. Formal evaluation and priority­ setting procedures should not be used as a basis for replacing ingenuity, serendipity, and scientific entrepreneurship with bureaucratic procedures. There is a wealth of informal evidence that a successful research program rests heavily on the spirit, imagination, judgment, and integrity of agricul­ tural scientists who are allowed freedom of enquiry. The role of research evaluation and priority setting is to help determine the boundaries within which free scientific enquiry occurs. Once decisions have been made about the numbers and types of scientists to employ and broad parameters have been placed on operating budgets, highly structured evaluation and priority-setting procedures can reduce the efficiency of the research system. A balance must be struck between the use of formal evaluation and priority-setting procedures and informal alterna­ tives at any level of decision-making. The trick is to ensure that the relevant economizing principles are involved in decisions regarding resource alloca­ tion, without overly managing individual scientists. Personal and profes­ sional incentives for scientists must be built into the system so that they respond to the demands of clients, generate new ideas, and produce high­ quality products. Occasional quantitative, economic, and social evaluations I. A major difficulty when assessing research impact at ~ project level involves apportioning observed or predicted changes in yields or reductions in unit costs to research-induced changes in particular components of a technology package while /widing other components oftha1 technology unchanged. Given the interrelationships inherent in many new technological packages, it is likely that spurious attribution, double counting, or both will result. Preface xxiii of individual projects or programs of research can be useful, but not whole­ sale, costly, quantitative evaluation of all potential and completed research projects. Some say that research evaluation and priority setting should be left to the scientists themselves. However, scientific merit alone is not sufficient to justify maintaining budgetary support for research or for setting strategic research priorities. Research administrators recognize that assessing the social value of research is useful for justifying budgets and is essential for making strategic decisions on research investments. Without economic analysis, it is difficult to assess the social value of scientific knowledge or new technologies and to make informed judgments about the trade-offs in allocating scarce scientific resources. This means there is a growing role for economists in research evaluation and priority setting, because biological scientists are no more capable of providing reliable answers to questions of economic value than economists are of evaluating the scientific potential of biological experiments. The roles of economists and other scientists are complementary. Scientists' opinions are needed to help define the possibilities of advancing knowledge or providing new technologies and information if resources are allocated to particular programs. The input of economic analysts is useful in estimating the eco­ nomic value of the research, including the distribution of benefits and costs, and in advising decision makers on procedures for incorporating economic principles when setting priorities and allocating research resources. Econo­ mists have made significant progress in developing methods for research evaluation and priority setting, but many research analysts and administra­ tors do not have a working knowledge or appreciation of them. Science under Scarcity has been written, in response to the demand from research admin­ istrators and the analysts working for them, to fill that knowledge gap. The title Science under Scarcity was chosen partly to convey the view that the increased current scarcity of resources for agricultural science implies an enhanced demand for methods that will help research administrators justify continuing budgetary support and economize within the constraints of their budgets. Of course, research resources are always scarce, and there is always potential for economizing in the allocation of those resources, but many budgets for agricultural research are increasingly tight and the likelihood of impact is greater at a time when people are being pressed to look for alternatives. If we are successful in assisting research administrators in making better decisions on research resource allocation, there is another sense in which scarcity might be alleviated through better information. In a world in which more than a third of the population live in poverty, the title Science under xxiv Preface Scarcity graphically expresses the important role that agricultural science has always played as the mainspring of economic progress in developing economies, and as an engine for lifting the technological constraints that limit the capacity of the global food and fiber system to produce more within ever-tighter natural resource constraints. We hope that by helping agricul­ tural scientists make better choices in their research resource allocations, this book can contribute in an economically meaningful way to a reduction in the problems that are associated with global scarcity of other, nonresearch resources. Objectives We were motivated to begin writing this book in the mid-1980s, when the International Service for National Agricultural Research (ISNAR) was faced with ever-increasing demands by national agricultural research systems (NARSs) to provide workable research-evaluation and priority-setting proce­ dures. A substantial body of literature had developed on ex post research evaluation, and a more disparate, gray literature on priority setting was emerging. We had been involved in several studies ourselves and were aware of the need for cost-effective and practical methods for evaluating and prioritizing research, but we were also conscious of the importance of having methods that could be defended at a conceptual level as being consistent with the relevant economic theory. Priority-setting methods that do not appeal to a consistent conceptual framework are likely to be ad hoc and to produce recommendations that are difficult to defend and, therefore, more easily dismissed by decision makers. An integrated treatment of research evalua­ tion and priority-setting procedures was needed, one that would relate procedures to theory and provide guidance for when and how to apply particular methods. The overriding goal of this book is to lower the cost of implementing conceptually sound research evaluation and priority-setting procedures. Re­ search administrators want procedures that • are cost-effective • can incorporate multiple research programs (defined by commodity, problem area, or spatial focus) • can assess trade-offs among multiple objectives Our premise in writing this book is that any method or procedure adopted should draw from a consistent conceptual framework. A range of approaches have been suggested and used, but they have not always been theoretically sound or consistent or appropriate. The specific objectives with respect to Preface xxv research evaluation and priority-setting methods are • to place them in a policy and scientific context • to describe their key theoretical or conceptual elements • to review, synthesize, and assess alternative procedures • to provide insights into issues associated with implementing these procedures To accomplish these objectives, we review appropriate theory, assess the literature, and draw on previous experience in implementing research eval­ uation and priority-setting procedures. Procedures are assessed with respect to their consistency with economic theory, their ease of implementation, and their appropriateness for the problem at hand. Previous Work The methods and lines of enquiry described in this volume trace back several decades. Willis Peterson, Vernon Ruttan, Burt Sundquist, and others at the University of Minnesota have addressed issues of research policy and evaluation in both developed and developing countries for many years. Vernon Ruttan's (1982) Agricultural Research Policy book raised many of the issues that we attempt to deal with in Science under Scarcity. The Minnesota work on research policy and evaluation had intellectual ties to the Uni versity of Chicago where studies by Schultz (1953a) and Griliches (1958, 1964) spurred a flurry of activity by graduate students such as Robert Evenson, Willis Peterson, and others, who evaluated the economic impact of agricultural research and extension. Work on research evaluation and prior­ ity setting spawned several conferences, beginning with one at Minnesota in 1969 that resulted in the book edited by Fishel (1971), Resource Allocation in Agricultural Research. An Airlie House Conference in 1975 led to the book edited by Arndt, Dalrymple and Ruttan (1977), Resource Allocation and Productivity,. that took an international focus. An interregional project (IR-6) funded by the U.S. Department of Agriculture from 1979 to 1991 resulted in three sets of symposium proceedings (Norton et al. 1981; Sund­ quist 1987, 1991) and numerous papers on research evaluation. Since the early 1970s, there has also been a broadening of the base of work in this area. Several major studies have been carried out in Australia, supported by the Australian government (including the Industries Assistance Commission report on Rural Research in Australia in 1976 and the mono­ graph by Edwards and Freebairn on Measuring a Country's Gains from Research in 1981), and a large number of smaller studies have dealt with particular institutions or industries. Similar developments have taken place xxvi Preface in Canada and the U.K., with major studies reported by Klein and Furtan (1985) and Thirtle and Ruttan (1987) and a host of monographs and journal articles. In relation to less-developed countries, economists working at or for the international agricultural research centers, particularly CIAT, CIMMYT, ICRISAT, IRRI, and ISNAR, have been involved in numerous efforts to evaluate the effects of research and assist with setting priorities for research in their own centers or for their clients. Beginning in the mid-1980s, the Australian agency ACIAR supported a series of research priority-setting projects in several Asian NARSs and also undertook an ex ante evaluation study at the global level (Davis, Oram and Ryan 1987). In addition many studies have been conducted by developing-country governments on their own behalf. There have been several reviews of parts of this literature, including Schuh and Tollini (1979), Norton and Davis (1981), Greig (1981), Scobie (1984), Parton, Anderson and Makeham (1984), Fox (1987), Scobie and Jardine (1988), and Schultz (1990). Most of the ideas in this book were borrowed from this general literature. It would take too much space to list all of the people to whom we owe an intellectual debt or to enunciate the size of the debt in any detail. We hope that those who read this book will think that it does a fair job of communi­ cating the ideas that have been developed by the work of those who have inspired us. Acknowledgments We have drawn on the patience, counsel, guidance, and collaboration of a good many people in preparing this book. We would especially like to thank Jock Anderson, Barbara Craig, Howard Elliott, John Freebairn, Grant Scobie, and Vince Smith, as well as Jim Chalfant, Matt Dagg, Cesar Falconi, James Houck, John Mullen, Vernon Ruttan, and Helio Tollini for reviewing and commenting on all or major parts of the manuscript. During the course of writing this book we benefited from the advice of Randy Barker, Doug Beach, Fred Buttel, Derek Byerlee, Jeff Davis, Klaus Deininger, Paul Driscoll, Wilhelmina Eveleens, Shenggen Fan, Bob Herdt, Bob Lindner, Luis Macagno, Will Martin, Alex McCalla, Bruce McCarl, and Brad Mills, and our NARS colleagues in several country studies, especially Modan Dey, Patricio Espinosa, Gustavo Ferreira, Melania Lima, Wilfredo Moscoso, Julio Palomino, and Stanley Wood. The institutional support we received from ISNAR, the University of California at Davis, the University of Minnesota, Virginia Tech, Cornell University, and the University of Saskatchewan is also gratefully acknowledged. Preface xxvii We are particularly thankful to Fionnuala Hawes who gave generously of her good humor, word processing, and desk-top publishing skills, to Kath­ leen Sheridan for outstanding editorial work, and to Nienke Beintema for proofing the manuscript: groetjes and bedankt. The final production assis­ tance of Leta Kelley and the design inputs of Richard Claase are also gratefully acknowledged. JULIAN M. ALSTON Professor University of California at Davis GEORGE W. NORTON Senior Fellow International Service for National Agricultural Research Professor Virginia Polytechnic Institute and State University PHILIP G. PARDEY Senior Research Officer International Service for National Agricultural Research Associate Professor University of Minnesota Introduction Governments have to decide what resources to make available for public­ sector research. In turn, research administrators have to allocate resources across research problems, programs, people, and places. In order to make these decisions effectively, decision makers have to evaluate alternatives and set priorities. This book reviews, synthesizes, and assesses research evalua­ tion and priority-setting procedures. The term research evaluation, as used in the book, refers to assessing the economic effects of research.1 The value of research evaluation is both as a means of accounting for the effectiveness of past research investments (i.e., ex post evaluation) and, looking forward, as a basis for setting priorities and allocating research resources (i.e., ex ante evaluation). Applying the principles and procedures identified in this book will provide new knowledge and help decision makers achieve their objectives. Even when the procedures described here are not adopted explicitly, incorporating the underlying economic way of thinking about research investments will help structure decision-making processes and improve their outcome. At the same time, the principles and practices described here should not substitute for the best judgments of scientists and policymakers in the research-priority-setting process. They provide a unifying framework within which to synthesize a wide range of scientific and economic data that would otherwise be difficult to reconcile and use. Thus, good judgments can be made even better, and poor judgments may be exposed. The evaluation procedures outlined in this book enable the various productivity-related, distributional, and environmental consequences of re- I. See Horton et al. (1993) for alternative approaches to research evaluation. xxix xxx Introduction search to be reported using a comparable money measure. By so doing, they can reveal research opportunities and consequences that at first sight may not be so obvious. The systematic approaches to research evaluation and priority setting described here pay due regard to the economic context in which research is funded, conducted, and adopted without abandoning the scientific basis of the research process. This book provides a set of basic principles and procedures that are differentiated primarily by their degree of detail and complexity, and it gives guidance to the application of methods judged most appropriate for different representative situations. Intended Audience The primary audience for this book are analysts working for agricultural research systems who have at least Master's-level training in economics. Most agricultural research systems have access to such expertise. Parts of the book should be of direct interest to research administrators as well, although much of the material (especially in chapters 3, 4, and 6) is fairly technical. Where possible, we have put the more difficult material in footnotes or appendices. However, the bulk of the book is targeted toward economists who will be carrying out and interpreting research evaluation or priority-set­ ting analyses for administrators. Some degree of technical sophistication is inevitable in a book of this kind. In presenting these evaluation methods and priority-setting procedures, we have been sensitive to the data and resource constraints that commonly confront analysts and decision makers in less-de­ veloped countries. Nevertheless, most of the material presented here is also directly applicable in a developed-country context. The principles and practices described in this book are also relevant to those concerned with research evaluation and priority-setting problems in sciences beyond agriculture and beyond the public sector. While the primary focus is public-sector agricultural research, the approaches described here are equally applicable (or with small modifications) to nonagricultural re­ search and to some private-sector situations. In particular, the ideas and methods in this book apply directly to the allocation of public-sector R&D resources between agricultural and nonagricultural research. Also, most of the evaluation and priority-setting methods are directly applicable to the decisions about research undertaken by producer organizations (e.g., using funds collected by a levy) and could be extended to consider choices about funding R&D versus product promotion and so on. Science under Scarcity is most likely to be used as a reference book by economists doing applied studies on the economics of research. For that Introduction xxxi reason considerable use has been made of headings and subheadings, and an extensive subject index is appended. The book can also be used as a supplementary text in graduate-level courses on agricultural development, applied production economics, and welfare economics. A basic understand­ ing of economic principles is required for most of the chapters, and for some, competence in statistics and mathematics is needed. A companion set of training materials for research evaluation and priority setting has been produced at the International Service for National Agricultural Research (ISNAR), including additional details to facilitate the application of the suggested methods. An interactive computer program Dream©, developed at ISNAR, that applies the economic surplus methods described in this book is included in the training materials. Topical Outline This book contains four major sections. In the first section, the institu­ tional framework and conceptual issues associated with research~Yllluation and priority setting are addressed. Chapter 1 considers the institutional, scientific, and policy contexts of agricultural research. It discusses the influence of the type of organizational structure and the scientific context for research, on research evaluation and priority-setting procedures. It addresses research as an instrument of social policy and discusses the reasons for public-sector involvement in research and the need for research to improve institutions as well as technologies. Chapter 2 presents the theoretical and conceptual issues in research evaluation and priority setting in a relatively nontechnical fashion. It intro­ duces the concept of a research production function and the key factors influencing the economic consequences of research. The concept of eco­ nomic surplus as a welfare measure is fundamental to the work in this book. Therefore, in chapter 2, the foundations of economic surplus measures are described, and criticisms of economic surplus as a welfare measure are reviewed and evaluated. Critical assumptions in the economic surplus model are also discussed and evaluated. Then we introduce a range of extensions of the model. In the second major section, econometric and economic surplus measures are reviewed and synthesized. This section is significantly more technically demanding for the reader than the others; it contains the theoretical and conceptual underpinnings for all of the approaches being discussed in the other sections of the book. Chapter 3 presents parametric, nonparametric, and index-number approaches to the measurement of agricultural productiv- xxxii Introduction ity. Primal and dual approaches to measuring the structure of production and (research-induced) technical change are considered, and measures of output, conventional inputs, and research and extension variables are detailed. Sta­ tistical issues related to specification error, multicollinearity, and simultane­ ity are briefly reviewed. Chapter 3 concludes with a discussion of the use of results from econometric models for measuring the effects of research directly. Chapter 4 presents a comprehensive treatment of the measurement of research-induced changes in economic surplus. Horizontal and vertical mar­ ket relationships are modeled, in large part to enable the benefits from research to be disaggregated among different groups of producers, consum­ ers, or input suppliers. Such disaggregations are important when there is an interest in the impact on particular groups (such as domestic producers and consumers for a traded good, or suppliers of labor). It is suggested that this multimarket approach can also be used to explore the impact of research-in­ duced quality change. Also in this chapter, the effects of market-distorting policies on the size and distribution of benefits are given some attention. Sustainability and other externalities are discussed briefly. Several of the more common variants of economic surplus models are presented in graph­ ical and mathematical forms, and a fairly general method for calculating benefits from research is described. In the third major section, implementation issues associated with research evaluation and priority-setting methods are presented. Chapter 5 builds directly on chapter 4 and focuses on the measurement of economic surplus with an emphasis on practical issues and using the results to make decisions. It addresses the different types of information required for different types of decisions and the implications of varying the degree of detail. Practical approaches are suggested for defining objectives, obtaining basic data and other information needs, developing the information base (including tables and questionnaires), data processing, and presenting and interpreting results. Step-by-step procedures for implementing the evaluation methods provided in the previous chapter are included in an appendix. Chapter 6 identifies mathematical-programming methods that can be used for analyzing the trade-offs involved in allocating scarce research resources across competing programs. It discusses the rationale for applying these optimizing methods and the modeling options available, and it identifies requirements for making these procedures operational. Chapter 7 presents the scoring approach that has been used as a shortcut method for evaluating alternatives and trading off multiple objectives. We review the use of scoring models in practice and contrast them with the methods described in chapters 5 and 6 in terms of consistency with the basic principles presented in earlier chapters. To con- Introduction xxxiii clude this chapter, we provide practical guidance on shortcut procedures and rules of thumb that can be applied when a more formal analysis is not warranted. The fourth and final major section of the book provides an overview and assessment of research-evaluation and priority-setting procedures. Chapter 8 gives a perspective on when each type of method is likely to be most useful. It highlights some of the difficult methodological issues that are yet to be resolved, suggests potentially fruitful areas of future model development, and concludes the book. The following conventions of style and nomenclature were adopted in the presentation of mathematical symbols in the text in general, but some exceptions were unavoidable (e.g., the use of s to denote shares of producer revenue and S to denote shares of consumer cost): Convention Usedfor Example 1. base font functions and operators natural logarithms In functions f(.) relative change in X (dX/X) E(X) 2. italics all variables 3. upper-case italics scalar variables output price p output quantity Q variable input price w variable input quantity X fixed factors Z relative supply shift down in the price direction K relative shift of supply to the right J relative reduction in producer price Z 4. lower-case italics rate of change of a scalar variable z absolute supply shift down in the price direction k 5. bold italics vectors and matrices Z 6. Greek letters parameters supply elasticities demand elasticities (usually absolute values) 1'\ elasticities of substitution (1 percentage tax (negative subsidy) 't per unit tax (negative subsidy) T technology index 't first difference operator ~ 7. italic t time index Part I Institutional and Conceptual Framework 1 The Institutional, Scientific, and Policy Contexts Worldwide, public-sector agricultural research is big business. By the mid-1980s public-sector agricultural research systems were spending $9.2 billion (1980 dollars) annually: $4.4 billion in developing countries and $4.8 billion in developed countries (Anderson, Pardey and Roseboom 1994). This global investment in public agricultural research represents a 2.6-fold in­ crease in real (i.e., inflation-adjusted) terms over the amount that was invested just two decades earlier. I It may be big business, but "business as usual" may not be sustainable. Questions are being increasingly asked about public-sector agricultural re­ search: How much should be spent? How should it be spent? Who should spend it? Who should pay for it? What is the role for the private sector? How can the resources be used more effectively? The questions being asked are similar around the world, but the answers might differ depending on the institutional, political, and scientific environment in which public-sector agricultural research is being conducted. For instance, in most countries agricultural research is funded and administered independently from tertiary education, but in the United States and India, for example, they are intimately connected. Even among countries with similar institutional arrangements, there can be considerable differences in the objectives of the agricultural research system as well as important constraints imposed by other policies or other aspects of the economic and cultural environment. I. Notably, the public research system in developed countries grew by 2.2-fold while the systems in developing countries spent 3.4 times more in the mid-1980s than they did in the early 1960s. 3 4 The Institutional, Scientific, and Policy Contexts In this chapter we consider the implications of three aspects of the agricultural research environment. First, there is the "institutional setting" for research: the nature of funding arrangements, the legal and other consid­ erations, the general objectives, the organizational culture, and the system of incentives and rewards. Section 1.1 discusses the implications of different executing agencies that carry out research, their organizational structures, and the funding arrangements. Second, there is the "scientific" context of research, including the general biological and physical sciences and the nature of the systems being studied. That context also includes research and adoption lags, as well as differing mixtures of basic and applied research. The importance of these factors for research evaluation and priority setting is addressed in section 1.2. Third, there is a "policy" context: agricultural research is only one of many policy instruments and its economic impact is conditioned by other policies. In section 1.3 we consider the economic reasons for government intervention in research, the implications for the appropriate forms of intervention, and the objectives for it. To conclude, we discuss research as an instrument of social policy, considering the availabil­ ity of other policy instruments to pursue different social goals. 1.1 Institutional Setting Agricultural research is conducted in a variety of institutional settings that may influence the procedures for allocating research resources. Three key features of the institutional setting are (a) the form of the "executing agen­ cies" that carry out the research, (b) the "structure" within those agencies, and (c) the "funding arrangements" for R&D. 1.1.1 Executing Agencies There is a wide range of organizational structures for public agricultural research systems. Typical systems include (a) those based in a university, (b) those based in a ministry of agriculture, (c) autonomous or semiautonomous research institutes, and (d) agricultural research councils. Mixtures of these systems are common, and public systems usually coexist with private ones. The best organizational structure for research will vary, depending on the size and resources of the economy and its stage of development, the nature of the government bureaucracy, the objectives for the research system, and other factors. Each type of research system brings with it special character­ istics. For example, a university system may consider the complementarity between research and education when setting priorities, while an institute The Institutional, Scientific, and Policy Contexts 5 system may not. A system based in a ministry of agriculture may give less budget autonomy to its agricultural research director than the autonomous­ institute system does, but it may have its goals tied more explicitly to the government's goals for the agricultural sector and less directly to those of particular commodity groups. The choice of priority-setting procedures for research may be affected too. For example, a value is placed on educational output when setting priorities in a university-based system but not when setting priorities in a ministry-of-agriculture system. The ministry-of-agriculture model has been the most common system for research on food crops in smaller countries and an important part of inte­ grated federal-state systems in large countries (Trigo 1986). The autono­ mous or semiautonomous research institutes have been most common in countries with a significant amount of plantation agriculture or other large­ scale forms of production. Several countries, particularly in Latin America, have recently attempted to increase the degree of budget and management autonomy for systems that have historically followed the ministry-of-agri­ culture model (Sarles 1990). The agricultural-research-council model is found primarily in Asian countries such as India, Bangladesh, and the Philippines. Council structures are often justified on the grounds that they improve the coordination of a system in which two or more of the other models are present (Jain 1989). 1.1.2 Structure within Agencies Most university-based agricultural research is organized within disci­ pline-based departmental units. This structure evolved in part because it facilitates teaching. The university-based systems are often the most difficult to evaluate because the units are neither strongly linked to commodities (in some cases they research issues that span both the agricultural and nonagri­ cultural sectors) nor solely focused on research. Nonuniversity-based systems have tended to include a combination of commodity-based, multidisciplinary units, as well as disciplinary units such as soil science, plant pathology, and agricultural economics. Research sys­ tems in developing countries most often follow this combined commodity­ discipline structure. Particular commodities or disciplinary units are then frequently assigned to individual stations or institutes within the system.2 The implication of the commodity-discipline structure is that research evaluation and strategic priority setting involves evaluating the benefits from 2 Sometimes these systems contain multidisciplinary, noncommodity-based units as well, such as natural resource management research units. 6 The Institutional, Scientific, and Policy Contexts commodity-based research programs, ranking them, and suggesting resource allocations, as well as evaluating, ranking, and suggesting allocations to disciplinary programs. As mentioned above, it is often easier to evaluate and prioritize commodity programs and the disciplinary components of com­ modity programs than it is to evaluate disciplinary programs that cut across several commodities or multidisciplinary programs that are not commodity­ based (e.g., natural resource conservation). Superimposed on the commodity and disciplinary foci in most countries is a spatial structure; programs are evaluated both regionally and nationally. Sometimes regional and commodity foci are correlated, but not always. One implication of the spatial structure is the need to assess regional as well as national priorities and the consequences of research results spilling over from one region to the next. These may involve spillovers of technologies themselves or the spillover effects of induced price changes. Some research agencies have a mandate that applies specifically to a subnational geopoliti­ cal region or a particular agroecological zone, and some agencies have mandates that are multinational as well as multi regional. In all of these cases, the technological and price spillovers may have significant consequences. 1.1.3 Types of Decisions Resources are allocated to research at different stages in the system, at differing degrees of aggregation, and they have an impact in several dimen­ sions. Research priorities are set across commodity programs, disciplinary (and multidisciplinary) programs, and research problems. Decisions on program emphasis affect the locational emphasis of research, the focus on particular factors of production (e.g., land, labor, and water), and the dis­ tributional effects of research (e.g., on different farm sizes, on producers versus consumers, and on people at different income levels and at different points in time). These are strategic decisions that guide a research system over several years. The generation of information to support such decisions is at the heart of this book. Decisions about allocating resources within research programs involve the selection of specific projects and experiments within projects. These are generally tactical or shorter-term decisions that affect the relative emphasis within commodity and disciplinary (multidisciplinary) programs. There is a danger of stifling the ingenuity and entrepreneurship of scientists by over­ formalizing this allocation process. However, it is desirable to have a process in place so that given suitable guidance, researchers can properly screen alternatives. In some situations, even for such tactical decisions, a full-scale formal evaluation and priority-setting study will be warranted. The Institutional, Scientific, and Policy Contexts 7 Cutting across strategic and tactical decisions are allocative decisions made with respect to operating funds versus human and physical capital. To a large extent these day-to-day operational decisions are the province of research management and beyond the scope of the formal procedures out­ lined in this book, although of course they ought to be consistent with the same underlying principles. 1.104 Funding Arrangements Funding for public agricultural research is provided by a variety of public and private sources. The federal-state (national-provincial) and public-pri­ vate sharing of funding responsibilities is justified in large measure on the notion that research benefits spill across geographical boundaries and across firms and, hence, a mix of funding sources will generate the optimal amount of overall research investment. Such spillovers may mean that federal fund­ ing of state research can also be justified normatively on distributional grounds. Funding of research in developing countries by more-developed countries can also be justified, in part, by the direct impact on the economic self-inter­ est of the donors or by distributional (moral) considerations. Strategic politi­ cal self-interest undoubtedly plays a major role (Ruttan 1989). Whether the motivation of the donors is to "do good" or to "do well," for many countries, particularly in Africa, external sources of support (grants and loans) can constitute the largest share of the budget (often more than 60%). The influence of these various factors on domestic research priorities can be substantial. The specific mechanisms for collecting financial resources and funnelling them to public agricultural research are also quite diverse. Money may be collected through targeted agricultural taxes, general export taxes or import tariffs, income or land taxes, or commodity check-off schemes, among other means. Most often the funds are derived primarily from the general revenues of the government. These public-sector resources may then pass through the ministry of agriculture in whole or in part, or they may be combined with private and external support provided to a semi-private foundation which then channels support to the public research system. The specific funding mechanism has implications for the incidence of research costs and the rate of return (Alston and Mullen 1992) and can influence the processes for evaluating the effects of research, setting research priorities, and allocating research resources. 8 The Institutional, Scientific, and Policy Contexts 1.2 Scientific Context General developments in biological and physical sciences have dramati­ cally changed the opportunities and requirements for successful R&D and the nature of the systems being studied. Such changes and other factors define the scientific context in which agricultural research is conducted, which, in tum, has implications for research benefits and costs and for procedures for evaluating R&D and setting priorities. In this section we consider three elements of the scientific context of research: (a) the varying nature of research itself (i.e., relatively basic as well as more applied and adaptive research), (b) the role ofthe linkages that transfer information and research results, and (c) dynamics - the fact that agricultural research is a time-in­ tensive process (taking time to produce results that require more time for adoption and that may eventually depreciate). 1.2.1 Basic, Applied, and Adaptive Research Not all research results are intended to be applied directly by farmers, policymakers, or other decision makers. Some research is intended to gener­ ate fundamental knowledge that other scientists can use when conducting more applied research and developing specific technologies or institutions. For example, applied plant-breeding research makes use of research results in genetics, molecular biology, and statistical theory. Molecular biology and statistical theory make use of even more basic research in mathematics. Therefore, agricultural research can be viewed along a continuum from very basic research in scientific disciplines to very applied and adaptive research with farm-level and policy-level applications (Huffman and Evenson 1993; Seaton 1986). Basic research provides the foundation for more applied research, and some applied research results may be further adapted or tested before being used in the field. Advances in biotechnology have influenced research at several locations along the continuum in recent years and have served to strengthen linkages between basic and applied research. In some important respects, modem biotechnology methods have changed the nature of the research process itself (Persley 1990). Because basic research does not directly result in changes in production or cost, it is relatively difficult to quantify the benefits arising from such research.3 Smaller NARSs tend to concentrate on applied and adaptive re- 3. lbis is not to say that benefits from basic or pretechnology research cannot be quantified using the conventional apparatus. To do so it is necessary to infer implications for changes in particular technologies affecting commodity lII31Xets arising from a particular pretechnology innovation. The Institutional, Scientific, and Policy Contexts 9 search;4 however, even large, resource-abundant NARSs conduct a good deal of applied and adapti ve research. The different types of research are comple­ mentary. This highlights the point that the users of research results are not neces­ sarily the economic beneficiaries of the research. Scientists in applied research may use the results produced by scientists in basic research. Pro­ ducers may use the results of applied and adaptive research. The ultimate beneficiaries, however, may be producers, consumers, and scientists who perhaps do not live in the same state, region, or country where the research takes place or where the research results are adopted, and who perhaps are not even living at the time when the research is undertaken or its results are adopted, in the case of work on sustainable agricultural systems. Sometimes the principal users of research results may be nonfarm producers in the private sector (e.g., suppliers of farm machinery, plant material, or chemical inputs), but even in these cases, the ultimate beneficiaries may be farmers or consumers rather than agribusiness firms. 1.2.2 Research and Technology-Transfer Linkages Some agricultural research systems have a mandate to transfer informa­ tion and technology to producers and policy makers. Even when the provi­ sion of extension services is not part of its mandate, a research institution has to obtain information on the current and potential problems facing producers and other clients and has to test new technologies under actual production conditions. Decisions about research investments must consider farm-level problems and the constraints on technology adoption, as well as the problems facing other users of research products. The role of on-farm research in relation to on-station research has been examined in recent years (Tripp 1991). Not all research can be done at the experiment station because too many factors are held constant, thereby reducing the applicability of the results. Likewise, not all research can be conducted on-farm because there are many factors that must be held constant at some stage in the research process, which is difficult to do at the farm level. Many agricultural research systems now contain a significant on-farm research component, which, among other things, is intended to keep scien­ tists aware of current and emerging problems and to help them understand how their research fits into the farm system. 4. Adaptive research involves bringing in and modifying technologies and institutions produced elsewhere. The smaller the country, the more economic it will likely be to focus on testing and adapting research results from other countries, particularly if results are available from regions or countries with similar agroecological, economic, and social environments. 10 The institutional, Scientific, and Policy Contexts Extension systems and on-farm research link research investments to current problems. Ideally, research and extension systems involve a two-way continuum of communication from basic research to transfer and adoption of information and technology, but in many instances the reality falls far short of this ideal. Farmers and ranchers are not the only users of the knowledge generated by agricultural research systems. Input suppliers, processors, policymakers, community development planners, and many others have become more significant clients for public agricultural research systems than they were a few years ago. Technology- and information-transfer mechanisms for these groups are often less formal than the farm-to-station linkages. Technology- and information-transfer issues are also linked to questions of location, size, and scope in agricultural research within a country and to questions of international spillovers of research results. However, there has been relatively little formal analysis of the optimal location, size, and scope of research facilities and programs (Ruttan 1982; Pardey 1986).5 A variety of factors influence the location, size, and scope of within-country transfers of agricultural research technology as well as international transfers. These include (a) the sensitivity of the applicability of research results to environ­ mental conditions - the similarity of climate, topography, farm size, and so on within the country, (b) the relative costs of site-specific research versus transfer and adaptation of technologies, (c) the complementarity with and availability of research results from other countries or international research centers, and (d) economies or diseconomies of size and scope (Evenson and Binswanger 1978; Pardey, Roseboom and Anderson 1991). 1.2.3 Dynamics The current stock of usable knowledge is the result of previous invest­ ment. It can grow as new research investments are made and diminish as current technologies, institutions, and pretechnology information depreci­ ates. Utilization of this stock has not only a spatial but also a time dimension. It often takes a long time for research knowledge to be developed and adopted, typically one to 10 years between the initiation of a research project and the dissemination of results. Borrowing research results (e.g., plant lines or varieties) from other countries can shorten research time in some cases, but many high-payoff research projects still cannot be completed in less than a year. Therefore, part of the purpose of research priority-setting activities is 5. Analysis of spillovers has received somewhat more attention. For example, see Edwards and Freebaim (1982), Brennan (1986), Davis, Oram and Ryan (1987), Evenson (1989), Griliches (1992), and Wood and Pardey (1993). The Institutional, Scientific, and Policy Contexts I I to make longer-run strategic decisions that, once made, can insulate certain types of research for long enough to allow successful completion. The length of time needed for research as well as for adoption is important for another reason. The sooner research results are achieved, the greater the potential economic returns. Benefits received today are worth more than the same received tomorrow (because those received today could be reinvested sooner to earn additional returns); so, research evaluation and priority-set­ ting procedures must recognize the need to discount future research costs and benefits. Not all applied research is intended to raise agricultural productivity from current levels. Significant research investments, particularly in entomology, plant pathology, and plant breeding are required just to maintain current levels of productivity. Estimates indicate that 35% to 70% of U.S. agricul­ tural research is needed to maintain previous research gains (Heim and Blakeslee 1986; Adusei 1988; Adusei and Norton 1990). Several developing countries have found that, with insufficient research support, productivity not only ceases to grow, but actually declines. Therefore both research evaluation and decisions on allocations of resources to research must con­ sider research depreciation and the fact that it may vary by commodity. 1.3 The Policy Context The institutional and scientific contexts interact with government policy to define the environment in which agricultural research takes place. To­ gether they determine the economic impact of R&D and they have im­ plications for research evaluation, priority setting, and resource allocation. These separate elements of the research environment are not independent: for example, the institutional arrangements arise in part from a consideration of the objectives and the scientific context. The main (normative) economic argument for government intervention is a "market failure" argument. In this section we layout that argument and related arguments, and we discuss the use of research to pursue objectives other than economic efficiency. Other factors that might explain public-sec­ tor agricultural R&D are discussed as well. Finally, social science research is presented both as a comparatively neglected component of agricultural research and as a complement to other types of agricultural research - such as when an economic way of thinking about R&D is integrated into the institutions where research is evaluated and prioritized, and research re­ sources are allocated. 12 The Institutional, Scientific, and Policy Contexts 1.3.1 The Economic Justification/or Government Intervention The primary justification for public-sector investment in agricultural research, from an economic-efficiency standpoint, rests on the assumption that there is a "market failure" in the private production and funding of R&D. That is, the market does not provide the private sector with incentives to support the quantity and mix of research that would be best from society's point of view. Such market failures arise when individuals cannot appropri­ ate all of the benefits from their R&D investments.6 Other individuals can "free-ride" on the investment. The ability of the private sector to capture the gains from research varies from industry to industry and from country to country because of differences in technologies and laws governing property rights, among other things (Evenson and Putnam 1990). When private benefits are less than social benefits from an incremental investment in R&D, there will an underinvestment in R&D from society's point of view and it will become appropriate for the government to intervene. One typical intervention is to make public funds available to support R&D that may be carried out in either the private or public sector - most often the latter. However, taxpayer funding of public-sector R&D is not the only way to correct a private-sector underinvestment. When the underinvestment is due to free-riding by producers on one another, it might be fairer and more efficient for the government to create an institution to carry out research on behalf of producers using funds collected by taxing output, for example (Alston and Mullen 1992). Because public resources are limited, when the objective is to correct a market failure, the public sector ought to focus its support more heavily on types of research that have a high social payoff but which the private sector has relatively little incentive to support. Historically, the private sector has concentrated much of its own research efforts in the areas of seed, ma­ chinery, and chemicals, where patents and licenses have generally been more easily obtained and enforced, thus avoiding or reducing many of the free­ rider problems that arise in other settings. Private firms also undertake R&D to develop new technologies for processing, typically post-harvest process­ ing, storage, and transportation technologies, where secrecy enables firms to capture the cost savings arising from such innovations. Because gains from more basic research may be difficult to capture privately, the public sector often supports research that cannot be transferred immediately into new technologies and institutions. In some cases, the private sector conducts this 6. Information and technologies developed from research have, to some extent, the characteristics of "public goods" (e.g., public defense or radio broadcasts), for which one person's use does not diminish their availability to others. The Institutional, Scientific, and Policy Contexts 13 research, in some cases it is done by the public sector, and in some cases the two sectors jointly complete the research. Farmers and ranchers have little incentive to conduct much of their own research. Their large numbers and relatively small firm sizes (and the fact that many of the products from research have the characteristics of public goods) typically mean that individual firms would not be able to capture much of the total benefits. In addition, there are often economies of size and scope in research, which means that a large, diversified organization is often able to do the same research at lower cost than a number of smaller ones could. The fact that, in some cases, research is most efficiently carried out by a large (public) organization does not necessarily mean that it should be paid for out of general government revenues. Spending and funding decisions are separable. In many countries, producers provide financial support to the public research system. This support may be generated through self-imposed levies by producer groups or through export taxes. For example, rice produ­ cers in Uruguay directly support rice research, and Colombia funds its research on coffee through a tax on exports. The relevant question may be what the cost-sharing arrangements should be between the different arms of government and producer levies in supporting particular research programs. And the answers may be expected to vary among programs. It has also been suggested that because there are economies of size and scope in research, small firms (i.e., firms that produce only a small fraction of total production) might not be able to undertake large-scale research programs and therefore may be at a disadvantage compared with very large firms in generating their own research. Thus, public research may provide knowledge that enhances the competitive structure of the market (Ruttan 1982). An additional reason sometimes offered for public-sector involvement in agricultural research is that research, higher education, and extension are complementary (Ruttan 1982). The U.S. land-grant university system of integrated agricultural research, teaching, and extension takes explicit ad­ vantage of these complementarities. However, R&D institutes, particularly in developing countries, are often totally divorced from teaching universities and are less well structured to take advantage of complementarities with teaching and, sometimes, extension. Finally, some say that, from society's viewpoint, the private sector will underinvest in R&D simply because research is risky - the payoff is uncertain because the scientific outcome is unknown and, even if the re­ search is successful, its economic impact depends on things that cannot be known with certainty. However, most economists would disagree. All eco- 14 The Institutional, Scientific, and Policy Contexts nomic activity is risky and the justification for government intervention must go beyond the presence of risk to show that the private sector is unable to spread its risk economically. It is not obvious on the face of it that research risk is special compared with other business risk.7 A further implication of the economic arguments used to justify govern­ ment intervention in R&D is that public resources for research should not be allocated in a manner that competes with or crowds out actual or potential private-sector research. Government intervention is warranted only if (a) incentives are such that markets fail to produce the socially optimal amount of research. (b) economies of size and scope in research threaten the competi­ tive structure of markets. or (c) opportunities exist for exploiting the com­ plementarities between research. education. and extension. Even when government intervention is warranted. the form of intervention might not involve using the general revenues of government to fund the research or doing the research in a public-sector institution. Thus. the predominant form of agricultural research. public-sector research funded by the government. is economically justified only in a limited set of conditions. It might be expected that those necessary conditions will be fulfilled more often in poorer countries than in richer ones. In less-developed countries there might be a greater chance of market failures in research associated with transaction costs, problems with property rights, or other distortions (e.g., capital market imperfections). In counterpoint, however. for similar reasons, in less-developed countries the opportunity cost of the general revenues of the government can be expected to be relatively high - the taxation system is likely to be relatively inefficient and there are many competing uses for the funds for health, education, rural infrastructure, and other capital invest­ ments that also have high rates of return. Thus. while there might be grounds for a greater role for the government in research in less-developed countries, the potential for public-sector research funded by general revenues might be smaller than in more-developed countries. 1.3.2 Research as an Instrument of Social Policy Agricultural research is conducted in the context of other economic and agricultural policies, but research is only one instrument of social policy, and most nonefficiency-related objectives are more effectively pursued using other policy instruments. Thus, public-sector research should be treated as one of several available instruments for attaining agricultural-sector goals, 7. There is no doubt about the riskiness of research. The argument is that such riskiness is not pertinent to the social evaluation of such investments (Arrow and Lind 1970). The Institutional, Scientific, and Policy Contexts 15 and decisions on research resources should reflect the reasons behind pub­ lic-sector involvement in research. In many places, stated objectives for the agricultural research system include economic growth, income distribution, and food security.s Growth in agricultural production is important for improved welfare and overall economic development in many countries. Even in wealthy coun­ tries, production growth can help keep food prices down, generate foreign exchange, and improve competitiveness in world markets. Agricultural re­ search, through its influence on productivity, is a major source of growth in agricultural production and income; however, research is just one of many activities that can contribute to growth. Some policies focus on income distribution among different income classes, geographic regions, different types of producers, and between pro­ ducers and consumers. Research can have significant distributional im­ plications, but this need not mean that research should be directed to pursue distributional objectives. Food security, the long-run sustainability of agricultural production sys­ tems, and the quality of the natural resource base are becoming more important objectives. Population pressures, outdated and inappropriate insti­ tutional structures, and a variety of other factors have created a series of problems with deforestation, soil erosion, desertification, and pollution. Agricultural research can either reduce or worsen these problems through its impact on technology and institutions. We mentioned earlier the question of whether the riskiness of research justifies government intervention. A separate idea is the impact of research on the risk associated with agricultural production. Statements of national goals and objectives often refer to desires to improve security and make incomes more stable. Alternative agricultural research portfolios may have different implications for the variability of agricultural production and hence for food security or income variability. Priority-setting exercises in research might consider its effectiveness as a risk-reducing strategy but ought to do so with due regard to the effectiveness of other public interventions designed to stabilize output, prices, and incomes. More generally, the use of public-sector agricultural research to pursue nonefficiency objectives can be questioned on two grounds. First, consider­ ing more than one objective adds greatly to the cost of decision-making, and second, there are usually better instruments for pursuing nonefficiency goals. 8. Environmental objectives are frequently voiced as weU but can be thought of as falling under growth, distributional, and security objectives. For example, environmental concerns often arise when measures of growth fail to include the external costs associated with environmental damage or when the distribution of benefits to future generations may be jeopardized. 16 The 1nstitutional, Scientific, and Policy Contexts Research administrators have a difficult enough time evaluating research programs or choosing a research portfolio when increasing total net benefits (sometimes referred to as growth or efficiency) alone is the objective. It is even more difficult when two or more objectives are involved. When multi­ ple objectives are considered, the evaluation involves not only identifying specific objectives and measuring the contributions of alternative research programs to each of them, but it also requires trading off or weighting the alternative objectives. Attaching weights is problematic because the subjec­ tive value judgments of individuals are required and decisions must be made about whose judgments are relevant. In addition, when multiple objectives are being pursued, it is important to assess the comparative advantage of research relative to other policy instru~ ments for meeting social objectives. Many economists view agricultural research as a blunt instrument for achieving nonefficiency objectives. Re­ search directors often agree, but other agricultural policymakers and interest groups sometimes make their support for research conditional on a consid­ eration of its distributional, security, and environmental consequences. These latter groups argue that (a) agricultural research has distributional consequences and (b) the transactions or political costs associated with using research to meet particular objectives are lower than those associated with alternative policies.9 The economic arguments support a singular objective of economic effi­ ciency. On the other hand, it appears that public-sector agricultural R&D is often driven, in fact, by its impact on particular groups. Thus, distributional objectives may be a fact of life in public-sector agricultural R&D. An important role for analysts evaluating research programs is to inform deci­ sion makers about the costs of biasing the research portfolio in pursuit of particular objectives. 1.3.3 Political Economy Perspective on the Demand/or Research The discussion above provides a normative perspective on the circum­ stances under which government should be invol ved in funding or executing agricultural research. It does not consider explicitly, however, the underlying political and economic forces that affect the demand for particular levels and types of research. Decisions on allocating resources to research are made in the context of these forces. The incidence of benefits and costs of research generates political pressures that influence the size and direction of research 9. Some economists (e.g., Gardner 1988; de Gorter, Nielson and Rausser 1992) have suggested that policymakers may even use research to try to offset the negative distributional effects of other policies. The Institutional, Scientific, and Policy Contexts 17 funding. Hence, this incidence must be understood by analysts attempting to inform decision makers. Furthermore, research policies and pricing policies may be jointly determined (Gardner 1988), and each set of policies can affect the economic costs and benefits of the other (Alston, Edwards and Freebairn 1988; Alston and Pardey 1991, 1993). The potential beneficiaries of research include producers, consumers, owners of factors of production, and even scientists and administrators themselves. Who benefits depends on many factors, including, among other things, the nature of the research-induced technological change, the nature of the market for the commodity being affected by it, and the incentive structure in the research system. The country's trade status in the commodity (i.e., whether it is an exporter or an importer, whether it is able to influence world prices), its price policies, the nature of the research, government regulations, and other factors also have an influence. The possible joint determination of research and price policies, combined with the unequal ability among producer groups and others to influence the direction of technical and institutional change (Ulrich, Furtan and Schmitz 1986; Roe and Pardey 1991), lends particular importance to analyses that demonstrate the trade-offs associated with alternative research portfolios. 1.3.4 Roles for Social Science Research Private incentives for research are especially lacking in the social sci­ ences. Socioeconomic research develops marketing and management tools, provides information to improve efficiency in the farm and marketing sec­ tors, aids in the design of new technologies, and supports improved policy decisions. Much of this work, especially market information and policy analysis, has a large public-good component. Agricultural research, particu­ larly in developing countries, is often viewed as synonymous with develop­ ing improved technologies (e.g., new crop varieties, methods of pest control, and livestock management practices). Socioeconomic research tends to receive little attention in agricultural research institutions, although it can enhance the benefits from technological innovations (Byerlee and Franzel 1993), as well as being valuable in its own rightY' Research evaluation and priority-setting methods should be able to take advantage of the types of outputs produced by socioeconomic research. However, as we will see later in this book, the methods of analysis presently 10. Within research institutions that focus on technology development, economists and other social scientists are often involved in planning, monitoring, and evaluating research. In addition, they can coordinate with the socioeconomic units in the planning and policy sections of the ministries of agriculture and finance to help design policies that facilitate rather than hinder the adoption of new technologies. 18 The Institutional, Scientific, and Policy Contexts available are much better developed for evaluating the impact of R&D leading to embodied technological changes (where the effects are reflected fairly directly in commodity or factor markets) rather than disembodied technological changes (such as those commonly produced by social science research).11 Our ability to evaluate the impact of R&D varies directly with the nature of R&D. Basic or pretechnology research is more difficult to evaluate from an economic perspective than applied or adaptive research and exten­ sion. Structuring research programs to encourage interactions between social and technical scientists can be beneficial, in part as a way of developing an institutionalized "economic way of thinking" about the role of the R&D effort in the economy, what R&D should be done, or any other questions about the economics of public-sector agricultural R&D, such as those posed at the beginning ofthis chapter. II. One exception may be social science research directed at policy reform, which, if successful in leading to a policy change, is directly reflected in commodity and factor I1IlIrl minus the costs, Ct+k, associated with the program, discounted at an appropriate rate, r (here, for simplicity, assumed to be a constant), as follows: NPVt = i Bt+k - Ct+k (2.7) k=O (1 + r)k The IRR is calculated: as the discount rate at which the NPV is exactly zero . .0 = i Bt+k - Ct+k (2.8) k=O (1 + IRR)k Further details on these alternative methods and their advantages and disad­ vantages can be obtained from a number of sources (e.g., Mishan 1981). For most purposes, the NPV method is preferred. In this approach, any program with a positive net present value is profitable. The disadvantage of 11. Evidence for the United States suggests that maintenance research represents about one-third of production-related agricultural research. See Adusei (1988) and Adusei and Norton (1990). Research Evaluation and Priority-Setting Principles 33 this method is that it does not provide a convenient ranking of alternatives because, although the scale of benefits is measured, the scale of the invest­ ment is not revealed. Concern over scale of investment is not an issue when funds are 'unlimited: all programs with a positive NPV are profitable. When funds are limited, an alternative is to express the net present value per unit of constrained input (e.g., per unit of research investment or per scientist) and rank programs accordingly. The IRR method does rank programs clearly in terms of their profitability, but it does not reveal either the scale of the investment or the value of the programs. According to this criterion, pro­ grams are profitable if the IRR is greater than the opportunity cost of funds. One criticism of IRR is that it assumes that the stream of benefits can be reinvested at the computed rate of return, which is implausible in many cases of agricultural research where very high rates of return are obtained. Often, a combination of IRR and NPV calculations can be used as complementary approaches to summarize the relevant information on the total returns to research. Typically IRRs have been used in ex post evaluation studies, while NPVs (per scientist or per unit of investment) have been used in ex ante evaluation and priority-setting studies. Two important, related questions transcend the choice of method: What is the appropriate rate of discount (a required rate of return either for use in NPV calculations or for comparison to calculated IRRs)? How should uncer­ tainty surrounding the estimated stream of benefits and costs be handled? Some people try to deal with both questions by requiring conservatively large rates of return. In our view (with the support of the mainstream of the literature on project appraisal), it is not appropriate to deal with uncertainty by adjusting the rate at which the streams of benefits and costs are dis­ counted. In addition, some have argued, especially recently, that we should be using low discount rates so as to encourage re.se!!t:Gnjnto technology that conserves natural resources (Cline 1992).12 Underlying' this is an implicit belief that we should attach greater weight to the welfare of future genera­ tions and that biasing the pattern of agricultural research is an appropriate way to achieve an intergenerational redistribution of welfare (Birdsall and Steer 1993). Ad hoc reduction of the discount rate is, however, unlikely to be a good way to account for externalities or to incorporate (intergenera­ tional) distributive weights into research evaluation. 13 In most situations, 12. Ruttan (1994) points out that the impact ofloweringthe discount (or interest) rate on the rate of exploitation of natural resources is not entirely clear and the relationships are not simple. Lipton (1991) discusses the effects oflow and high interest rates on sustainability. See also Pearce, Barbier and Markandya (1990). 13. Mikesell (1991) suggests taking explicit account of resource depletion in project analysis, instead. 34 Research Evaluation and Priority-Setting Principles there are likely to be less costly means of achieving environmental objectives (or intergenerational transfers) than biasing the pattern of agricultural re­ search specifically towards environmentally friendly projects. As a final note on this topic, sometimes it is helpful and important to distinguish between marginal and average effects of research and rates of return to research. It is often suggested that the research production function is characterized by diminishing returns. Indeed, if this were not the case, it might well be better to specialize much more in a smaller number of research programs or projects within programs. With diminishing returns, the mar­ ginal return to increasing the budget for a particular program will be smaller than the average return to the total investment in the program. When the relevant decision is whether to continue a program or close it down, the average return is the appropriate measure. When the relevant decision' is how to allocate an increase in the total research budget among programs (or, more realistically in the current environment, how to distribute a budget cut among programs), the appropriate measure may be the marginal rate of return. Sometimes programs will be ranked differently according to marginal re­ turns than according to average returns. Often both marginal and average rates of return are useful for providing a more complete picture of the opportunity costs of program alternatives. 2.1.3 Other Issues - Uncertainty and Adjustment Costs Uncertainty Agricultural research is an intrinsically risky activity, with probabilities of success akin to those in mining for diamonds or looking for oil. The uncertainty that is inherent in virtually all aspects of the research process and its effects on production and markets creates difficulties both for research administrators and scientists making decisions about their work and for the economist attempting to measure research benefits and costs. Representing uncertainty appropriately in agricultural research evaluation and priority setting is not straightforward. Clearly, any estimation of the benefits from research inevitably involves some estimations of, or assump­ tions about, all of the relevant uncertain variables. In some cases, the results may be insensitive to these assumptions or estimates; in some cases the results will be highly sensitive. Most studies of research benefits do not deal with this question very well,14 As Anderson (1991, p. 103) puts it: 14. A few studies have looked at taking formal account of risk in research benefit calculations (e.g., Fishel 1970; Dyer, Scobie and Davis 1984; Anderson 1991; Scobie and Jacobsen 1992). Research Evaluation and Priority-Setting Principles 35 Most of the fonnal literature on agricultural research per se, whether of a managerial or evaluati ve orientation, implicitly treats research and its setting as being deterministic. In fact, of course, the process is intrinsi­ cally uncertain. Most agricultuml sectors are highly variable and the observed variability is extremely unpredictable so that it is, technically speaking, risky. The conjunction of an uncertain research process with an uncertain physical and economic environment is the reality of agriculture that makes it all an extremely risky business. There is thus a considemble mismatch between nearly all the literature on research resource allocation and that on decisions about investing in research in the risky environment in which this takes place. The analyst's uncertainty: Uncertainty surrounds most of the variables and parameters involved in the calculation of the returns to research. It is almost tautological to note that there are uncertainties in the research process itself. The time taken to complete research is not precisely known, the scientific outcome of a particular line of research is uncertain, and the impact ofthe resulting new knowledge on yields, costs, and so on are also unknown at the time the research begins. It is not known in advance whether a project will lead to a commercially successful result, and the time lags and the adoption path are uncertain as well. In ex ante studies it is important to take the possibility of the failure of research into account through the use of some measures of probabilities of success, which vary by scientist, commodity, and type of research. The results of research aimed at varietal improvement in wheat, for instance, may be fairly predictable because of the constraints imposed by the laws of quantitative genetics. On the other hand, the results of research looking at the possibility of nitrogen fixation in wheat are surely much less predictable. Even when the scientific outcome of research is known (as happens in ex post evaluation studies), the measures of benefits are uncertain because the measures of annual flows of benefits involve market parameters that are uncertain. These include elasticities and functional forms of supply and demand and the values of those functions, as well as government policies, among other things. Also, the costs of research are uncertain. In addition, future market outcomes (either with or without successful research) are characterized by uncertainty, some of which is due to uncertainty about government policies. It is clear that investing in a particular agricultural research project is a risky business. But the approaches being developed here are intended to be applied more at the level of the research program, involving a portfolio of individual projects, than at the level of individual projects. The riskiness of large programs may be very different from the riskiness of individual projects (formally, this depends on the covariance of returns among the 36 Research Evaluation and Priority-Setting Principles individual projects within the program). It is likely that risk will be less serious at the program level. However, even when risk or uncertainty is not a concern, it may be important to account for its effects on the mean project performance in an investment analysis (Anderson 1991, p. 126); uncertainty may affect the expected value of the stream of research benefits and costs. In some analyses, it might be sufficient to consider uncertainty only in so far as it affects the expected benefits, but the measures of expected benefits have a distribution around them generated by the distributions surrounding the underlying variables. In some contexts, the variance and skewness of the distribution of research benefits also may be of interest, along with the expected values. 15 Mean-variance trade-oft's in the research portfolio: The fact that mar­ kets are characterized by uncertainty, and that research investments are risky, means that research investments can be evaluated in terms of their relative riskiness as well as their expected benefits. An issue to be addressed is whether the riskiness of alternative programs ought to be considered in the evaluation and, if so, how these considerations should be introduced. Whether the intrinsic uncertainties of doing research have a bearing on attempts to appraise the social worth of the enterprise depends on the acceptance or rejection of the controversial (but generally accepted) argu­ ments of Arrow and Lind (1970, 1972) as to the relevant criteria for appraisal of public investments. 16 The idea here is that government can effectively "pool risk into unimportance" through its large and diversified investment opportunities (Anderson and Dillon 1992). Presuming that the risks of individual investments (be they projects, programs, or whatever) were sta­ tistically independent, Arrow and Lind (1970, 1972) showed that when such risks are publicly borne, the total cost of risk bearing is insignificant, and accordingly, for most practical purposes governments should ignore these sources of uncertainty in appraising public investments. In other words, risky research projects or programs should not be discriminated against (Parton, Anderson and Makeham 1984). An exception to this general approach is when the research is on a commodity that accounts for a large fraction of the national (or local) earnings so that national (or local) income is correlated with research bene­ fits. In cases such as this, or when the decision makers perceive the riskiness of research investments to be relevant, the research investment decision IS. These higher moments may be estimated crudely using sensitivity analysis for key parame­ ters. In doing this it is important to be aware that the individual stochastic elements are unlikely to be independent, in which case covariances among variables and parameters might be important. 16. See Parton, Anderson and Makeham (1984) for a more complete discussion of this issue in this context. Research Evaluation and Priority-Setting Principles 37 might involve trading off risk against expected benefits when choosing between investments anticipated to give a higher expected payoff (and higher risk) and those anticipated to give a lower expected payoff (and lower risk).17 In such cases, diversification strategies to reduce the riskiness of research investments are relevant. Diversification can have many dimen­ sions in a research context. These include approaches taken, commodities chosen, sites selected, problems addressed, disciplinary perspectives used, and different investigators' perceptions about what will best contribute to knowledge (Anderson and Hardaker 1992). If one accepts the Arrow-Lind notion that government should generally be risk neutral in its attitude towards risky research projects or programs, then policy makers and administrators of publicly sponsored research should seek to maximize the expected (i.e., mean or average) social value of research. But this in no way implies that knowledge of the range of possible outcomes of a particular line of research, as opposed to knowledge ofj ust the "most likely" outcome, is irrelevant for decision making. If the distribution of outcomes is not symmetric, then the most likely (or modal) outcome will be a biased estimate of the expected (or mean) outcome. In addition to these statistical quirks, there is the issue of presenting cogent information to decision makers that reflects the real-world uncertainties of the research enterprise and our imprecise perceptions of these uncertainties. In ex ante assessments of the effects of research, precision statistics (even if purely subjective) can be a useful adjunct to estimates of the expected or most likely outcome of research (Anderson 1992). This is particularly true when the expected net present values of research generated by a number of proposed activities are more or less equal. Knowledge about the confidence that can be placed on those prior expectations - as captured by the likely dispersion around the respective mean outcomes - can aid decision making. Production risk and food security: A separate issue is the impact of research on the riskiness of agricultural production itself. Agricultural pro­ duction in marginal areas, especially where rainfall is scarce and erratic, makes farmers particularly vulnerable to the vagaries of the weather. This is especially true for the semiarid areas of Africa, the Middle East, and Aus­ tralia. But even in the humid tropics, where rainfall is relatively abundant and less sporadic, climate-induced fluctuations in agricultural outputs and prices (e.g., as a result of pest and disease outbreaks in intensi ve production systems such as irrigated rice) are also a policy concern. Statements of national goals and objectives often refer to desires to improve security and make incomes more stable. Alternative agricultural 17. This may be so even in a "small commodity" case; for instance in a privately sponsored research unit or a quasi-public research agency that is supported in part by producer funds. 38 Research Evaluation and Priority-Setting Principles research portfolios may have different implications for the variability of agricultural production and hence for food security or income variability. Research priority-setting exercises may need to consider the effectiveness of research as a risk-reducing strategy and may need to do so with due regard to the effectiveness of other public interventions designed to stabilize output, prices, and incomes. The Arrow-Lind argument for focusing decision-making attention on the expected (or average) outcome should not be misinterpreted to mean that the production- and market-related risks perceived by producers and consumers are not a legitimate social concern. Successful R&D might lead to technical changes that alter the riskiness of production, and such effects might be of value to producers and consumers in ways that are not reflected in conven­ tional (deterministic) measures of benefits. To place a social value on the effects of research often requires that explicit attention be paid to the inherent variability of agricultural production and markets, the consequent uncertain­ ties in the agricultural sector, and the degree to which research may modify these types of risks. This variability is due primarily to the variability of the natural environment over time and space, the nature of the economic envi­ ronment for agricultural commodities, and political uncertainties. Anderson and Hazell (1989, p. 340 f.), summarizing the results of Weber and Sievers (1985) and others, note that the baseline levels of production variability are high in many countries, especially in semiarid areas. In addition, production variability tends to be smaller in larger countries because of the greater risk-pooling effects across crops, and regions. Prices can vary markedly from year to year and in unexpected ways within a year. Production changes at home and abroad as a result of changes in weather can cause relatively large variations in prices for many agricultural commodities. These price swings are often especially large for commodities that are produced and marketed locally but which are difficult to transport to national or international markets. The swings can also be quite large for internationally traded goods produced in a relatively small number of loca­ tions worldwide or for goods for which one or two countries are dominant producers. Another cause of uncertainty is potential changes in the political environment. Price policies can change dramatically from year to year, and severe disruptions in an economy can occur as a result of macroeconomic adjustments, wars, coups, and other political changes. When decision makers perceive any of these types of risk to be socially relevant, the problem of decision making becomes one of trading off percep­ tions of risk reductions from research against expected (or mean) benefits foregone as a consequence of this sensitivity to risk. In this instance, the weight placed on the nonefficiency objective (i.e., risk reduction) ought to Research Evaluation and Priority-Setting Principles 39 reflect the decision makers' notion of the social value of such risk reductions. In determining the priority to be given to risk-reducing research, decision makers need to be aware of the efficiency of research relative to other forms of public investments (e.g., improved irrigation services) or other interven­ tions (e.g., price stabilization schemes or buffer stocks) to achieve a certain degree of risk reduction. I8 Adjustment Costs The results of priority-setting exercises for research may suggest modified research programs. There are costs (related to human and physical capital) associated with changing a research program, particularly if the changes are made relatively rapidly. It generally takes some time and expense to train or retrain personnel, while newly installed buildings and equipment usually require a shakedown period before reaching their productive potential. If the cost of "organizational capital" required for growth or change is an increas­ ing function of the speed of adjustment (as assumed, for example, by Lucas 1967 and Prescott and Visscher 1980), then rapidly growing or changing systems will have higher average cost structures than slower-growing ones. These costs have to be considered when short-term research priorities are selected, although they become less important in the medium to long term. The implication is that the agricultural knowledge supply curve differs between the short and long run because of asset fixity. Investments in human capital (e.g., training) and physical capital are costs that must be considered when developing short- and long-run research priorities. A related cost to consider is the possible value foregone when research projects already initiated are not completed. Even when a particular commodity or type of research appears to be of lower priority than other commodities or types of research, it may pay to invest additional resources in the short run to complete work that is already underway so as to obtain the benefits of the previous research investment. Most studies have treated the question of adjustment costs informally, if at all. In a notable exception, Scobie and Jacobsen (1992) use an explicit adjustment cost function. 18. See section 2.5 for a discussion concerning trade-offs between efficiency and nonefficiency objectives. 40 Research Evaluation and Priority-Setting Principles 2.2 Measuring Benefits and Costs Using Economic Surplus Concepts The most common approach for analyzing the welfare effects of agricul­ tural research in a partial-equilibrium framework has used the concept of economic surplus. Griliches (1958), Peterson (1967), and Schmitz and Seck­ ler (1970) provide early examples of applying the economic surplus concept to ex post evaluation of agricultural research, while Davis, Oram and Ryan (1987) and Norton, Ganoza and Pomareda (1987) provide more recent examples of applying the concept in an ex ante setting. Underlying these analyses is a body of theory and set of assumptions that are not always explicitly stated. Harberger (1971, p. 785) defended the general approach and defined three postulates that he suggested should be accepted as providing a conventional framework for applied welfare eco­ nomics. These three postulates are (a) that the competitive demand price for a given unit measures the value of that unit to the demander, (b) that the .competitive supply price for a given unit measures the value of that unit to the supplier, and (c) that when evaluating the net benefits or costs of a given action (project, program, or policy), the costs and benefits accruing to each member of the relevant group (e.g., a nation) should be added without regard to the individual(s) to whom they accrue. When these assumptions are valid, consumer benefits from consumption may be measured as the area beneath the ordinary demand curve, net changes in consumer welfare may be mea­ sured using Marshallian consumer surplus, and the area beneath the supply curve is a measure of total costs, so changes in the net welfare of producers may be measured using producer surplus. In spite ofthe intuitive appeal ofthese assumptions, the approach has not been without its critics. In this section, we will illustrate the basic approach in the context of a simple closed-economyl9 case of a research-induced supply shift, review the criticisms of the economic surplus approach, and consider alternatives to the partial-equilibrium economic surplus model. Our conclusion is that, for most purposes, the partial-equilibrium economic surplus model is the best available method to evaluate returns to research. 19. The term "closed economy" refers to a situation where the commodity of interest is not traded internationally and its price is determined inside the country (or region) of interest. The most important feature of this case is that consumer welfare is affected (in the typical case of a small 'country with an open economy, the price is exogenous). It is important to distinguish between the closed-economy case that arises as a consequence of natural protection (e.g., transportation costs) and a case where there is no trade because the border has been closedby government intervention. In the latter case the simple closed-economy model is inappropriate; a more appropriate model accounts for the effects of government intervention, as described in chapter 4. Research Evaluation and Priority-Setting Principles 41 2.2.J Basics of Economic Surplus Measures In figure 2.4 the supply curve for a commodity under the original technol­ ogy is denoted by So, and the demand curve by D. The origimil price is Po and the quantity supplied and demanded is Qo. Using Harberger' s postulates, the total consumer surplus from consumption of the commodity is equal to the triangular area FaPo (the area beneath the demand curve less the cost of consumption). Similarly, the total producer surplus is equal to the triangular area PoaJo (total revenue less total costs of production as measured by the area under the supply function). Total surplus is equal to the sum of producer and consumer surplus, as shown by the triangular area Fa10 which is equal to the to~al value of consumption (the area under the demand curve) minus the total cost of production (the area under the supply curve). Changes in producer, consumer, and total economic surplus are measured as changes in these areas. Cost-reducing or yield-enhancing research and adoption of the resulting new technologies shift the supply curve to SI, resulting in a new equilibrium price and quantity of PI and QI' The change in consumer welfare (surplus) from the supply shift is represented by the area P(p-bP, and the change in producer welfare (surplus) is represented by the area P ,b11 - Pcp./o. Consumers necessar- Figure 2.4: Producer and consumer surplus measures Price o , Quantity/Year 42 Research Evaluation and Priority-Setting Principles ily gain because they consume more goods at a lower price. In general, the net welfare effect on producers may be positive or negative depending on the supply and demand elasticities and the nature of the research-induced supply shift. This is so because there are two effects working in opposite directions. The producers sell more goods, but they must sell at a lower price. Both costs . and revenues are affected. Producer benefits are assured if costs fall and revenues rise. But under plausible conditions, in some cases (i.e., an inelastic demand) revenUe falls when supply increases. In addition, when supply shifts in a pivotal fashion against an inelastic demand, revenue falls faster than costs, and producer losses are assured. The nature of the supply shift can clearly have important implications for the distribution of benefits. In the case drawn in figure 2.4, with a linear supply curve shifting in parallel, producers necessarily benefit (by an amount equal to area P,bcd =P ,bl, - Pr,aIo). The total (or net) welfare effect is equal to the sum of the changes in producer and consumer surplus,/r,abl, (which, in this case of a parallel supply shift, is also equal to area p(,abcd). The sum of producer and consumer surplus changes measures the net welfare change in the sense that the gainers from technological change could, in principle, compensate losers and still /be better off by the amount loabl,. In the model in figure 2.4, compensation could mean reducing consumer benefits, perhaps through taxes, in order to provide subsidies to producers. The compensation principle assumes that such transfers could be made in a lump-sum fashion without any tax-induced distortions in consumption or production.20 When all losers are fully compensated and there are still some net gains, the new technology constitutes a welfare improvement according to the Pareto criterion. Usually no compensation is paid following the adoption of new technologies produced through research. The Kaldor-Hicks criterion (KHC) for a net welfare improvement is that the gainers could afford to compensate the losers and still be better off, but the compensation need not take place. The KHC is a weaker, and more debatable, criterion than the Pareto criterion but it is also more practicable and underlies applied welfare economics.2 ' However, several other questions have been raised about the 20. A true lump-sum transfer is therefore not possible, because some resources necessarily are consumed in making the transfer. New technology is making these costs smaller. But, even nowadays, at a minimum the cost would still be that of the labor and capital used to make an electronic transfer of funds. More usually there are substantial deadweight costs in collecting funds to finance transfers and in making transfers. 21. Bieri, de Janvry and Schmitz (1972) point out that, if compensation is not paid, the total welfare gains from research equalloablI only if equal welfare weights are attached to producers and consumers. Attaching equal weights is clearly a value judgment. Some might argue that higher welfare weights should be attached to those who lose from technological change if compensation is not paid and if those who lose are already at the lower end of the distribution of income, and lower Research Evaluation and Priority-Setting Principles 43 use of consumer and producer surplus as welfare measures, and these are addressed below. 2.2.2 Criticisms of Economic Surplus as a Welfare Measure The use of consumer and producer surplus has been criticized. ftom several perspectives. Some criticisms have centered around the accuracy of what is being measured, others around the value judgments that are i~plied, and others around the perceived lack of policy relevance. For ease of exposition, we have grouped the complaints into six types of criticisms of surplus analysis, which we evaluate in tum. Normativeness The long-standing debate about the merits of positive (what is) versus normative (what should be) economics has spilled over into discussions about the merits of consumer-producer surplus analysis. As such, the criti­ cism is broader than an attack on economic surplus; rather, it is an attack on the implicit value judgments associated with welfare economics. Mishan (1982, p. 23) points out that, at times, this criticism leaves one with the feeling that "value judgments involve nothing less than an indecent surrender of one's methodological chastity." While normative economics must draw on methods of positive economic analysis, were economists to restrict themselves to positive economics, they would be ignoring many of the vital issues of concern to society. Perhaps more fundamentally, positive econom­ ics (indeed all science) is far from value free. As Boulding (1970, pp. 121-2) puts it, "as science develops, it no longer merely investigates the world, it creates the world which it is investigating ... [and] as science moves from pure science toward control, that is, toward creating what it knows, what it creates becomes a problem of ethical choice." In other words, value judg­ ments are inevitable in any scientific endeavor (including economics). The important thing when conducting consumer and producer surplus analysis is to make those judgments as explicit as possible (Chipman and Moore 1978). The issue of making value judgments explicit is important because, as noted earlier, the validity of using changes in consumer and producer surplus to measure welfare changes rests in part on the compensation principle.22 weights should be attached to those who gain. Weights could be patterned after the relationship between the marginal utility of income and the income levels. Gardner (1988) shows a Cobb-Doug­ las policy-preference function with the characteristic that losers are given more weight than gainers as we move from an initial position with equal weights. Harberger's (1971) third postulate explicitly assumes equal welfare weights within relevant groups. 22. When unequal weights are attacbed by society to gainers and losers (e.g., Harberger 1978), 44 Research Evaluation and Priority-Setting Principles Because of the difficulty in making nondistorting lump-sum transfers, distri­ bution issues are indeed relevant.23 Usually, compensating transfers are not made, and the welfare analysis requires an implicit or explicit value judg­ ment. Measurement Error Most of the literature that discusses the validity of consumer and producer surplus analysis has focused on the conditions that must hold if consumer and producer surplus measures are to provide an accurate indicator of changes in social welfare. This vast literature will not be reviewed here. Instead, we present the key conclusions that emerge from this debate, as we see them.24 There are several alternative measures that have been proposed as money metrics for the consumer welfare change due to price changes. Whereas consumer surplus, CS, as defined by Marshall (1890) is the excess of the price the consumer would be willing to pay over the actual cost of the good, equivalent variation, EV, is the amount of additional money (income) that would leave the consumer in the new welfare position if it were possible to buy any quantity of the commodity at the old price. And compensating variation, CV, is the amount of additional money (income) that would leave the consumer in the initial welfare position if it were possible to buy any quantity of the commodity at the new price. McKenzie and Pearce (1982) argue that the best cardinal representation of the ordinal utility function for the individual is a money metric related to the equi valent variation concept defined by Hicks (1946). This measure is an exact representation of an individual's utility function and can be derived from a Taylor's series expansion and written as a "fixed weight combination compensation schemes that assume equal weights will not ensure net welfare gains from technolog­ ical change. Furthermore, even if compensation is paid, the distribution of income can become increasingly skewed through time if losers are only compensated to the extent that they are no better or worse off than before the technological change. This latter result occurs if gainers are better off than before the technological change, even after they pay compensation, and if losers are in a lower income bracket. The issue of defining unequal welfare weights has received little empirical attention in the literature. ' 23. These transfers are difficult because government does not have sufficient information to make them. Furthermore, tax-subsidy schemes are themselves costly. 24. Some useful and relatively recent references that provide more detail on this topic are those by Currie, Murphy and Schmitz (1971), Willig (1976), Chipman and Moore (1978,1980), Just and Hueth (1979), Hausman (1981), Mishan (1981), Just, Hueth and Schmitz (1982), McKenzie and Pearce (1982), and McKenzie (1983). Alston and Larson (1992,1993) reviewed some more recent literature on the choice between Marshallian and Hicksian measures and discussed the issue of precision, as well as bias, in welfare measures. Research Evaluation and Priority-Setting Principles 45 of product prices, the fixed weights being constructed from first and higher order elasticities of demand and individual income changes, with elasticities evaluated at a base point" (McKenzie and Pearce 1982, p. 681). It is preferred to consumer surplus (as measured off the ordinary or Marshallian demand curve, which holds money income constant) because it accurately captures the income effect associated with a price change. For example, as price declines with a shift out in the supply curve against a downward sloping demand curve, the real income of the consumer increases, which, in effect, shifts consumer demand for the good, thereby increasing welfare. Why then does consumer surplus continue to be used? Undoubtedly, familiarity plays a part, but more important reasons are likely to be that (a) contrary to McKenzie and Pearce (1982), consumer surplus calculations are often made with less information than would be required for calculating their exact money metric and (b) for reasons discussed below, a correct answer is not always preferred to an incorrect one, regardless of how difficult it may be to compute. Consumer surplus studies often do not begin by estimating demand functions. For example, evaluations of the benefits from agricultural research often begin with estimates of ordinary demand elasticities gathered from diverse sources, but comparable information may not be available for calculating the income effects of price changes. There is little question that the McKenzie and Pearce (1982) money metric is the most accurate measure of individual utility under competitive equilib­ rium conditions.25 No doubt, if the functional form of the Marshallian demand curve(s) is known exactly (or at least if its derivatives are known), one can deduce exact measures of the underlying preferences. But this is theoretical sophistry. As a practical matter, in most cases, the econometrics can provide little more than a local approximation to demand at a point. We must recognize that we cannot know the functional forms of supply and demand. The issue then is whether consumer surplus provides an adequate approximation of the market-level analogue to individual welfare changes, given real-world limitations on available information. An intermediate option is to use a linear (first-order) approximation of the area behind the Hicksian demand curve (corresponding to consumer surplus behind the Marshallian demand curve), which measures the compensating variation or equivalent variation for the research-induced price change. This requires a little more information than the consumer-surplus method, but it is explicitly an approximation and does not require knowledge of the func­ tional form for demand. 25. It is not clear that this superiority over consumer surplus applies also to aggregate (market level) measures as well. The issue of aggregation over consumers might swamp the issue of income effects in aggregate welfare measures. 46 Research Evaluation and Priority-Setting Principles Figure 2.5 duplicates the curves in figure 2.4 and includes, as well, two Hicksian demand curves along which money income adjusts to hold utility constant. One of these curves, Ho, holds utility at the preresearch value, Un, corresponding to Po and Qo; the other, HI, holds utility at the post-researfh value, u" corresponding to P, and Q,. The Hicksian demands are shown as less elastic (i.e., less price-responsive) than the Marshallian demand, as applies for a normal good with a positive income elasticity of demand. The Marshallian consumer surplus measure is area Pl'pbP,. The corresponding area behind the first Hicksian demand (area p(pcp, behind Ho) is the com­ pensating variation for the price change from Po to P" and the corresponding area behind the second Hicksian demand (area Pr,ebP, behind H,) is the equivalent variation for that price change. Thus, the consumer surplus overstates compensating variation and understates equivalent variation in the case of a price decrease for a normal good. When information is available on the income elasticity of demand, 'TljY, and the share of the good in total expenditures, Sj, as well as on the Marshallian demand elasticity, 'Tlj, for a good i, the quantity change along the initial Hicksian curve (Q, - Qo) can be calculated (using the Slutsky equation) and used in the consumer-surplus formula instead of the quantity change along Figure 2.5: Accuracy of consumer surplus Price SI Po PI 10 D HI(u l ) II Ho(uo) 0 Qo Q; QI Quantity/Year Research Evaluation and Priority-Setting Principles 47 the MarshalIian demand. This use of the Hicksian quantity change provides a measure of the compensating variation for the price change. The error from using Marshallian rather than Hicksian demand to measure compensating variation is equal to triangle abc. which corresponds to the error of using the Marshallian demand elasticity. 1li. rather than the Hicksian demand elasticity (1li* =1 li + SiTJiY)' The income effect associated with a price decrease (-Si1liY x %M) is positive for normal goods (reinforcing the substitution effect) and negative for inferior goods. so consumer surplus overstates the compensat­ ing variation measure of the welfare change due to a research-induced price decrease for normal goods and understates it for inferior goods. Because triangle abc is only a small fraction of total benefits (and a very small fraction for small price changes). errors in approximating it are unlikely to loom large in estimates of research benefits. In other contexts. where measurement of the triangle is the main question. these errors of approximation may become relatively important. as argued by Hausman (1981). Hausman (1981) showed how to obtain "exact" Hicksian measures of welfare change. given a Marshallian demand function. for the commonly used functional forms and argued that there was no reason not to make the correction for income effects to go from consumer surplus to either compen­ sating variation or equivalent variation. Why use a biased measure when it is easy to correct the bias? One response may be drawn from some of the recent literature. mostly in the field of environmental economics. that has drawn attention to the precision of welfare measures.26 Because the param­ eters of demand equations are random variables. transformations of them used to measure welfare are also random variables. If demand is estimated imprecisely. the bias in the consumer welfare effect or the deadweight loss might not be statistically significant (e.g .• Kling 1992). Alston and Larson (1993) point out that. when correcting for the income effect. an additional source of imprecision in the welfare measure. variance of the income elas­ ticity of demand. is added in order to reduce bias. There may be a trade-off of variance against bias. Using a mean-squared error criterion. they showed that the MarshalIian (biased) welfare measure might be preferable to the Hicksian (unbiased) measure. The issue of precision of welfare measure­ ment. which arises from recognizing that welfare measures are random variables. is relatively new and its implications remain as yet mostly unre­ solved. However. the work that has been done illustrates that considering precision may well weaken the arguments in favor of correcting for income effects to obtain Hicksian welfare measures. 26. This literature - including work by Bockstael and Strand (1987), Hayes and Porter-Hudak (1987), Kling (1988, 1991, 1992), and Smith (1990) - has been reviewed and extended by Alston and Larson (1992, \993), 48 Research Evaluation and Priority-Setting Principles When the income effect associated with price changes is small, consumer surplus is not a bad approximation. Willig (1976) provided empirical evi­ dence that the bias introduced by ignoring the income effect is relatively small for most goods (less than five percent), at least in a relatively devel­ oped country. The income effect is likely to be small when the income elasticity of demand for the good is small, or when a small proportion of the consumer budget is spent on the good.27 T,he significance of "small" can be assessed in the context of other biases introduced from other sources during an analysis. Substantial income effects can be associated with price changes for certain foods in developing countries, but for most, the income-effect bias is likely to be swamped by errors in measuring positions of curves or shifts in them. That is to say, there are a number of sources of potential errors in the analysis and attention ought to focus on the more important ones, which in this case are not due to using consumer surplus instead of either the exact money metric or an approximate measure of compensating or equivalent variation. The above discussion centers on consumer surplus. What about producer surplus? Producer surplus is the excess of the return to the factor owner above that necessary to induce him or her to provide the factor, and it is analogous to the concept of consumer surplus (Mishan 1981). It is meant to measure the change in producer welfare associated with a change in eco­ nomic profit. However, the income effect associated with a change in a factor (or product) price is often substantial, making producer surplus a much less reliable measure of the corresponding equivalent variation or compensating variation. A change in an individual's return to labor or the value of an individual's land can have a major income or welfare effect. There are also problems with the accuracy of producer surplus as a measure of economic profit. Producer surplus measures the return to fixed or quasi-fixed factors and thus may be thought of as corresponding closely to the concept of economic profit: income over and above the opportunity cost associated with variable factors. A voidable fixed costs - costs that do not vary with output, once a decision is made to produce, but which can be avoided by choosing not to produce at all- add complications. An example may be the opportunity costs of specialized equipment. These avoidable fixed costs do affect profit but they may not be reflected in the producer surplus measure that accounts only for variable costs. However, changes in producer surplus are likely to provide a more accurate reflection of changes in profit, especially when the effects of relatively small changes in prices are 27. Alston and Larson (1992, 1993) reinforced Willig's (1976) result and provided some theoretical and simulation results for the case of a research-induced supply shift. The biases in measures of aggregate research benefits were indeed negligible. Research Evaluation and Priority-Setting Principles 49 measured, so that the impact on avoidable fixed costs (i.e., the impact on decisions whether to produce at all or to shut down) is minor. An additional criticism of both consumer and producer surplus is associ­ ated with errors introduced when individual demand or supply curves are aggregated into market demand or supply curves. The underlying assump­ tion usually is that tastes, money income, and the prices of other goods are constant across individuals in the economy (Mishan 1981).28 However, these assumptions underlie alternative measures of welfare as well and, even if they do not hold exactly, are not likely to introduce more serious errors than other simplifying assumptions. Finally, when estimating the benefits from agricultural research, errors are inevitably introduced by assumptions about (a) functional forms for supply and demand, (b) elasticities of supply and demand, (c) other market param­ eters, (d) the nature of the research-induced technical change and the corre­ sponding shift of supply or demand, (e) the size of the research-induced productivity improvement, and (t) the timing of the flows of benefits and costs. As a practical matter, from our experience, the potential for errors from these sources is greater by orders of magnitude than that associated with the income effects of price changes or any other imperfections in the economic surplus measures of welfare changes. The results are sensitive to these aspects of the analysis and it is not possible to obtain definitive support for particular choices. Accurate measurement of even ordinary demand-and-supply curves, par­ ticularly along their entire length, is very difficult. And it is very hard to predict the nature of shifts in these curves (Scobie 1976; Lindner and Jarrett 1978; Rose 1980). Supply-and-demand curves may be nonlinear, but they are often assumed to be linear to simplify consumer and producer surplus calculations. Errors associated with this simplification may not be too severe, but errors associated with unavoidable assumptions about the nature of the research-induced supply shift (i.e., parallel, pivotal, or some other shift) can be major (e.g., see Voon and Edwards 1991c). A parallel shift almost doubles the benefit compared with a pivotal shift. Of course, these errors only arise in consumer-producer surplus applications in which curves are shifting, such as when new technologies are being adopted. What does economic theory tell us about the nature of these shifts? Unfortunately, not very much. To be confident about this aspect of the problem would require either (a) precise econometric evidence or (b) de­ tailed information on the effects on individual agents, details of industry 28. Homothetic preferences would be sufficient to permit aggregation across consumers without introducing such errors. Otherwise. the much more restrictive assumption of equal money incomes across individuals is necessary. 50 Research Evaluation and Priority-Setting Principles structure including details on exit and entry of firms, and a complete theory of aggregation. This information is not available; assumptions are unavoid­ able. The consequences of assumptions for potential error in using producer and consumer surplus for predicting research impact in priority-setting exercises should be kept in mind. In most cases, the effect on producer surplus will be greater than the effect on consumer surplus. Partial Welfare Analysis The validity of a partial welfare analysis, which ignores the complex interrelationships with other product and factor markets in the economy, has been called into question by Little (1960) and others. Broadly speaking, there are two important considerations that influence the legitimacy of employing partial-equilibrium welfare analysis. The first is whether optimality condi­ tions are fulfilled elsewhere in the economy. "If prices equal marginal costs in the rest of the economy, then, assuming no external economies or dis­ economies, the prices of factors used in the industry under consideration will reflect their value to society elsewhere. If, however, prices exceed marginal costs elsewhere, the private costs of expanding the output of this industry will understate the real costs to society" (Currie, Murphy and Schmitz 1971, p.788). For example, if other industries are monopolistic, it may not be best for price to equal marginal cost in the sector being studied. Second, even if price equals marginal cost elsewhere, the uncounted welfare effects in factor and product markets elsewhere can be substantial if large adjustments in those sectors occur as a result of changes in the sector being studied. The "second-best" issues arising from market distortions elsewhere, and other multi-market impacts, are not really deficiencies of economic surplus analysis, which can (at least in principle) be modified to reflect their effects. Rather, they are problems that may invalidate an economic surplus analysis if empirically inappropriate assumptions are made in the face of data limita­ tions or other constraints on an analysis. Two criticisms of economic surplus (existence of externalities and transaction costs) are closely related to this general point in that they are criticisms of the conventional practice rather than criticims of the measures in principle. In later chapters we suggest explicit adjustments to accommodate externalities and other market distor­ tions and show more general approaches to welfare measurement, still in the economic surplus mold, that are less vulnerable to such criticisms. Externalities and Free Riders External economies or diseconomies (externalities) can arise which cause the marginal social value to differ from the private value or market price. Research Evaluation and Priority-Setting Principles 51 This type of market failure can occur when the welfare of one person is influenced positively or negatively by the consumption or production activ­ ities of another and when compensation is not made for these external effects.29 If all transaction costs were equal to zero, and all property rights were assigned, then externalities should be corrected by the market. The existence of substantial, particularly negative, externalities in the world is an indication that transaction costs are important (Mishan 1981). Unless the welfare implications of externalities are explicitly accounted for in the analysis, the usual calculation of consumer and producer surplus does not include them. A concept closely related to externalities is that of a "collective good" or "public good. " We mentioned earlier that research can represent such a good because the benefits produced by certain types of research cannot be appro­ priated entirely by those producing or financing the work. Therefore, "free riders" become a problem because firms can receive some research benefits without incurring the full research costs. They may incur some costs in screening, adaptive research, or replication, but when these costs are much less than the initial costs of development or discovery free riding is a significant problem. When this happens, incentives to conduct research are reduced and firms produce less research than is socially optimal. This provides a justification for government intervention, such as public research, and implies that decisions to invest in public research should consider the degree of "collectiveness" of each type of research. It is also the primary justification for patents. Transaction Costs and 1ncomplete Risk Markets Consumer and producer surplus, and neoclassical welfare economics in general, has been criticized for ignoring transaction costs that arise because of asset fixity (sunk costs), imperfect information (bounded rationality), and the willingness of people to profit at the expense of others (opportunism).30 The presence of fixed or sunk costs associated with capital items and people means that reallocating resources will involve adjustment costs that make it uneconomic to transfer capital and people immediately to their otherwise preferred uses when conditions change. Imperfect information arises be­ cause the future is uncertain. It can lead to incomplete risk markets, in which 29. Externalities do not refer to price effects associated with production or consumption, but are rather the consequences of unintended activities, such as pollution, that result from some other legitimate form of activity (Mishan 1981). Taxes or subsidies are often suggested as possible means of bringing the marginal social cost equal to the market price. 30. For instance, see Williamson (1985), Baumol (1986), and North (1984, 1987). 52 Research Evaluation and Priority-Setting Principles case competitive equilibrium is no longer Pareto optimal (Hart 1975; Stiglitz 1982, 1985; Runge and Myers 1985). Imperfect information and incomplete risk markets make it difficult to assess accurately the true impacts of changes in consumer and producer surplus on economic efficiency and distribution (Stiglitz 1985; Runge and Myers 1985). Because there are bounds on people's ability to calculate, it is impossible to set up complex contracts that foresee every contingency (Baumol 1986, p. 280). If it were known that participants would not try to exploit opportunities to profit at the expense of others, then contracts could be drawn up loosely and specified in more detail as circumstances became clear, but this is not the case in modem society.31 A conventional welfare analysis that ignores transaction costs will pro­ vide results that overstate the benefits of activities with high transaction costs both absolutely and relative to activities with relatively low transaction costs. New technologies produced through research can involve significant trans­ action costs because of sunk costs, imperfect information, and risk. Schmitz and Seckler (1970) and Hertford and Schmitz (1977) have pointed out that the effects of resource displacement need to be accounted for if those resources fail to find employment immediately. When a farmer's labor, for example, is displaced, there can be significant adjustment costs associated with moving the labor to a new occupation, even if it is fully employed once it arrives there. If there are adjustment costs, they have to be subtracted from the benefits. Measuring these costs is difficult. If the resource were labor, one might measure the cost to society of supporting unemployed people until they find new employment. However, it is difficult to measure the true cost to society of such adjustments. Criticisms related to transaction costs and incomplete risk markets (which are part of a broader range of criticisms of economic surplus measures that center on "market failures" or "second-best" problems) pertain more to the assumptions conventionally used in economic surplus measurement rather than the neoclassical paradigm, which can, of course, accommodate im­ perfect information, transaction costs, and so on. The real issues in deciding just how to modify the empirical analysis to accommodate such problems are empirical ones. Policy Irrelevance Several critics of consumer and producer surplus analysis argue that the concepts are irrelevant for policy analysis. Cochrane (1980), for example, 31. North (1987) points out that in primitive economies, social pressures reduce opportunism because everyone knows everyone else. Thus, transaction costs are reduced. As economies grow, transaction costs rise with the impersonal nature of exchange. Research Evaluation and Priority-Setting Principles 53 argues that producers, consumers, and policymakers understand the im­ plications of price changes but not changes in economic surplus. The "policy irrelevance" criticism tends to be based on two factors: the first is an implicit or explicit recognition of the five criticisms described above; the second stems from the way the results of consumer and producer surplus calcula­ tions are presented. They are often reported as if they came from a black box, the·important assumptions and variables that drive the results are not ex­ plained, and distributional value judgments are not made explicit. The measures of changes in consumer and producer surplus arising from agricultural research may be more useful when they are interpreted in terms of cost-reducing or yield-enhancing effects, effects on production and con­ sumption, price effects, and other factors relevant to the tYJle of economic surplus analysis being conducted. If a value judgment is made that income received by different people (consumers, producers, low- versus high-in­ come people, people from different regions, and so on) has equal worth to the nation, regardless of who recei ves it, this should be stated. Unless care is exercised to relate the consumer and producer surplus calculations to the goals and objectives of the decision makers, those calculations are likely to be treated as irrelevant. Summary Six major criticisms have been leveled against consumer and producer surplus as welfare measures. Some criticisms (ignoring transaction costs, externalities, general equilibrium effects, and certain measurement errors) can be addressed, at least partially, by refinements in the measures of benefits or costs. Value judgments cannot be avoided but can be made more explicit. Policy relevance can be improved by clearer explanation of the implications of the results (or the factors driving them) and explicit consideration of distributional and other objectives. Most procedures for assisting priority setting in agricultural research described in part III of this book use consumer and producer surplus or attempt to approximate the results of consumer and producer surplUS. All of these procedures are only approximations to the "true" money metric, cor­ rected for transaction costs, externalities, and general equilibrium effects and weighted by society's values in a policy-relevant fashion. It is impossible to obtain the truly correct welfare measure but "even those of us who sin should recognize where virtue lies."32 32. Comment by Martin Bailey in a different context during a conference on improved proce­ dures for agricultural productivity measurement at the Economic Research Service, u.s. Department of Agriculture, Washington, D.C., March 31, 1988. 54 Research Evaluation and Priority-Setting Principles 2.2.3 "Alternatives" to Economic Surplus Analysis Cost-Benefit Analysis Cost-benefit analysis is sometimes represented as an alternative to eco­ nomic surplus analysis for assessing research benefits in a partial-equilib­ rium framework. In fact, cost-benefit analysis uses the concept of economic surplus and changes in such surplus measures, either explicitly or implicitly. For example, in a formal cost-benefit analysis, when research benefits are explicitly measured as changes in consumer and producer surplus, as repre­ sented in figure 2.4, these economic surplus changes are subsequently distributed and discounted over time. Internal rates of return, net present values, or benefit-cost ratios are calculated both to capture the time value of money and to incorporate research costs so that net benefits and not just gross benefits are calculated.33 Economic surplus changes may not be explicitly measured, but economic surplus calculations are still implicitly being made when internal rates of return, net present values, or benefit-cost ratios are calculated to place a value on the extra output or the inputs saved (cost reductions) because of re­ .search.34 Moreover, one of the following two simplifying assumptions is being used.35 The extra production is valued at a single market price that assumes that the supply curve is vertical and shifts against a horizontal demand curve (figure 2.6a). Alternatively, the value of inputs saved (cost reduction) at the current level of production is calculated, which implies that a horizontal supply curve is shifting down against a vertical demand curve (figure 2.6b). The change in economic surplus is equal to abQIQo in the first case and abP1Po in the second.36 The potential advantage of employing this type of implicit consumer surplus analysis is that polar demand or supply elasticities are simply im­ posed on the analysis by assumption, eliminating the need to obtain elasticity estimates. The disadvantages are that the implicit economic surplus calcula­ tions ignore all regional and international price effects that are due to the 33. See Davis, Oram and Ryan (1987) or Norton, Ganoza and Pomareda (1987) for examples. 34. See, for example, the discussion of benefit-cost analysis in Bottomley and Contant (1988). While Bottomley and Contant do not mention economic surplus analysis, they in fact are presenting an example of simplified economic surplus analysis. 35. This type of approach was introduced as an economic surplus measure of returns to research by Schultz (I 953a, pp. 114-22) and Griliches (1958). 36. A parallel shift down of ali near supply function, as shown in figure 2.1, yields a gross benefit equal to a rectangle (area [ooch) and a triangle (area abc). The approximation used in cost-benefit analysis in the latter case (horizontal supply shifting down against vertical demand) corresponds to the rectangle, and therefore represents a lower-bound value of benefits. Research Evaluation and Priority-Setting Principles 55 Figure 2.6: Implicit assumptions in cost-benefit analysis a) The value of extra production b) The value of inputs saved Price Price o Pol---__a :.... .:,.b ___ D o QuantitylYear 0 QuantitylYear research, as well as any distributional effects. In other words, the implicit economic surplus analysis is subject to the criticisms mentioned earlier, and it adds two more. This is not to argue for or against this type of simplified economic surplus analysis, but only to point out its implications. Econometric Models Earlier we mentioned that some studies use econometric methods to estimate directly the relationship between past investments in agricultural research and extension and agricultural production or productivity in equa­ tions similar to equation 2.6 (see chapter 3 for more detail). The results may be used to indicate the value of the reduction in inputs for a given quantity of output or the value of additional output from a given quantity of inputs attributable to research spending. As in the case of benefit-cost approaches, this step of valuing additional output or savings in inputs is an implicit economic surplus analysis that makes extreme assumptions about market conditions (effectively assuming either perfectly inelastic or perfectly elastic supply). In this sense, the econometric approach is a variant of the economic surplus approach rather than an alternative to it. More generally, the econo­ metric methods (or the results thereof) can be combined in a complementary fashion with a less restrictive economic surplus model to estimate the economic consequences of agricultural research investments. 56 Research Evaluation and Priority-Setting Principles Domestic Resource Cost (DRC) Models An alternative method that has been suggested for guiding resource allocations to research while incorporating the welfare-distorting effects of government policies is a procedure called domestic resource cost (ORe) analysis.37 ORe analysis involves calculating the ratio A/(B - C), where A is the social value of nontraded inputs (e.g., fixed capital, labor, and land) used to produce a unit of the commodity, B is the social value of gross output, and C is the social value of traded inputs used to produce the commodity. Outputs and traded inputs are valued at their world prices. Therefore B - C is the foreign-exchange value of the output minus the foreign-exchange value of traded inputs used to produce it. A ORe ratio can be used as a cost-benefit ratio. If the ORe ratio is less than one, then it is socially profitable to produce the commodity. If the ORe ratio is greater than one, it is not socially profitable. The concept of a ORe ratio bears a close relationship to measures of comparative advantage because it provides a measure of the opportunity cost (in terms of domestic resources) of providing a net marginal unit of foreign exchange.38 The real appeal of a ORe ratio for policy analysis is that it provides a relatively simple measure of the social value of inputs used to generate a unit of net output valued at its true social value. Unfortunately, as currently applied, ORe analysis is seri­ ously flawed as a single or primary procedure for setting agricultural re­ search priorities. Its fatal flaw is that it ignores one of the major determinants of the social value of conducting research on a particular commodity: the number of units (hectares, animals) to which the research benefits will apply.39 Because research costs are relatively independent of the number of hect­ ares that will eventually be affected by the research results, the benefits will be substantially greater if the research affects 10,000 hectares rather than 10. The ORe does not take this into account. One could argue that if a country has a lower ORe for one commodity than for another, it should expand the number of units as well as lower the per unit costs through research. But as soon as production of the commodity expands very much, the opportunity 37. Recent applications of domestic resource cost analysis in developing countries are found in Monke and Pearson (1989) and, with reference to prioritizing research, in Byerlee (1985). For a discussion of the conceptual basis and limitations of domestic resource cost analysis see Bruno (1972), Srinivasan and Bhagwati (1978), and Tower (1984). 38. ORe is not it true measure of comparative advantage because no attempt is made to remove policy distortions on the same commodity or on other commodities in other countries that in tum are affecting world prices. 39. It also ignores the probability of research success on a particular commodity and the adoption rate of research results, but these items could be factored into a ORe analysis. For example, the denominator of ORe could be mUltiplied by the probability of success. Research Evaluation and Priority-Setting Principles 57 costs of the domestic resources used in production would change, input substitution would take place, and the initial DRC ratio would change. Thus a DRC is an accurate cost-benefit ratio only for marginal changes in produc­ tion. In addition, if the country is a large exporter or importer, world prices could also be affected by production changes in the country. These price changes cannot be calculated without elasticities. But if elasticities are available, one might as well conduct an economic surplus analysis and include the policy distortions directly. The Congruence Rule The congruence rule has been used widely as a crude procedure for allocating resources to research. In this approach, funds are allocated so as to equate research intensities - research investment in relation to the value of output in gross or value-added terms - across areas. That is, research is funded in proportion to the value of production, an approach that requires minimal information. Congruence will result in maximum economic surplus from the portfolio of research investments when ail projects or programs are subject to the same per unit research (knowledge) production function (Evenson 1991).40 The conditions under which this would occur are unlikely to be met, but in some cases there might be insufficient information available (on the flows of benefits and costs from alternative projects and programs) to justify a more complete analysis. In such cases, an application of the congruence rule might be consistent with an ultrasimplified economic surplus approach in that at least some account is taken of the scale of the industry. 2.3 Determinants of the Size and Distribution of Benefits and Costs In section 2.1 of this chapter, we discussed the nature of agricultural research as a process that augments the stock of knowledge, which provides a flow of services as inputs to agricultural production. In that context, agricultural research is a component of a dynamic agricultural production system whose effects are spread over long periods. Using the familiar framework for supply and demand, we showed how these dynamic relations can be represented as a series of comparative-static, market equilibrium displacements. In section 2.2, we introduced the economic surplus approach 40. Evenson (1991) demonstrates this with a research production function (or "discovery func­ tion") of the semi-logarithmic form: D = IX + pln(R), where D is the increment to knowledge due to research, R. 58 Research Evaluation and Priority-Setting Principles to measuring the annual economic welfare effects of a research-induced supply shift. Following a critical appraisal of that approach, we concluded that, although they are imperfect, economic surplus methods are the best available means for measuring the flows of benefits and costs of agricultural research. The purpose of this section is to extend the discussion by elaborat­ ing on (a) the key elements of the economic surplus approach and the critical choices that influence the estimates, (b) the extension of the approach to incorporate alternati ve market characteristics, and (c) an examination of the distributional implications of agricultural research. 2.3.1 Critical Assumptions in the Model In figure 2.4 we schematically showed measures of producer, consumer, and total economic surplus changes associated with a research-induced supply shift. In order to measure those welfare areas explicitly, it is necessary to define explicit mathematical functions to represent supply-and-demand equations and the supply shift. In most instances it is not possible to measure these relationships econometrically with high levels of precision (if at all) and it is necessary to make assumptions rather than rely on data alone. But some aspects ofthe measures are sensitive to assumptions about (a) supply­ and-demand elasticities, (b) functional forms of supply and demand, (c) the nature of the research-induced supply shift, and (d) the nature of the technical change. That assumptions must be made is inevitable. That they are import­ ant may be regrettable but need not be fatal to the analysis. The purpose of this section is to illustrate the nature of the sensitivity of results to assump­ tions in order to assist analysts in making informed choices about assump­ tions and in focusing their sensitivity analyses in appropriate directions. Elasticities of Supply and Demand To recapitulate, in figure 2.4, supply and demand are represented by linear functions. Adoption of new technology resulting from research causes sup­ ply to shift in parallel from So to S,. The resulting welfare changes are (a) MS (change in producer surplus) = area hpb!, - PrpbP, = P,bcd, (b) IlCS (change in consumer surplus) = area PoabP" and (c) IlTS (change in total surplus) = IlCS + MS = area !nab!, = P(pbcd. Mathematical formulas to measure these areas in terms of the size and nature of the supply shift and market parameters are provided in chapter 4. For the present purpose it is sufficient to look at these areas informally. What are the effects of assumptions about supply-and-demand elasticities on the welfare measures? First, it is helpful to notice that in this instance, Research Evaluation and Priority-Setting Principles 59 each of the surplus areas (IlPS, llCS, and llTS) can be represented as the sum of a rectangle and a triangle. In each case the width of the rectangle is the original (preresearch) quantity, QI), and the width of the triangle is equal to the change in quantity produced (IlQ =Q I - QI). The heights of the rectangles and triangles depend on the absolute size of the vertical shift in supply, the per unit cost saving due to research (for llTS), the change in price (for llCS), and the difference between the change in cost and the change in price (for MS). In the case of total benefits, elasticity assumptions do not affect the rectangle, PrPcd, but they do affect the triangle, abc: the more elastic supply or demand is, the larger the triangle and the larger the total welfare gain. As a practical matter, however, in the context of estimating research benefits, the triangles are typically very small relative to the rectangles and total benefits are relatively insensitive to elasticities of supply and demand.41 Elasticity assumptions (or estimates) are much more important in relation to the distribution of benefits. In particular, the more elastic supply is relative to demand, the greater the consumer share of total research benefits (and the smaller the producer share) and vice versa. In the extreme case of perfectly elastic supply with downward-sloping demand (as might occur in a compet­ itive industry in a closed economy or in a large-country case), all of the research benefits go to consumers because the research-induced change in price is equal to the research-induced cost-saving and there is no producer surplus either before or after the supply shift (figure 2.7a). When demand is perfectly elastic (as in the case of a small, open economy which cannot affect the world price for the commodity, probably the predominant case for agricultural goods), all of the benefits go to producers because there is no research-induced reduction in price (figure 2.7b). When the elasticities are of equal magnitudes (albeit opposite signs), the benefits from research are shared equally between producers and consumers (figure 2.7c). These conclusions about the role of elasticities, obtained using linear supply-and-demand functions with a parallel shift, apply fairly generally but there are some important additional considerations when different types of supply shifts are used. In particular, when we have a pivotal supply shift, whether producers benefit at all from research depends on the elasticity of demand - when demand is inelastic, producers lose! 41. For a lOOK percent shift down of a linear supply function, the rectangle is equal to KPoQo and the triangle is 112 K2poQo£Il/(E+1'), where E is the supply elasticity and 1') is the absolute value of the elasticity of demand. The triangle is equal to SOKET)/(E+1')) percent of the rectangle. When the supply and demand elasticities are equal to one (or smaller), the triangle is equal to 2SK percent (or less) of the rectangle. For commonly used research-induced supply shifts of less than 10 percent of the initial price (i.e., K = 0.1), the triangle would be less than 2.S percent ofthe rectangle. Even when demand is perfectly elastic (1') = +00) the triangle is only SOK percent of the rectangle; i.e., S percent when K = 0.1. 60 Research Evaluation and Priority-Setting Principles Figure 2.7: Effects of elasticities on distribution of benefits (a) Perfectly elastic supply (b) Perfectly elastic demand (c) Equal supply-and-demand elasticities Price Price Price 50 5 , 5, Po Po D Po P, P, 5, 10 D D I, o Quantityl Year Functional Forms of Supply and Demand Figure 2.4 uses linear supply-and-demand curves that make it easy to calculate the geometric areas of surplus changes using simple algebra. Linear supply-and-demand curves have been used for that reason in the majority of studies of research benefits. With such curves, the elasticities change as quantity changes along the curve, and one must be explicit about where the assumed elasticities apply - before or after the research-induced market displacement. One hazard with linear supply functions is that, when the function is inelastic at the supply-and-demand equilibrium, extrapolating back to the origin implies a negative intercept on the price axis (implying that positive quantities would be supplied at negative prices). Various authors have criticized the use of linear supply curves with point elasticities of less than one for that reason.42 A linear curve does imply a negative intercept, but this can be averted by kinking the supply curve (Rose 1980). The economic surplus calculated after kinking the supply curve at the original quantity is the same as the economic surplus calculated without the kink.43 The real problem is the poor estimate of the proportionate cost reduction due to research. With an inelastic supply curve, the proportionate cost reduction 42. See Kim, et al. (1987), Ciodyn, Brennan and Johnston (1 987), and Voon and Edwards (I 991 c). 43. This is true whether the supply shift is pivotal or parallel. Research Evaluation and Priority-Setting Principles 61 implied by a proportional rightwards shift of supply can be unreasonable, giving rise to overestimated returns.44 The introduction of arbitrary kinks of supply (as by Rose 1980 and Hertford and Schmitz 1977) is effectively an abandonment of the linearity assumption. Alston and Wohlgenant (1990) suggest that when a parallel shift is used, the functional form is largely irrelevant, and that a linear model provides a good approximation regardless of the true functional form of supply. Based on that, it is safe to proceed using the algebra from the linear model, ignoring the question of any implied negative intercepts on the price axis so long as the supply shift is parallel. The main alternative assumption that has been used is one of supply-and­ demand curves of constant elasticity (e.g., Ayer and Schuh 1972; Scobie and Posada 1978). The constant-elasticity model has the supply function, regard­ less of its elasticity, passing through the origin. Thus, as with the linear model (in the case of inelastic supply), the constant-elasticity model has some implausible implications when we extrapolate far from the initial equilibrium. Typically the constant-elasticity assumption is combined with an assumption of a proportional (or pivotal) supply shift - because it is difficult to use a nonproportional shift with a constant-elasticity model - and, as we shall see below, that is the most important consequence of the functional form choice. A constant-elasticity supply function with a propor­ tional supply shift is shown in figure 2.8. The surplus areas are !l.TS =a rea Oab, !l.CS = area PepbPI> and !l.PS =! l.TS - !l.CS =a rea Oab - area PoabP •. What are the implications of choosing linear versus constant-elasticity forms? The nature of the consequences will depend in part on the approxi­ mating formula being used. For example, compared with using a constant­ elasticity model, the formula for a kinked supply curve provided by Rose (1980) will increase the economic surplus measure for an elasticity of supply greater than one while the formula provided by Hertford and Schmitz (1977) (for a pivotal shift) will reduce the measured economic surplus. The opposite effects would occur for a supply elasticity of less than one. The results of Rose (1980) and Hertford and Schmitz (1977) differ because Rose kinks the supply curve at the original quantity while Hertford and Schmitz kink it at the new price. Some authors have used linear approximations to calculate 44. Actually, if one uses an elastic supply curve, the benefits can be underestimated as well, even though the curve intersects the vertical axis at a positive price. One way that these overestimates or underestimates arise is when the proportionate vertical supply shift is calculated from the proportion­ ate horizontal supply shift using the expression K = lie, where K = the vertical supply shift, I = the horizontal supply shift, and e = the elasticity of supply. This relationship only holds in a small neighborhood around the original price and quantity and can give very unrealistic results for linear supply curves that are either very elastic or very inelastic. The relationships between these alternative measures of research-induced supply shifts are dealt with in detail in chapter 5. 62 Research Evaluation and Priority-Setting Principles Figure 2.8: A proportional supply shift in a constant-elasticity model Price PI ------------------- D o Quantity/Year economic surplus under an assumption of a constant supply elasticity. For example, the formula used by Akino and Hayami (1975) as a linear approx­ imation to a constant-elasticity function overestimates the economic surplus for elasticities of supply less than one and underestimates the economic surplus for elasticities of supply greater than one. A third alternative has been suggested by Lynam and Jones (1984) and used by Pachico, Lynam and Jones (1987). This is a constant-elasticity form based on a positive intercept with the price axis, so, in fact, it is not a constant-elasticity function. This model has the virtue of greater realism than either the linear or constant-elasticity models - in that it allows a positive price intercept independent of elasticity assumptions - and it is flexible in relation to the nature of the research-induced supply shift. For instance, in - figure 2.9 we have a verticaly parallel shift of a constant-elasticity form of supply function. In practice the advantages of this flexibility may be illusory because of difficulties in identifying the nature of the research-induced supply shift. In addition, since this approach requires a nonlinear algorithm for its solution, the extra effort involved in solving for price and quantity changes may not be warranted (Alston and Wohlgenant 1990). Of course, none of these is likely to be the true and generally unknown Research Evaluation and Priority-Setting Principles 63 Figure 2.9: A parallel shift down ofa "constant-elasticity" supply function Price D o QuantityIYear functional form. The pertinent question is whether the functional form used is an adequate approximation for the purpose. It turns out, empirically, that measures of total research benefits and their distribution between producers and consumers are quite insensitive to choices of functional form. They are much more sensitive to the related but separate choices concerning the nature of the research-induced supply shift and elasticities.45 The Nature of the Research-Induced Supply Shift There has been a great deal of discussion in the literature about the effects of different types of research-induced supply shifts on the size and distribu­ tion of research benefits, and rightly SO.46 This choice in the analysis is crucially important, and by comparison, the choices about functional forms and elasticities pale into insignificance. For example, given a linear supply 45. See Lindner and Jarrett ( 1978), Norton and Davis (1981), Alston and Wohlgenant (1990), and Voon and Edwards ( 199 1c ) for details and evidence. 46. The point was comprehensively addressed by Lindner and Jarrett (1978), leading to some debate, with comments from Rose (1980) and Wise and Fell (1980) and a reply by Lindner and Jarrett ( 1980). The matter has never been resolved satisfactorily. Norton and Davis (1981) provide a summary discussion of the history of the question and the main points. 64 Research Evaluation and Priority-Setting Principles function, total benefits from a parallel shift are almost twice the size of total benefits from a pivotal shift (of equal size at the preresearch equilibrium). When supply shifts in parallel, producers always benefit from research unless supply is perfectly elastic or demand is perfectly inelastic, and even in these extreme cases, producers are no worse off as a result of research. On the other hand, as noted above, with a pivotal shift, producers benefit only when demand is elastic; when demand is inelastic, producers necessarily lose from a pivotal supply shift (e.g., Lindner and Jarrett 1978). Unfortunately, economic theory is not informative about either the func­ tional form of supply and demand or the functional form (parallel, pivotal, proportional, or otherwise) of the research-induced supply shift. The indus­ try curve is based on the aggregation of supply curves for individual firms. Shifts in the industry curve depend on the effects of new technologies on the marginal costs of existing firms and on entry and exit of firms. One would need to examine the characteristics of individual firms that affect marginal costs and technology adoption in order to predict which types of firms would benefit from a particular new technology. In addition, with current tech­ niques and typically available data, it is not possible to settle these questions econometrically. We might hope to obtain plausible estimates of elasticities at the data means, but definitive results concerning functional forms are unlikely and it is impossible to get statistical results that can be extrapolated to the price or quantity axes (i.e., the full length of the function) with any confidence. Thus, assumptions about the nature of the research-induced supply shift are unavoidable. Our conclusion is that it is important to be aware of the consequences of different assumptions.47 Our preference - in the absence of the information required to choose a particular type of shift - is to follow Rose's (1980) suggestion and employ a parallel shift. Rose (1980, p. 837) argued that "For most innovations, the best information available may be a cost-reduction estimate for a single point on the supply curve .... [It] is unlikely that any knowledge of the shape of the supply curve, or the position at which the single estimate applies, will be available. The only realistic strategy is to assume that the supply shift is parallel." We find the arguments of Rose persuasive, and therefore we are inclined to assume vertically parallel research-induced supply shifts. Under this assumption, the functional forms of supply and demand are unimportant and it is convenient to use a local linear approximation, as suggested by Alston and Wohlgenant (1990).48 47. One could always use a pivotal shift rather than a parallel shift in order to generate conservative estimates of total economic benefits. 48. A parallel shift implies the change in average cost equals the change in marginal cost at every point along the curve. Vince Smith (pers. comm.) has pointed out that often we have a reasonable Research Evaluation and Priority-Setting Principles 65 An additional advantage is that this assumption simplifies some calcula­ tions and permits consistency in the evaluation among projects and programs across a range of commodities. For priority-setting purposes, this consis­ tency is important. Errors in assumptions are less important in relation to ranking priorities than they are in absolute terms. A side issue is whether the shift is expressed as vertical (in the price direction) or horizontal (in the quantity direction). Of course any supply shift may be expressed, with care, equivalently in either way. The issue has arisen in going from experimental or industry yield increases to supply shifts (see chapter 5). Expressing a K percent yield increase (or cost saving) as either a K percent vertical shift or a K percent horizontal shift has different im­ plications unless the supply elasticity is unitary.49 2.3.2 Extensions to the Basic Model The basic economic surplus model may be modified in various ways (a) to disaggregate effects across multiple markets horizontally (across geopo­ litical regions, socioeconomic groups, or commodities) and vertically (across stages of production or among factors of production), (b) to allow interregional spillovers of research-induced price changes and research re­ sults, (c) to incorporate general-equilbrium feedback and economy wide adjustments, (d) to accommodate market distortions and the effects of research on the creation (or amelioration) of market distortions caused by government intervention or production externalities, and (e) to account for the costs of taxes to finance government spending. International and Interregional Trade When goods are traded internationally, the basic economic surplus model measures total research benefits from a global perspective. With free trade, the total supply-and-demand functions are the horizontal summations of the underlying supply-and-demand functions of individual nations or other sub­ aggregates, respectively. In this setting it is a relatively simple matter to estimate of the change in average cost at the current equilibrium, and this is what counts for estimating the rectangle that dominates gross annual research benefits (GARB). A similar point was made by Cooke and Sundquist (1993). Thus, we do not need to know what has happened to marginal or average cost at any other point along the curve. An error in estimating the change in marginal cost, given an estimate of the change in average cost, will not cause important errors in GARB (it only affects the comparatively small triangle) although it may have important implications for distribu­ tion. 49. The equivalent horizontal supply shift, J, for a K percent vertical supply shift is given by using the definition that dQIQ = EdP/P and, therefore, J = EK, where E is the elasticity of supply. 66 Research Evaluation and Priority-Setting Principles disaggregate the consequences of research among regions or countries. The total consumer surplus measured behind the total demand curve is the sum of the component consumer surpluses measured behind the component demand curves. Similarly, total producer surplus may be decomposed into subcomponents. The situation may be analyzed in a disaggregated fashion either by modeling all nations in a multicountry model that explicitly involves the supply-and-demand curves of each nation (i.e., n markets for n countries), by modeling supply and demand in a country of interest and in an aggregate "rest-of-the-world" (ROW) (i.e., two markets), or by using excess supply and demand from the home country and the ROW (i.e., one market). In addition, the same approaches could be applied instead to analyze trade patterns and the distribution of research benefits among regions of a country, with regions playing the role of nations in the multinational modeling approach. All of these approaches are conceptually identical and, if implemented consis­ tently, in a particular application give identical results for total research benefits in both the ROWand the home country. However, the different approaches yield different disaggregated details. In deciding how to analyze research benefits for a traded good, the main question is how much detail is warranted. While it is conceptually straight­ forward to disaggregate to any level, in practice the information problems­ of obtaining suitable measures of quantities, prices, and elasticities of supply and demand - become greater relatively quickly as one proceeds to finer disaggregations. Thus, from this perspective, it is sensible to disaggregate to the point where the requirements of the analysis are served, but no further than that. From another perspective, it may be sensible to disaggregate somewhat further in order to get a more accurate picture of the research-in­ duced supply shift. When a technology is applicable in multiple regions that differ in their responses to the new technology, it may be necessary to dis aggregate some aspects of the study to avoid aggregation biases, even if all regions are not of specific interest from a research-policy or priority-set­ ting perspective. From a global perspective, the total benefits from research and their distribution between "consumers" and "producers" are as defined in the basic model and are sensitive to assumptions about elasticities, functional forms, and so on, as in the basic model. When the interest is in returns from a narrower (say national or subnational) perspective, the story is altered somewhat. In addition to the factors identified in the basic model, the key determinants of national research benefits in the home country are (a) the extent to which the research r.esults can be adopted by the ROWand (b) market power in trade (i.e., the elasticities of supply and demand that Research Evaluation and Priority-Setting Principles 67 determine the distribution of benefits among nations).5o These effects may . be considered in terms of technology spillovers and price spillovers. Technology and Price Spillovers across Geographical Areas The results of agricultural research can have two major types of spatial spillovers. First, new knowledge or technologies produced in or targeted for one country or region can spill over into other countries or regions. Second, new technologies can affect prices, not only in the region and country adopting the technology, but in other regions or countries where the product is produced or consumed. These spillovers influence both the size and distribution of research benefits and should be considered when research priorities are set. The ease of technology transfer across geographic boundaries depends on a variety of factors, as described by Ruttan and Hayami (1973) and by Evenson and Binswanger (1978). The relative costs of direct technology transfer and of adaptive versus comprehensive research programs, the com­ plementarity between screening of existing technologies and carrying out applied research, the environmental sensitivity of the technology, and differ­ ences across locales in their factor scarcities are particularly important (Evenson and Binswanger 1978). Technologies themselves, as well as the capacity to create and adapt technologies, are potentially transferable. Basic research is often less environmentally specific and thus more likely to be transferred than applied research. The fact that certain research results can be transferred from one country to another or from one region to another means that the calculations of total benefits of research to the world as a whole should consider these spillovers. Individual countries, however, have little incentive to consider the effects of their own research on other countries, except in so far as the transfer of new technology to other countries may reduce their own competitive positions in the market. Countries also have an incentive to borrow technologies when they can obtain results for less than the full cost. Smaller countries in particular may transfer in a high proportion of their new technologies because they usually cannot afford extensive research programs. There is thus an incentive to underinvest in research in the world as a whole, partic­ ularly for basic kinds of research, which suggests a need for international agricultural research centers that can generate technologies applicable to 50. Trade also affects distribution of benefits between producers and consumers in the home country by changing the effective elasticities and thus modifying the consequences of technical change for prices and quantities. These issues are described and analyzed. for example. by Edwards and Freebaim (1982. 1984) and Davis. Oram and Ryan (1987). 68 Research Evaluation and Priority-Setting Principles several countries. This is one rationale for the current system of centers supported by the Consultative Group on International Agricultural Research. The existence of technological spillovers has important implications for resource allocations to research within individual countries. First, individual countries may be wise to consider new technologies being produced in international centers and in research systems in countries with similar human, natural, and physical resource bases so that research programs can be structured to complement the research conducted abroad rather than duplicating it excessively. Second, local research capacity may be needed just to enable the country to transfer in and adopt technologies developed elsewhere. Third, the differential effects of research on agricultural produc­ tivity across regions in a country may need to be considered. International and interregional adoption of technologies implies that re­ search in one location induces supply curves to shift out in other countries and regions, as discussed by Edwards and Freebairn (1984) and by Davis, Oram and Ryan (1987). This results not only in changes in agricultural productivity, but it also effects output prices across countries and regions. For example, if new technologies are adopted in one region (country) but not in another, producers in nonadopting regions (countries) can experience price reductions without a corresponding reduction in costs. Consumers can also benefit from the technology even though they live in regions (countries) not adopting new technologies.51 These factors may need to be considered in an ex ante assessment of research priorities. They also imply a need to consider the consistency between regional and national priorities within a country. Large countries may find it difficult to achieve a consensus on research priorities because of a divergence between regional and national priorities. Even for research that is regionally specific in its application and impact, a country must decide on its relative allocation across regions. For research with both technical and price spillovers, reconciling regional and national priorities is even more difficult. A desire to help landless laborers in one region may'conflict with a second objective of increasing agricultural pro­ ductivity in another region in order to provide low-cost food for urban consumers. The procedure developed for research priority setting may need to consider, and preferably to quantify, the tradeoffs associated with alterna­ tive regional allocations of research resources. 51. See Davis, Oram and Ryan (1987) and chapter 4 for a graphical and mathematical description of these effects. Research Evaluation and Priority-Setting Principles 69 Processing and Marketing Research Two conceptual issues arise with respect to processing and marketing­ sector research. First, it is frequently argued that because most of the final value of agricultural products is added beyond the farm gate, much of public agricultural research should focus on the postharvest sectors. Second, some have argued that, under reasonable assumptions, producers, processors, and consumers receive the same share of total industry benefits from research, irrespective of whether the research generates new technologies at the production, processing, or marketing stage. Considering these two, related, ideas has led some to infer that agricultural research has neglected opportu­ nities beyond the farm gate. What determines if research affecting the processing and marketing sectors should receive priority and whether it makes a difference if research affects agricultural commodities pre- or postharvest? Figure 2.10 is a schematic representation of production and price determi­ nation in a multistage production system for a traded good. The two stages of production include farming (that uses inputs supplied by farmers as well as purchased inputs) and processing and distribution (that uses the farm product from the first stage plus marketing and other inputs). The stages are integrated in that production and the prices of inputs and outputs at all stages of production are mutually dependent. Potential beneficiaries from research include final consumers (including foreign consumers) and suppliers of all factors used in production (farmers, farm input suppliers, and marketing input suppliers). Research may affect the supply of any of these inputs and may also affect the technology in the two stages of production. This is very similar to the multistage models analyzed by Freebairn, Davis and Edwards (1982) and Alston and Scobie (1983). As discussed in chapter 1, the rationale for public-sector involvement in particular types of research rests on the assumption that the private sector has insufficient incenti ves to conduct the socially optimal quantity of those types of research. If private research incentives are insufficient, the priority placed on research in marketing and processing would then depend on the ability of researchers to generate greater net benefits in marketing and processing (by contributing to efficiency, distributional, and security goals) than in research directed at primary production. Advocates of a greater emphasis on process­ ing and marketing research usually imply that efficiency gains would be greater if research efforts focused more on postharvest activities. This is an empirical question. However, the issue of private incentives is often ignored. Private-sector incentives may be greater in marketing and processing sectors than in primary production because of a greater ability to patent and license 70 Research Evaluation and Priority-Setting Principles Figure 2.10: Schematics of multimarket effects of R &D Price and quantity of ret;li l product Pricing Production of ." ~£' jinalgoods ~,' r Processing technology t ~~"'**",~. *.::: Prices and quantities of farm product and processing inputs Demand for Supply of farm product Pricing farm product Production .of farm product Prices and Demand for quantities of inputs supplied I¢:=:=; ~ Inputs supplied by farmers Pricing by farmers inputs used by farmers ~----.....-.:­-- Demand for Supply of purchased purchased farming inputs Pricing farming inputs research results and thereby to appropriate the returns to research. Much processing and marketing research is related to mechanical items and new products, both of which may be easier to patent than biological innovations. In relation to the question of who receives the gains from research at different stages in the production, processing, and marketing chain, the degree of substitutability among inputs in processing and marketing be- Research Evaluation and Priority-Setting Principles 71 comes quite important. Freebairn, Davis and Edwards (1982) argued that, in a processing industry using a farm and a nonfarm input, producers, proces­ sors, and consumers receive the same share of total benefits regardless of whether the research shifts the farm or processing supply curves. Alston and Scobie (1983) pointed out that this result only holds when substitution between inputs in processing is not possible. They developed a framework, allowing input substitution in processing, that can be used to assess the benefits from research at different stages in the marketing chain, and their analysis showed that a small degree of input substitution can have a large effect on the distribution of research gains among primary producers and processors. As the elasticity of substitution increases, producers generally receive larger benefits if new technologies are directed at their sector. 52 For most agricultural products, the degree of substitution possible be­ tween a raw product and processing inputs is small. However, some substi­ tution is possible. An important source of input substitution in the beef subsector, for example, has been the use of new technologies such as boxed beef to reduce shrinkage and spoilage. Mullen, Wohlgenant and Farris (1988) found an elasticity of substitution of 0.1 between beef and processing inputs in the U.S. beef sector that, although seemingly small, was enough to make a significant difference. Their empirical results reinforced the analyti­ cal findings of Alston and Scobie (1983). Other assumptions can be relaxed as well. Freebairn, Davis and Edwards (1983) consider the case in which nonfarm inputs and marketing services are not available in completely elastic supply. They also consider the case of a monopolist in the marketing or input supply sector. Furthermore, calculation of benefits to research at the sector that supplies inputs to farmers (e.g., fertilizers or pesticides) involves additional complications because these inputs are used to produce several commodities and some of the commodi­ ties are produced jointly (e.g., meat and milk).53 It is unlikely that practition- 52. This analysis of the "functional distribution" of research benefits has been extended in a number of recent studies, including Freebaim, Davis and Edwards (1983), Mullen, Wohlgenant and Farris (1988), Mullen, Alston and Wohlgenant (1989), and Holloway (1989). These studies have shown that the total research benefit and the functional distribution of research benefits among factors employed at different stages of production (or among nations when semi-processed products are traded) depend on several things in addition to those represented in the basic model unless the factors are used in fixed proportions. These additional aspects include (a) the stage of the production process to which the research applies, (b) the nature of the research-induced technical change, and (c) more detailed technological and market parameters. These results have implications for the allocation of private or public research resources within a country between programs applying at different stages in multistage production processes. 53. See Duncan (1972) for an example of a study that evaluates research on an input (pasture) which is used by another farm commodity (grazing livestock). 72 Research Evaluation and Priority-Setting Principles ers conducting priority-setting exercises will have the time and resources to consider all these potential effects for every commodity and type of research. Nevertheless, there are cases, particularly for certain types of postharvest research, where these refinements are necessary. In less-developed countries, these issues may be thought to be of reduced importance (compared, say, with the United States or Australia) because research investments are primarily directed at production agriculture, per se, and the functional distribution of returns from research directed at further processing may be largely irrelevant for policy. However, anyone conduct­ ing an exercise in research evaluation or priority setting ought to be con­ scious of the implications of different types of research directed at different stages of production for the size and distribution of benefits, even when there is a strong presumption that all research efforts will be directed at farming. There are four reasons for this: (a) opportunities beyond the farm gate ought not to be neglected for the wrong reasons, (b) most places do engage in some research into postharvest handling and storage technology, (c) it is becoming increasingly fashionable to redirect research resources away from traditional agricultural technology to issues beyond the farm gate (often in pursuit of "value-adding" objectives), and (d) components of a new technology that results in direct cost savings off-farm may be embodied in outputs that are generated on-farm (e.g., new varieties that have both improved yields and improved storage or processing characteristics). In addition, international trade in agricultural products means that the international distribution of benefits from research among nations is con­ nected to the global distribution of benefits from research between producers and consumers. When goods are traded in raw or semiprocessed form, the international distribution of benefits also depends on where the research is applied in the chain of production (e.g., Alston and Mullen 1992). Thus, even when the interest is only in national research benefits, in the context of traded goods, the functional distribution of research benefits among stages of production or among factors of production becomes relevant. Effects on Factor Returns The distribution of income among factors of production is influenced by technical change because of the incentives that technical change provides for reallocating inputs among production alternatives. These incentives arise within agriculture and between agriculture and the nonagricultural sector. Several forces influence the impact of agricultural research on factor mar­ kets. The most important ones are the elasticity of demand for the commodity in question, the input bias ofthe technical change, the elasticity of supply of Research Evaluation and Priority-Setting Principles 73 factors, the elasticities of substitution between factors of production in each sector and the input mixes for the various commodities.54 When the demand curve has a constant unitary price elasticity, the value of the commodity and therefore the value of the inputs to produce it are constant, no matter how the supply curve shifts. Therefore, total resources are neither entering nor leaving the sector and, with a neutral technical change, the composition of those resources is unaffected. However, when the demand is elastic, resources are drawn into the sector as supply shifts out, and when demand is inelastic, resources are forced OUt.55 When new technol­ ogies are biased toward or against a factor, even if total resources are not displaced, the factor being biased against will lose relative to the other factor(s) and will probably lose absolutely (unless the product demand is very elastic). The elasticity of substitution between factors influences the impact of a technical change. The greater the elasticity of substitution between two factors, the more equal will be the effect of the technical change on both factors, regardless oftheir relative elasticities of supply. Finally, the relative factor intensities of the sector affect the size and distribution of research benefits. However, this effect in tum depends on the elasticity of demand for the product and the factor bias of the technical change. These qualitative effects are demonstrated formally in chapter 4 using a simple two-factor model. Cross-Commodity Effects Benefits and costs of research-induced technological changes might not be confined to the producers and consumers of the commodity whose production is affected directly by the new technology. Research that affects one commodity can also affect other commodities through cross-price ef­ fects, particularly on the demand side, and through technology spillovers. The cross-price effects on commodities that are substitutes or complements in demand will be most important if the commodity at which the research is directed has a relatively inelastic demand (i.e., a commodity for which the price will fall the most following the adoption of a new technology). The cross-price effects on supplies of commodities that are substitutes in produc- 54. Note that we have not mentioned the initial labor intensity or the number of people employed in producing the commodity. Research administrators often voice the desire to conduct more research on those commodities that employ a higher proportion of the agricultural work force than others as a means of increasing employment. The fact is, however, that in many cases, research on commodities that currently have a high labor content will reduce labor's share of income and employment unless conditions favorable to employment also hold. 55. The product demand curve is more likely to be elastic for commodities that are predomi­ nantly exported or imported than for those that are primarily produced and consumed internally. 74 Research Evaluation and Priority-Setting Principles tion may tend to be small because the lower per unit costs of production due to new technologies are partly offset by lower prices for the commodity affected. Most studies on research priority setting are not likely to incorporate many cross-commodity effects. Cross-price elasticities are often unavailable and cross-commodity spillovers of technologies may be difficult to judge. Often it is reasonable (as well as convenient) to assume that the producer and consumer surplus measures in the market of direct interest represent the total effects, which can be true if the supply and demand curves are defined sufficiently generally (as shown by Just, Hueth and Schmitz 1982). Alterna­ tively, conceptually at least, a more explicit accounting for these cross-com­ modity effects involves relatively straightforward extensions of the models presented in this and later chapters and may be developed for a limited set of commodities in some countries - depending primarily on the availability of data (e.g., see Chang et al. 1992). Quality Change Some types of agricultural research are intended to improve the quality of a commodity. An approach suggested by Unnevehr (1986, 1990) for evalu­ ating the effects of research on quality is to estimate implicit prices for individual quality characteristics. Quality-enhancing research supposedly shifts the product demand curve upward to reflect the notion that consumers will demand more of a product at each price if it contains a higher proportion of a relatively higher-priced characteristic (see also Voon 1991; and Voon and Edwards 1991 a). What really happens, however, is that the supply of the characteristics changes; the demand-shift representation is useful only under a limited set of circumstances. An alternative way to conceptualize the issue is to view the market supply­ and-demand curves for the commodity as an aggregation of a set of supply­ and-demand curves for different types (or qualities) of the product. Each type has its own homogenous set of characteristics. Research that affects quality lowers the cost of supplying the type of the commodity with the improved characteristics. The aggregate market supply curve for the commodity also shifts to the right, and more is demanded because the aggregate price has fallen Gust as the price of the higher-quality type of the commodity has fallen). At the same time, substitution effects would lead to a reduction in demand for the lower-quality types. Because of these cross-price distribu­ tional effects, in many cases the aggregate net change in economic surplus may be relatively small compared with the distributional effects on produc­ ers versus nonproducers of the higher-quality type of the commodity. Research Evaluation and Priority-Setting Principles 75 Market Distortions It can be argued that countries should focus research resources on those commodities in which they have a comparative advantage. Unfortunately, comparative advantage can be difficult to assess because it can be obscured by misaligned exchange rates, tariffs, subsidies, and other policies, and it changes over time in response to a number of factors, including research-in­ duced technical change. Therefore, assessments of comparative advantage must be made either with subjective judgments or with detailed analyses of the various macroeconomic and sector-specific policies affecting the coun­ tries under comparison. The latter is preferable because the existence of these policies implies some social losses that modify research benefits when supply curves shift out due to new technologies. International prices them­ selves are influenced by the policies of all countries. However, unless those policies are expected to be changed in a particular direction, an individual country usually is forced to take those international policies as given, particularly when assuming that the country is a relatively small producer in the world market for the commodities. Harberger's first two postulates for the use of economic surplus analysis are that supply-and-demand functions represent both private and social benefits and costs. When markets are distorted, either by government poli­ cies or by externalities in production or consumption, these postulates are be violated and an extra effort is needed to obtain measures of social and private costs and benefits of production and consumption. Building on some initial work by Edwards and Freebairn (1981), Alston, Edwards and Freebairn (1988) presented a conceptual overview of the effects of a range of commod­ ity programs on the size and distribution of research benefits. Subsequent work has provided some further theoretical results and empirical evidence (e.g., Oehmke 1988b, 1991; Zachariah, Fox and Brinkman 1989; de Gorter and Norton 1990; Voon and Edwards 1991b; Murphy, Furtan and Schmitz 1993; Alston and Martin 1992). This literature has identified the potential importance of market distortions for affecting both the size of research benefits (properly measured) and the potential size of errors introduced when the distortions are ignored. The results suggest that it may be unwise to ignore market distortions when conducting an assessment of research bene­ fits, especially when the distribution of benefits and costs is being measured. Some more recent, related literature has argued that it is logically incon­ sistent and perhaps unrealistic to treat market distorting policies and research policy as independent. Gardner (1988), Oehmke (1988b), Oehmke and Yao (1990), Roe and Pardey (1991), Alston and Pardey (1991, 1993) and de Gorter, Nielson and Rausser (1992) have argued for the use of political 76 Research Evaluation and Priority-Setting Principles economy models in which research policies and commodity market policies are chosen jointly to maximize a welfare function in which producer welfare is weighted more or less heavily than that of consumers. This line of argument is more relevant for explaining policies than for evaluating eco­ nomic welfare consequences of past or future investments in research. The results from the work on the implications of commodity market distortions for calculating research benefits would carryover fairly directly to the situation where markets are distorted as a consequence of distortions in the exchange rate. However, there is one important difference. Exchange rate distortions result in distortions of prices of all traded goods, and the consequences for distortions of supply and demand in a particular commod­ ity market are potentially quite complicated and difficult to assess (e.g., Krueger, Schiff and Valdes 1988). This topic has been neglected in the literature on research benefits. Given the pervasive nature of exchange rate distortions, especially in less-developed countries, some preliminary analy­ sis is provided in chapter 4. Environmental Sustainability A further type of market distortion involves externalities in production and consumption. This topic is becoming increasingly important in the context of agricultural production and includes all of the "green" issues such as environ­ mental pollution, the greenhouse effect, sustainability, animal welfare, and organic farming. Concern over environmental degradation has increased in several developing countries in recent years. Deforestation, soil erosion and degradation, desertification, silting of rivers and flooding, and pesticide pollu­ tion have become serious problems around the world, and research programs related to natural resource management and conservation are receiving in­ creased emphasis. In a number of countries, environmental deterioration has become so severe that what used to appear to be potential long-run problems have become short-run problems as well. In some countries, higher incomes have meant that the demand for natural resource preservation has increased. Furthermore, research is increasingly exposing the magnitude and conse­ quences of the problem (e.g., Pingali and Roger 1995). Throughout the world, these issues are becoming increasingly topical, and government intervention in response to concerns about them is becoming more pervasive. Procedures for setting research priorities must be capable of considering the effects of alternative research programs on the sustainability of the agricultural resource base. The analyst now faces the problem of accounting both for the externalities and for the government intervention to correct for them. This is far from straightforward conceptually and as an Research Evaluation and Priority-Setting Principles 77 empirical matter it is very difficult. To a great extent the literature has neglected this topic. Lichtenberg, Parker and Zilberman (1988) have pro­ vided a conceptual analysis of the link between commodity programs and the costs of externalities and environmental regulation in U.S. agriculture, but the issue of measuring research benefits in the presence of externalities has been neglected.56 In addition, and perhaps worse for the economic analyst, some research may be directed specifically at reducing such problems. In such cases, it is imperative to pay specific attention to the consequences of research-induced technical changes for the amelioration of externalities. The Full Cost of Government Funds Most of the discussion above has related primarily to the benefits side of the equation. Typically economists assume that the marginal opportunity cost of government spending is the amount spent. Papers by Browning (1976, 1986, 1987), Ballard, Shoven and Whalley (1985), Findlay and Jones (1982) and others have shown that it is appropriate to adjust the amount spent to include the deadweight cost of taxation in order to measure the full social costs of government spending. Fox (1985) introduced this argument into the evaluation of agricultural research investments. More recently, Dalrymple (1990) presented a synopsis of the relevant literature. The social opportunity cost of government revenue has significant im­ plications for the calculation of net research benefits. Typically the results from the empirical studies have suggested that the social cost of government spending is in the range of 1.2 to 1.5 times the amount spent (e.g., Browning 1987). Fullerton (1991) reviewed this literature and his reconciliation ofthe different results leads to a suggestion that a much lower marginal welfare cost of taxation may be appropriate - implying a marginal social cost of government spending of, say, 1.07 to 1.24 times the amount spent for the United States. He shows how the answer depends on what is assumed about the disposition of the tax revenue and the income effects resulting from that disposition. Those estimates are for more-developed countries where the marginal deadweight costs of government spending may be relatively small because of relatively efficient taxation mechanisms. As a further complica­ tion, the deadweight costs of government spending apply to other forms of spending in addition to agricultural research. Alston and Hurd (1990) illus­ trate the issues in the context of commodity programs. Thus the complete analysis of research benefits and costs in a commodity market that involves market distortions must take account of the full opportunity costs of govern- 56. See, also, Capalbo and Antle (1988), Just and Antle (1990), and Beach and Alston (1993). 78 Research Evaluation and Priority-Setting Principles ment revenues, from the point of view of measuring both direct research costs and the effects of research on the costs of market -distorting policies that involve government revenues.57 2.4 Economy-Wide (General-Equilibrium) Implications of Research58 Research-induced technological change in agriculture can have important economy-wide implications for employment and returns to factors of pro­ duction, as well as production and consumption of nonagricultural products. Through output market adjustments, technical changes in agriculture affect the relative prices of agricultural and nonagricultural products not directly affected by the new technology. These indirectly induced changes in product markets can lead to further changes in factor markets. Thus, agricultural productivity changes can affect foreign exchange earnings, food prices, domestic capital generation, labor use in nonagricultural production, rural markets for nonagricultural goods, and relative factor prices. The impacts on nonagricultural production of these research-induced factor market re­ sponses are difficult to predict in general and vary depending upon the nature of the technical change, among other things. Usually, however, one would expect impacts on nonagricultural production to yield additional benefits to augment the direct impacts measured from a partial-equilibrium model of agricultural-sector benefits. 2.4.1 Distinguishing between Partial- and General-Equilibrium Models In practice, and even in theory, the distinction between partial equilibrium and general equilibrium is not always clear. We suggest that it may be thought of in terms of (a) ceteris paribus assumptions and (b) the variables of interest that are endogenous.59 At one extreme is the typical model of a 57. Care is needed, however, when comparing summary statistics, like internal rates of return to research, that take account of this cost with rates of return to alternative investments that do not. 58. Several authors have drawn attention to the general-equilibrium effects of agricultural research, including Schmitz and Seckler (I 970), Ayer and Schuh (1972, 1974), Musalem (1974), and Binswanger (1980). 59. Will Martin (pers. comm.) has suggested to us that there is an important distinction between (a) models in which the budget of the households in the economy is exogenous and (b) models in which household budgets are endogenous and are affected by product market equilibrium and factor payments. The latter, he suggests, is a true general-equilibrium model: the defining characteristic of a general-equilbrium model being the endogenous household budget constraint. Research Evaluation and Priority-Setting Principles 79 commodity market that takes the price and quantity of that commodity as endogenous, treating prices of all other goods as constant and exogenous to the analysis. Usually though, even in this extreme case, it is not assumed that all factor prices are entirely exogenous. At the other extreme are the detailed economy-wide models in which all prices and quantities are endogenous to, and measured in, the analysis so that extreme mutatis mutandis (everything allowed to change) replaces extreme ceteris paribus (everything else held constant). In practice, most "general-equilibrium" models impose some restrictions, so that not all economic variables are endogenous. Most eco­ nomic analyses fall somewhere between these two extremes. Often when analyzing a particular commodity market, it is inappropriate to take the prices of all other goods as being exogenous, but at the same time, it is inappropriate to explicitly measure all of the endogenous adjustments. This is a quasi-general equilibrium analysis in which some prices or quantities are taken as given exogenously. The most important issue here is not what the analysis is called but rather to be clear and consistent about the ceteris paribus assumptions. An alternative way to distinguish between partial- and general­ equilibirum analysis is in terms of the techniques of analysis. For instance, when Marshallian supply-and-demand models are used, the analysis is typically regarded as being a partial-equilbrium analysis, whereas when a social accounting matrix (SAM) is involved, it is regarded as a general-equi­ librium analysis.60 In chapter 4, we consider research benefits in a multimar­ ket setting that considers effects in related factor and product markets as well as in the product market of primary interest. This is an approach to incorpo­ rating general-equilibrium effects in a partial-equilibrium framework. 2.4.2 Practical Approaches for Research Evaluation Unfortunately, economists have not developed general-equilibrium mod­ els both practical and detailed enough to provide much guidance for allocat­ ing research resources. Some models that have been developed to examine linkages among sectors (such as input-output models) are not very helpful because they fail to capture relative price changes and resource adjustments caused by technological change. The reason for the lack of practical general­ equilibrium models is that any relatively complete general-equilibrium mod­ el for the allocation of research resources would require vast quantities of 60. A SAM uses a matrix of current inputs and outputs across activities to extrapolate the impacts of an exogenous change. This approach is general equilibrium in that it considers an entire economy. but it is limited in that it does not allow for any price responses in consumption or production. 80 Research Evaluation and Priority-Setting Principles information, beyond what it would generally be economic to collect.61 Some recent agricultural-sector models provide a potential compromise. For example, Chang et al. (1992) have developed a programming model of the entire U.S. agricultural sector that they have used to simulate the size and distribution of the effects of various research-induced changes in technol­ ogy. This is a general-equilibrium model of the agricultural sector, but only a partial-equilibrium model in the broader sense in that the rest of the economy is exogenous. Martin and Alston (1992, 1994) have shown how to use a balance of trade function approach, in a modern, dual framework, to obtain Hicksian measures of the size and distribution of benefits from research in a full general-equilibrium setting, allowing for any number of market distortions. This approach, they argue, can be applied in a very aggregative way (say, for only a two-good model) as well as for detailed, disaggregated models. An alternative would be to take advantage of an existing general-equilibrium model.62 2.5 Reconciling Multiple Objectives of Research Improved efficiency and equity are the primary objectives in most if not all countries, but reduced income risk and national food security or self-suf­ ficiency are often expressed as important objectives as well, especially in less-developed countries. The concern over income risk is evident in several small countries heavily dependent on one (or a few) export crops. The desire for food self-sufficiency may reflect a feeling that there is a market failure in the world economy, or a perceived military security need. In this section we review these various social objectives and consider whether they can be achieved through public-sector agricultural research or can be better pursued using alternative policies. 61. Computable general equilibrium (CGE) models have been developed, but to keep the analysis manageable, these models often employ highly simplified assumptions with respect to individual sectors (see Robinson 1986 or de Melo and Robinson 1981), and typically, the agricultural sectors in such models are aggregated to a much higher degree than would be desired for an agricultural research priority-setting study. This is a natural consequence when models are developed for other purposes, and it would be uneconomic in most cases to pay the full cost of developing a CGE model of an entire economy only for agricultural research evaluation and priority setting. 62. Such models include the ORANI model of the Australian economy for which the agricultural sector has been developed in relatively detailed form (e.g., see Dixon et aI. 1982 and Higgs 1986), Tyers and Anderson's (1992) model of global agricultural trade, Hertel's (1991) model of the global economy, the World Bank-OECD Rural Urban North South (RUNS) model (see Burniaux and van der Mensbrugghe 1991). To date we have seen very little use of these, or other, CGE models for agricultural research evaluation or priority-setting work. Martin and Alston (1993) have used the RUNS model to simulate a variety of technical changes. Research Evaluation and Priority-Setting Principles 81 2.5.1 Economic Efficiency In chapter 1 we introduced the conventional economic arguments for public-sector involvement in agricultural research. It is widely accepted among economists that there is a market failure in the provision of agri­ cultural research by the private sector and that, without action by the gov­ ernment, there would be an underinvestment in research. Specifically, a convolution of factors (incomplete markets, inappropriability of returns to invention, economies of scale, and perhaps, risk in research) can cause private returns from research investments to be lower than returns to society as a whole. Thus the private sector will be expected to neglect research opportunities that would be profitable from the point of view of the nation as a whole. These ideas are reinforced by the evidence of typically high, estimated, net social returns from public-sector research investments (e.g., Echeverria 1990). This argument gets most of its force from the application of the welfare economics perspectives (producer and consumer surplus and the compensation principle) described in this chapter. With this argument as a basis for the intervention of government, the use of those techniques to guide the intervention seems particularly appropriate. For many economists, the application of the same set of principles and arguments leads to the conclusion that the only defensible justification for government intervention in agricultural research, and the only legitimate and achievable objecti ve of research, is the pursuit of economic efficiency. This is not to say that other objectives (such as personal and functional income distribution and food security) are illegitimate, irrelevant, or unimportant but that there are likely to be alternative, less costly ways to achieve these other objectives than by biasing the agricultural research portfolio away from programs that will maximize total national net income. These arguments are compelling. As a further consideration, maximizing with respect to a single objective is a much simpler problem than trying to maximize a trade-off among a variety of objectives - especially when the terms of that trade-off are unclear and subjective. At the same time we must recognize political realities. Alternative agri­ cultural research programs have different - and occasionally profound - implications for the distribution of income, patterns of trade, regional em­ ployment, food security, and so on. These effects may be of significant concern and, if other policy instruments are not in place to pursue those objectives and correct for any negative impact of research, it may not be appropriate to ignore them in setting research priorities. The best solution, always, is to have policy instruments that aim closely at the objective, and agricultural research is a very blunt instrument for the pursuit of objectives 82 Research Evaluation and Priority-Setting Principles other than economic efficiency (e.g., see Corden 1974, for a discussion of this general question). On the other hand, when the appropriate instruments are not being used, or they are being misused, a blunt instrument may be better than none.63 In such circumstances, it is appropriate and important for the economist assessing research priorities to point out the opportunity cost of efficiency foregone when using agricultural research as an instrument of general economic, institutional, or other policy. 2.5.2 Equity (Income Distribution) and Security Objectives A number of objectives other than efficiency are often raised in the context of agricultural research evaluation and priority setting. These fall into two broad categories: (a) equity or income distribution objectives (i.e., among different producer groups or consumer groups; nutritional status of the poor) and (b) security objectives (e.g., income risk and food self-suffi­ ciency). We have seen that different types of agricultural research programs can have different implications for the functional distribution of income both geographically and between different groups. Agricultural research also generates a wide range of distributional effects related to farm size, income, location, and so on.64 In some cases it is possible to measure the distribution of costs and benefits from agricultural research between producers, consum­ ers (and various categories of producers and consumers), and other partici­ pants in the food production, distribution, and marketing chain. This may be useful as a guide to appropriate ways of financing agricultural research as well as informing choices about alternative programs of research that have different implications for functional income distribution (distribution among factors of production). To go beyond the effects on functional income distribution to personal income distribution requires combining information about factor ownership and consumption with information on functional income distribution effects. If research policy is used to pursue some objective of income distribution, the objective should be made clear so that the contribution of research to it can be measured explicitly. While it is tempting to make gross simplifica­ tions - such as equating the interests of the poor with a low price for staple food crops and presuming that the poor have little interest in export revenues from a capital-intensive cash crop - such gross simplifications are likely to 63. See Krueger (1990) for a discussion of "government failure" that is relevant in this context. 64. David and Otsuka (1994) provide a comprehensive study of the distributional consequences of modem rice varieties in Asia. Research Evaluation and Priority-Setting Principles 83 involve gross errors. In any event, in many cases there are more effective and less costly ways to pursue a cheap-food policy (when that is the aim) than a distorted research policy. Different Producer Groups, Farm Size, and Tenure The impact of technological change on the distribution of income among producer groups can be assessed in many dimensions (corresponding to several of the distributional objectives described in chapter 1). Producers with different incomes, with different farm sizes, in different locations, and with diverse tenure situations can gain or lose depending on the suitability of the new technology to their particular situations. The supply curve can be disaggregated to allow these distributional consequences to be measured within the economic surplus framework (Binswanger 1977; Hayami and Herdt 1977). It is difficult to generalize about the effects of different types of technol­ ogies or the effects of research on particular crops on the incomes of tenants versus landlords. One might expect that biological technologies would augment land and thus help tenants. But the distribution of income gains depends on contractual arrangements for sharing costs and returns, and new technologies can induce changes in those arrangements that may offset the direct effects on tenants. A true picture of producer benefits would need to consider the incidence of taxes paid to support public research. Scobie and Posada (1977, 1978) made such estimates (for both producers and consumers) for rice in Colom­ bia. They divided producers into those producing upland and irrigated rice and then distributed producer benefits across farm sizes, according to esti­ mates of production based on census data. Research costs were also distrib­ uted by farm size. In their study, small upland producers lost the most, but their losses were more than offset by gains among low-income consumers. The issue of whether improved agricultural technologies benefit large farms more than small farms has received a great deal of attention in the literature (e.g., Ruttan 1977; Lipton with Longhurst 1989; Hazell and Ramasamy 1991). The more recent evidence suggests that neither farm size nor tenure has been a major impediment to adoption of new biological technologies, the major focus of agricultural research in less-developed countries (Scobie 1979a,b). Large farms, however, do tend to adopt new technologies first. This probably results from their economies of size in obtaining information about those technologies, their additional experience and education, and a greater ability to absorb risk. Small farms in the same region as the large farms do eventually adopt the technologies, but because 84 Research Evaluation and Priority-Setting Principles the large farms adopt technologies first, they recei ve greater gains than small farms. And, even if all producers in a given region were to adopt a new scale-neutral technology at the same time, absolute income differences would widen (Scobie 1979b, p. 23). This reflects the effects of the unequal distribution of productive assets. The implications for a research manager attempting to allocate research resources is that it will be more difficult to help small farms than large farms in a particular region unless the small farms are growing crops that are different from those grown on large farms or the technologies are biased toward small farms. Many technologies, for reasons beyond the natur~ of the technology per se, are scale-neutral or biased toward large farms, so it may be difficult to generate technologies that have a disproportionately large impact on small farms. This implies that research may not be an effective policy tool for achieving a distributional objective based on farm size. Another implication is that a desire to help small farms through research might be addressed better by focusing research on commodities grown on small farms in regions where small farms predominate instead of focusing on commodities grown on both small and large farms in regions where large farms predominate. Different Consuming Groups Consumers are major beneficiaries of agricultural research. There are direct benefits to all consumers when agricultural research results in a larger quantity being available at a lower price, as occurs when supply shifts out against a downward sloping demand curve. But because consumption pat­ terns and demand response to price changes both vary with income, there will be a differential impact among income classes and potentially an impact on the distribution of income. Most food products are normal goods (i.e., they have positive income elasticities of demand - richer people consume a greater absolute quantity than poorer people) and richer individuals benefit absolutely more from lower food prices' than do poorer individuals. The opposite of this may be true for some staple commodities that are inferior goods (i.e., having negative income elasticities) for which the poor consume absolutely more than the rich. Individually, low-income consumers tend to benefit relatively more from research on staple food commodities than research on other items because low-income consumers spend a high propor­ tion of their budget on food. The reverse is likely to hold for richer consumers who spend a relatively small proportion of their incomes on food. Research Evaluation and Priority-Setting Principles 85 Nutritional Implications ofR esearch Some advocate using agricultural research to improve the nutritional status of the poor. Agricultural research can influence human nutrition through several mechanisms. Perhaps the four most important ones are (a) by affecting household income, (b) by altering prices paid by consumers for food commodities, (c) by influencing downside risk associated with fluctu­ ations in food production, prices, and incomes, and (d) by increasing the production of foods consumed by the households that produce them.65 The effect of research on improving the purchasing power of the poor - both by raising their incomes and by lowering the prices of staple food products - is probably the major source of nutritional gains associated with agricultural research. Only the poor go hungry. Because a relatively high proportion of any income gains made by the poor is spent on food, the income effects of research-induced supply shifts can have major nutritional implications, particularly if those shifts result from technologies aimed at the poorest producers.66 Effects on incomes in agriculture arise in a number of ways. A shift out in the supply curve for agricultural products generates additional income streams to producers. It may also affect the demand for labor, and thereby labor income in agriculture, in either direction, depending in particular on the type of technical change. The primary effects of agricultural research on the non-farm poor are through lower food prices. As supply shifts out against a downward sloping demand curve, consumers benefit from lower food prices. The nutritional effects due to price changes will be influenced by the price elasticity of demand for the commodities, by the inherent nutritional value of the com­ modities, and by the importance of particular commodities in the diets of the poor. The more elastic the demand curve is, the lower the price effect for consumers but the greater the income effect for producers. Research can influence fluctuations in production, prices, and income and thereby alter nutrition. In years when rural incomes are low, because oflower than normal production or prices, severe malnutrition can occur in rural areas. Research can influence commodity diversification as well as the susceptibility of commodities to drought, insects, and diseases, and research 65. Pinstrup-Andersen (1984) lists 10 possible influences of research on human nutrition including these four. See also Pinstrup-Andersen, Ruiz de Londono and Hoover (1976) and Perrin and Scobie (1981). 66. However, the econometric evidence to date indicates that the income and price elasticities of demand for nutrition are small, even in countries with comparatively low per capita incomes (such as China), and correspondingly relatively high price and income elasticities of demand for food. Nutrient content does not always correspond closely to food quality as perceived by consumers. 86 Research Evaluation and Priority-Setting Principles can thereby affect the variability of production, incomes, and nutrition. Improved technologies for subsistence goods can augment the nutrients available for consumption (particularly food energy). While research can do all of these things, it is difficult to draw general conclusions about the nutritional implications of particular research portfo­ lios because the different factors affecting the nutritional impact of research often counteract one another. Because domestic prices are strongly influ­ enced by world market prices for most goods, nutrition is most likely to be improved by research that generates the largest income (efficiency) gains in general, particularly if those gains are realized directly by low-income producers. Therefore, research administrators should resist the temptation to emphasize research on particular commodities just because they have high nutritional content or are important in the diets of the poor. Income Risk Security objectives might also be rooted in a concern about the distribu­ tion of the impact of variablity among different groups of people. Economies heavily dependent on one or a few export crops such as sugarcane, bananas, or coffee often experience extreme income variability due to variability in both production and price. These countries may place a value on programs or policies that help diversify their income sources. While policy tools other than research may be more effective at reducing risk, agricultural research programs can be structured to complement diversification policies or pro­ grams. Research evaluation and priority-setting models may need to recog­ nize this objective in certain countries. This discussion relates to income risk from a national aggregate perspec­ tive. Income risk may also be of concern at a less aggregated level, such as a particular region, right down to the level of individual producers or consumers. For example, research aimed at producing a more drought-toler­ ant crop variety might contribute to reduced year-to-year or season-to-season income variability in regions (or on farms) highly specialized in its produc­ tion. Alternatively, research could develop other crops to be used in those regions (or farms) in order to reduce income variability through diversifica­ tion. This latter strategy might also alleviate the price variability that can be a problem for highly specialized producers. At the level of aggregation of national, regional, and individual producers, research can contribute to a goal of reduced production or income variability. However, research alone cannot contribute much to this objective - e.g., the drought-tolerant variety or alternative crop must be adopted to have any effect. In most cases research contributes little to reduced income variability. Research Evaluation and Priority-Setting Principles 87 There are likely to be much more effective, and less costly, ways to reduce income variability than by distorting the pattern of research investments. Self-Sufficiency and Foreign Exchange An expressed desire for self-sufficiency can be considered a security objective, and it often reflects a concern for national pride or military security. Some recently expressed desires for import substitution may repre­ sent a hope of saving foreign exchange to meet foreign debt obligations. Research administrators ought to consider whether these obligations could better be met by exploiting comparative advantage. Foreign exchange earned from agricultural exports (or saved on imports) can be used to ameliorate foreign debt problems. It is debatable, however, whether a country should place any importance on a separate criterion of generating foreign exchange when setting research priorities. The implicit assumptions are that foreign exchange earnings are worth more than other income gains to the economy, so focusing research on activities that generate the largest efficiency gains will not maximize the country's ability to repay debts. This is a strong assumption. In any event, it would seem to be more appropriate to make adjustments in the shadow prices to be used in a more conventional efficiency analysis than to assign extra weight to foreign exchange earnings per se. Calculation of the "net" foreign exchange effects of research is complicated by the fact that foreign exchange is also saved or spent on additional inputs used in the production of the commodities being researched. These effects are very difficult to calculate.67 Davis and Bantilan (1990) present some arguments and analytical results on these effects in the context of Philippine agriculture. 2.5.3 Trading offM ultiple Objectives It is relatively easy to choose a research portfolio that maximizes eco­ nomic efficiency, once the economic consequences of the alternatives have been assessed. It is much more difficult to set priorities when two or more objectives are involved, since to do so requires (a) a measure or the contri­ bution of each of the alternative investment programs to each of the objec- 67. Finally, it appears that the desire to focus research on commodities that generate foreign exchange through exports results in part from an implicit distributional objective to help large farmers. Farmers reap a higher proportion of the benefits from commodities with high rather than low own-price elasticities of demand. and these tend to be export commodities. These commodities are often grown by large farmers with political influence. Therefore. research administrators and politicians may be reacting to pressures to assist a certain group of farmers when they express a desire to increase foreign exchange earnings through research. 88 Research Evaluation and Priority-Setting Principles tives and (b) information on the relative values to be attached to the alterna­ tive objectives. The first part is difficult, sometimes impossible. The second part is extremely difficult because it involves the subjective value judgments of individuals and decisions about whose judgment is relevant. Both of these aspects are addressed in detail in later chapters. The idea of a trade-off between multiple social objectives has been referred to by some as a social welfare function (SWF). The idea of a SWF has been criticized (e.g., Arrow 1963) but analogous ideas continue to be used widely in economic analysis. An example is the analysis of agricultural price policies using a surplus or benefit transformation curve (BTC), and a political trade-off between the welfare of producers and consumers (e.g., Gardner 1988; Alston and Hurd 1990).68 The same type of model can be used to illustrate the trade-off between economic efficiency - maximizing the total economic benefit from research measured by the change in total eco­ nomic surplus due to research - and other objectives of research, such as equity or security. The simplest case of multiple objectives involves two objectives. To illustrate the ideas, in figure 2.11 the horizontal axis represents economic efficiency, E, and the vertical axis measures equity, V (e.g., E could represent total income or economic surplus in society and V could represent income or 68. Gardner (1988) has suggested extending the same idea to analyzing the joint optimization of agricultural research and price policies (see also de Gorter. Nielson and Rausser 1992 and Alston and Pardey 1993. 1994). It is a short step from the conceptual notion of a BTe and a political trade-off (perhaps called a SWF) to begin measuring the nature of the trade-offs involved in policy choices. In this type of analysis, the concepts of revealed preference are invoked to argue that policy choices indicate policymakers' marginal valuation of the achievement of one objective relative to another (welfare of consumers versus producers; equity versus efficiency). Efficiency in policy choices as described by Gardner (1983) requires that the benefits and costs of alternative programs be equated at the margin (i.e., there is a tangency between the SWF and the BTe). and the slopes of the curves at the equilibrium are sometimes called welfare weights (e.g .• Harberger 1978). Political economy models measure and attempt to explain those welfare weights in terms of the political influence of various interest groups. Similarly. the social willingness to exchange efficiency for equity or some other objective could be measured. Until the 1970s, price policies for agricultural commodities were customarily analyzed as being designed to correct market failures, primarily directed towards efficiency but also in consideration of other objectives in a multiobjective context (i.e .• in much the same way as we have described the approach to research policy in this section). More recently. agricultural economists have to an increasing extent abandoned that approach and have sought political economy explanations of farm programs instead. Similar work has barely begun on the political economy of agricultural research and much of the profession continues to treat agricultural research policy in the way we used to treat agricultural commodity price policy. Political economy models of agricultural research policy could mimic those used to study farm programs: agricultural research policy could be explained in terms of the consequences of self-serving behavior of politically powerful interest groups. Such an approach might have more empirical predictive content than the idea of trading-off efficiency versus equity and, if it did. the virtue of building such trade-offs into priority-setting exercises may be further weakened. Research Evaluation and Priority-Setting Principles 89 Figure 2.11: A trade-off of equity and efficiency using research policy alone Equity (V) a V* ------------------~-------- o E* Efficiency (E) economic surplus of low-income families).69 The curve BTCR represents the range of maximum possible combinations of economic efficiency and equity that can be achieved by varying the mix of research programs in the portfolio. It is drawn as a trade-off, so in order to obtain more of one objective, some of the other must be sacrified, as must be so in the relevant range,1o Each point represents the maximum economic efficiency that can be achieved for a given equity outcome and vice versa. Points below the curve are attainable but clearly inferior to points on the curve; points above the curve are unattainable. Point c represents the result if the research portfolio were chosen simply to maximize economic efficiency atEMAX' Moving back along the curve we can see how much economic surplus must be foregone in order to increase equity by shifting the research portfolio away from the one that maximizes economic efficiency. The other curve on the diagram is an indifference curve, ICo' that represents the policy maker' s willingness to 69. The same diagram could also be used with V representing performance against some other objective, such as security, which could be measured by income variability, for instance. 70. Clearly some technological changes increase the total income and the income going to disadvantaged groups and also reduce the variance of income. Others involve a trade-off: they increase total income at the expense of greater variability (or vice versa) or increase total income at the expense of disadvantaged groups (or vice versa) and so on. However if it is possible to increase benefits in all dimensions from a given R&D expenditure by changing the mix of projects, the portfolio does not lie on the efficient frontier. In such a case there can be substantial gains from inframarginal reallocations of research resources. Once the efficient frontier is reached, by defini­ tion, trade-offs are involved in any change. 90 Research Evaluation and Priority-Setting Principles substitute efficiency for equity. This particular indifference curve is tangent to the BTC and thus represents the highest level of benefits (given those preferences) that can be achieved by varying the combination of economic efficiency and equity through a change in the research portfolio. Thus, the optimal research portfolio is the one that corresponds to point b (E*, V*). To increase equity from VM IN to V* involves an opportunity cost of economic surplus foregone of (EMAx - E*), but given these preferences, this sacrifice is deemed worthwhile. This analysis could readily be extended to a case with three or more objectives as arguments of the policymaker's preference function. Three related comments are in order here. First, deciding whose prefer­ ences are to be used to define the indifference curve is not straightforward. It might depend, for example, on whether the research evaluation or prior­ ity-setting work is being undertaken on behalf of a national government, a provincial government, or a particular research agency. Even at the national level, the preferences expressed by the ministry of finance might differ, for example, from those expressed by the ministry of agriculture. Second, there has been some success in eliciting weights for this type of trade-off among research decision makers, but it is not clear that such decision makers have been fully informed about the costs of making the trade-off through the research policy rather than other mechanisms (Norton, Pardey and Alston 1992). In any event, the work that has been done suggests that research administrators are unwilling to sacrifice very much economic efficiency for other objectives. This implies that the indifference curves are twisted away from point a toward point c. Third, the analysis has usually been conducted as if there were no other policy instruments available. A more complete analysis would allow the use of the best possible policy instrument for substituting economic efficiency for equity. In the extreme example of a lump-sum transfer, for example, there would be no sacrifice of economic efficiency to achieve an increase in equity - the BTC for a lump-sum transfer would be a vertical line through EMAx. The hypothetical lump-sum transfer involves transferring income without any effects on the economic actions of either the people taxed to provide the funds or the people who receive the transfer. True lump-sum transfers are not possible in practice, and to achieve an increase in equity necessarily involves some sacrifice of economic efficiency, which arises because people do respond to being taxed and to receiving transfers. Thus, any policy to improve equity necessarily involves some loss of efficiency; the best policy is the one that involves the smallest sacrifice of economic efficiency in order to achieve the desired equity outcome. The . relevant BTC for policy is the one that involves the use of the best possible Research Evaluation and Priority-Setting Principles 91 (least-cost) policy instruments. Figure 2.12 duplicates the curves in figure 2.11 but includes two additional curves. BTC* is the optimal benefit trans­ formation curve that represents the combinations of economic efficiency and equity that are possible from changing the combinations of the research portfolio and another policy instrument (say tax and income transfer). This BTC is always above the one that holds when only research policy is involved. lC I is the highest indifference curve that can be attained as a tangent to that BTC. In figure 2.12, the optimal outcome (point d) involves higher levels of both equity, V", and efficiency, E", than the optimum from research policy alone (point b) because the alternative approach of combining the research and nonresearch instruments is a more efficient means of pursuing the equity objective than research alone. An extreme outcome - but not an unlikely one - is where the research portfolio is chosen to maximize efficiency without regard for the equity objective, which is pursued most effectively with other policy instruments. We can relate the ideas in figure 2.12 loosely to "Harberger triangles" of efficiency loss associated with market distortions. In the case where research policy is distorted to pursue equity, point b involves a deadweight loss equal to (EMAx - E'). The deadweight loss when using the optimal mix of research and nonresearch instruments is smaller, (EMAx - E"). This comparison is biased because point d involves a higher level of equity as well. The deadweight loss from using the combined policy to achieve V' will be Figure 2.12: A trade-off of equity and efficiency using the least-cost policy combination Equity (V) V** V* ________________________ L __ IL __ •I __ I I I I I o E* E** E' EMAX Efficiency(E) 92 Research Evaluation and Priority-Setting Principles smaller still at (EMAX - E'). The economic surplus analysis alone can be used to identify the opportunity costs involved in using research policy as an instrument of social policy. What it cannot do is indicate the least-cost way of achieving non-efficiency objectives. Such considerations strengthen the case for viewing research policy formulation in a holistic fashion, with regard to the availability of other instruments of social policy. From this view, it may be argued that research policy should focus, perhaps exclu­ sively, on efficiency objectives while other policy instruments are used to pursue equity objectives. 2.6 Conclusions and Discussion A variety of conceptual issues must be considered when designing and implementing a research priority-setting procedure. Some of these relate to the perception of the way agricultural research affects agricultural knowl­ edge, production, and markets; some relate more to the way we translate those effects into measures of the benefits and costs of research; some are related to how we would use that information to evaluate research programs and set priorities. Each of these conceptual issues comes part and parcel with empirical challenges. Economic surplus concepts are intimately involved in any method of estimating research benefits and, in fact, underlie the conventional economic rationale for government intervention in the provision of public-sector agri­ cultural research. These concepts are subject to some criticisms. In im­ plementing them in the context of agricultural research, the more important aspects are the assumptions about the nature of the research-induced supply shift and the measure of its magnitude. There are additional difficulties when markets are distorted by government policies or externalities and when international or interregional spillovers in prices or technology are involved. Methods are available to adjust for all of these factors and to consider the incidence among factors of production in a multistage system, internation­ ally or among rich and poor people. The information and data requirements for the analysis are the binding constraints rather than the methods of economics. While the research program may be designed primarily to increase the size of the national economic pie, inevitably the shape of the pie and the way it is sliced among groups will be affected to some extent by the choice of research priorities. There is a broad consensus among economists that agricultural research is a poor way to achieve national objectives other than economic efficiency. Still, unless other policies are in place that can correct Research Evaluation and Priority-Setting Principles 93 fully for any unintended side effects of agricultural research on other objec­ tives, it may be necessary to trade off the efficiency gains from research against other objectives such as equity or security. The partial nature of measures chosen for the analysis, incorporation of distributional and security concerns, and the nature of the research produc­ tion function each present unique challenges and suggest caution on the part of both the analyst and the people who use the results from the analysis. In the next two chapters we review alternative methods that have been used or suggested for evaluating research and setting research priorities. We discuss the extent to which these procedures can handle the important conceptual issues described in this chapter. Part II Measuring the Effects of Agricultural Research 3 Econometric Measurement of the Effects of Research Econometric and nonparametric approaches have been used to relate measures of output, profit, or costs directly to past investments in research (and extension). Using these methods, the nature and extent of changes in technology resulting from investments in research can be computed along with the measures of research~induced savings in costs or gains in output or profit. Estimates of these effects provide summary indicators of the impact of past research investments. This chapter reviews these methods for assess­ ing supply responses to agricultural R&D, emphasizing the evaluation of the effects of past investments in agricultural research rather than the assessment of the potential impact of current or possible future research investments. However, to the extent that the past is a useful guide to the future, such measures may also be useful as an indicator of the likely payoff to further investments in agricultural research and in designing future research strate­ gies and priorities. Parametric approaches involve specifying an explicit functional form that links inputs to outputs; either primal or dual methods or supply equations can be used. Primal approaches involve estimating either production junctions (in which output is the relevant dependent variable), response junctions (in which output is expressed per unit of a single input, usually land), or productivity junctions (in which output is expressed per unit of aggregate input). In each of these alternative primal representations of the agricultural production technology, research and extension may be included directly as explanatory variables in the statistical model of production. Dual procedures are also feasible and, in some circumstances, preferable. The empirical 97 98 Econometric Measurement of the Effects of Research approaches in these instances call for research and expenditure variables to be included in either a profit function or a cost function and in the associated systems of factor demands and/or output supply equations. Nonparametric procedures can also be used to assess the effects of past agricultural research investments. This type of approach avoids the use of functional forms altogether (hence the term "nonparametric") (Varian 1984). Instead, the data are checked for consistency with axioms of rational producer behavior, such as the weak axiom of cost minimization (WACM) or the weak axiom of profit maximization (W APM), without the imposition of additional restrictions, such as functional forms, as joint hypotheses. In the event that the data are inconsistent with cost minimization (or profit maximization), pro­ gramming algorithms have been developed that can be applied to deduce the minimum set of adjustments to the data (in the form of measures of quantity changes attributable to factor-biased and -neutral technical changes) necessary to restore consistency. This is analogous to the use of a technology index, such as time, to estimate the impact of technical change in the parametric approach. Alternatively, by incorporating measures of expenditures on research (and extension) in the analysis (e.g., Chavas and Cox 1992) the changes in outputs or inputs that are not attributable to changes in input or output prices or scale of production may be used to measure the effects of research (and extension) on output and productivity. This is analogous to incorporating research (and extension) as explanatory variables in a parametric model. Index-number procedures can be used as simple accounting (i.e., aggre­ gating) devices, or they can be used either directly or in conjunction with econometric approaches to assess sources of growth in agricultural output or agricultural productivity. In this way, the share of the growth of output or "total" factor productivity (TFP)' attributable to agricultural research invest­ ments can be identified, distinguished from other sources of growth, and quantified. Index-number approaches represent an additional, and intuitively appealing, means of documenting the effects of agricultural research. Many decisions must be made to make these econometric approaches operational. For example, when a parametric approach is chosen, a decision must be made about which particular primal or dual method to use. Decisions must also be made about functional form, the degree of spatial, commodity, and temporal disaggregation, the variables to be included in the model, and how to specify a stock-of-knowledge (or research-and-extension) variable in the model. All of these choices are governed by the nature of the question at I. The teon total is used here in deference to its common usage in the literature. Measured 1FPs are more appropriately described as multifactor productivity indexes in recognition of the fact that measured inputs do not capture the totality of all factors of production. For more discussion on this point, see Schultz (1956) and Alston, Anderson and Pardey (1994). Econometric Measurement of the Effects of Research 99 hand and the availability of data and resources for the analysis. These modeling decisions can have a substantial impact on the insights to be gleaned from the data concerning technical change and the contribution of R&D to such change. This chapter begins with a review of the relevant theory and practice in production economics as it relates to assessing research-induced technical changes. Problems with data measurement, model specification, and statis­ tical estimation are considered. Then we discuss how to use the results from applying these econometric methods to quantify various aspects of the economic effects of research. We describe how growth-accounting tech­ niques can be used to identify the sources of output growth, the contributions of research and extension, in particular. Then we review the procedures for translating the parameters obtained from production, productivity, and cost functions into measures of the economic benefits of research. 3.1 Conceptual Models of Production, Productivity, and Technical Change Evaluating the effects of agricultural research and extension can be viewed as a particular application of the more generally applicable methods of production economics. But some special problems arise in applying the methods of production economics to evaluating agricultural R&D - notably the long lags in the relationship between an investment in R&D and the effects of that investment on production. And the evaluation goes beyond estimating the relationships between inputs and outputs. When evaluating past research investments, we are usually more interested in the relationship between production (or productivity) and investments in research and extension than in the relationship between conventional inputs and outputs. However, in order to isolate the effects of R&D, it is usually necessary to measure the effects of R&D and the effects of other variables at the same time in a complete model of production. If R&D effects are important, models of production that do not account for these effects will be misspecified and the resulting estimates are likely to be biased. If R&D affects output directly or if it affects the relationship between conventional inputs and output, vari­ ables representing R&D belong in the model for econometric reasons, regard­ less of whether the primary purpose of the analysis is to estimate the effects of R&D or to estimate, say, the output response to fertilizer. The methods for evaluating agricultural R&D have developed along with more general developments in production economics. Until the early 1970s, production economics used almost entirely primal approaches in which the 100 Econometric Measurement of the Effects of Research quantity of output was modeled as a function of input quantities. Some of these models included time trends for technological change; some adjusted inputs for quality change and incorporated other variables that might be thought of as representing sources of technological changes, distinguishing between conventional and unconventional inputs; and some used explicit measures of research and extension as inputs.2 Simultaneity between inputs and outputs is a general problem with these primal approaches. Typically, the models have used annual data. But usually some input decisions (e.g., pest-control inputs or harvesting inputs) are made during the year after some information has become available on weather and other factors that are usually treated as exogenous, random, and part of the residual or error term of the model. This means that the error term and some included inputs move together, so that some inputs are not independent of the random part of the model. As Marshak and Andrews (1944) first pointed out, routine regression procedures are inappropriate in such circumstances. A second statistical problem with primal approaches has been multicollinearity. When all of the explanatory variables tend to move together - a common feature of highly trending time-series data - it is difficult to isolate statistically the effects of any particular variables (e.g., R&D variables) on output, indepen­ dent of changes in other variables. Various devices have been developed to handle these problems, but they may introduce problems of their own. The 1970s saw the beginning of a "new wave" in production economics, driven by two related innovations in the technology of economics: flexible functional forms and duality models of production.3 This new wave was sustained in part by the continuing process of innovation in data processing technology and attendant developments in econometric methods. The new methods allowed researchers to tackle the estimation problems posed by the new models and to take advantage of expanded computing capacity. Dual specifications, most often based on locally flexible functional forms (such as the translog), have come to dominate the theoretical and empirical literature in production economics over the past twenty years and, perhaps to a lesser extent, the literature on estimating returns to research ex post. An offshoot from the literature on flexible functional forms has been the development of globally flexible models, such as the Fourier flexible form (Gallant 1982; Chalfant 1983), which are sometimes termed "semi-nonparametric" because the link 2. See, for example, Griliches (1964), Evenson (1967, 1968), Bredahland Peterson (1976), and DaVis (1979). 3. Theoretical papers in these areas began with, for example, Diewert (1971, 1973, 1974 ),Christensen, Jorgenson and Lau (1971, 1973), and Lau (1976). Applications to agriculture began aImost inunediately afterwards, including Binswanger (l974a and b, 1975), and Lau and Yotopoulos (1971). Chambers (1988) documents the literature. Econometric Measurement of the Effects of Research 101 between any single parameter and an economic concept of interest (e.g., an elasticity of substitution) is broken.4 More recently. nonparametric models of production have been developed and applied to measure the impact of research and extension on agricultural production (e.g., Chavas and Cox 1992). These methods are relatively new. and they have considerable appeal because they avoid imposing restrictions that are not derived from economic theory as joint hypotheses when the properties of the data are evaluated (and when hypotheses about the data-generating process are tested). Much ofthe literature on agricultural production economics has acknowl­ edged the importance of incorporating technological change in the specifica­ tion. Indeed. how to incorporate technological change and how to distinguish factor-biased technical changes from price-induced substitution effects or factor-neutral technical change from economies of scale have been the focus of a significant fraction of the literature.s Most studies have used a time index as a proxy for technological change. Only a small fraction have explicitly incorporated measures of R&D: typically, these have been studies where the focus has been on estimating the impact of research.6 In principle, there may be only one appropriate specification of a particu­ lar production relationship, reflecting the true data-generating process, and this specification should govern the econometric estimation regardless of the purpose of the analysis. In practice, the true data-generating process cannot be known, and empirical specifications are usually dictated by the purpose of the analysis or the available data. The choices of procedure to use, variables to include (including the representation of technical change), and functional form are intimately related and are typically made jointly in light of the objecti ve of the analysis and in consideration of data constraints. When the primary aim is to measure a factor-substitution relationship, it may be appropriate to spend relatively more degrees of freedom attempting to increase precision in that aspect of the relationship, sacrificing accuracy 4. The notions of local and global flexibility. as they pertain to the choice of functional fonn, are discussed further below. s. Once a functional form has been chosen (whether for a production function. cost function, profit function, or supply function), a technology index - typically a function of research and extension expenditures or time - could be incorporated either (a) as an argument of tre function (an explanatory variable), (b) as a modifier of the variables already in the model of the function (e.g., defining actual and effective prices and quantities of inputs and outputs to represent input- or output-augmenting technical change), or (c) as a modifier of the parameters of the function. 6. Most studies have used time-series data, in which case a time-trend variable can be used as a proxy for changes in technology, but in studies using cross-sectional data, there is no natural ordering of the observations in terms of a "path of technology evolution," so some other index must be used. When panel data (i.e., time series of cross-sections) are used, it may be appropriate to index technology both over time (perhaps using a time trend) and cross-sectionally. 102 Econometric Measurement o/the Effects 0/ Research (precision and, perhaps, unbiasedness) elsewhere in the model. For instance, a functional form may be chosen that is relatively flexible in its representa­ tion of substitution responses but parsimonious in its representation of technical change. Alternatively, when the relationship between R&D and production is the focus, it may be efficient to choose a specification that sacrifices some accuracy in estimates of scale and substitution effects in order to concentrate on the response to R&D. For instance, a relatively parsimonious (and restrictive) production function may be chosen (e.g., a Cobb-Douglas) but with flexibility added by allowing its parameters to be functions of research (e.g., Fulginiti and Perrin 1992). We now briefly review the conceptual underpinnings of these various approaches to problems in production economics and to the ex post evaluation of returns to R&D. Later sections address (a) the econometric issues and problems that arise with the various approaches and methods to mitigate their effects and (b) the more common measurement problems that arise, along with practicable methods to cope with them. Some of these problems are of the type that accompany any empirical econometric analysis. The discussion empha­ sizes the aspects that are peculiar to measuring research benefits, referring to the more general literature for information on the more general problems. 3.1.1 Parametric Approaches Most econometric studies of returns to research have used either paramet­ ric models or index-number approaches to estimate the productivity growth attributable to R&D. In this section we consider the conceptual underpinnings and the strengths and weaknesses of the various parametric approaches. Since we are interested primarily in the effects of research on supply functions, the literature on the analysis of supply response is examined. In recent years, several review articles on the state of the art of empirical supply analysis have been written: Colman (1983), Wall and Fisher (1988), and Just (1991, 1993). However, these review articles paid relatively little attention to the incorporation of research and extension variables, so we focus on the treatment of research and extension within the frameworks suggested by the previous literature and leave out much of the more general detail because it can be found in those studies. Colman (1983) classified econometric approaches to estimating supply functions into three broad categories in order to "facilitate examination of the role of economic theory in the different approaches, and to help assess their empirical properties and problems" (p. 202). We consider three corresponding categories of approaches to estimate supply response to, and benefits from, research and extension: (a) primal (two-stage) models, (b) dual (two-stage) Econometric Measurement of the Effects ofR esearch 103 models, and (c) direct single-equation supply models. As can be seen in figure 3.1 (adapted from Colman 1983, p. 205), these three approaches are intimately connected in terms of the (static) theory of the firm. The first estimates a production function and then imposes upon that function some behavioral assumptions in order to deduce the implied supply response (route 1 in figure 3.1). The second estimates a cost function and its corresponding input-demand functions, in which behavioral assumptions are embedded, and then uses derivative properties to deduce supply response (route 2 in figure 3.1), or it estimates a profit function jointly with its input-demand and corresponding output-supply functions (route 3 in figure 3.1). The third estimates the supply response directly so that behavioral assumptions are minimized, but as a result, there is no assurance that the estimates are consistent with any particular set of behavioral assumptions. Figure 3.1: Economic relationships between supply functions and other functions in the theory of the competitive firm Produclionfunction Min.csubjecttof Costfunction ......... Q"f(X,K,Z) C'''c*(Q, W,K,Z) ROUTE I R07 (CostMin.)lnputdemandsX*' "x·'(Q,W.K,z) ,, , (Profit Max.) InputdemandsX* "x·(P,W.K.zJ .... -! , (ProfitMax.)OutputsuppliesQ·"q·(P,W.K,z) <0(-' ROUTE 3 Projilfunclion Transformationfunction ......... It'" n*(P, W,K,Z) Max. It subject to F F(Q.X,K.Z) " 0 The key contrast is between the third category (i.e., directly estimated supply functions) and the other approaches. As will be seen below, both of the first two categories (using the production, profit, or cost function) draw comparatively heavily on the theory of the firm in order to impose restric­ tions on parameters and (presuming the restrictions to be correct) to improve the efficiency and internal consistency of the estimates. But they also involve comparatively strong restrictions on behavior, such as assumptions of static cost minimization or profit maximization and perfect knowledge. Thus, such models allow relatively little flexibility in behavioral assumptions and a 104 Econometric Measurement of the Effects of Research rudimentary role for dynamic responses, uncertainty, and expectations. For these reasons, the ability of these models to describe and predict has been called into question, and their performance has often been disappointing in that regard. Models in the third category - the so-called ad hoc single-equa­ tion supply-response models that in fact predominate in the econometric supply-response literature - use relatively little of the static theory of the competitive firm under perfect knowledge and, instead, draw more heavily on the theory that has been developed for modeling dynamic response, expectation formation, and decision making under uncertainty. In comparing these approaches, Colman (1983, p. 224) concludes that It is clear ... that there exist major problems with the time-series regression approach to single-commodity supply response analysis. However, this is also true of its major competitors, and it remains the case that aggregate time-series analysis is the most often used and preferred of the methods. The most significant factors in its favor are that it operates directly on the aggregate supply data which are the object of interest for projection purposes, and it handles dynamic adjustments to supply in ways in which the other procedures do not. It is the simplest of the procedures in terms of estimational methods and data requirements. . . . Finally, it is a technique which has shown itself capable of generating acceptable and useful results. Studies of returns to research, on the other hand, have been dominated by what Colman (1983) referred to as two-stage procedures, where either a primal (production function) model or its dual equivalent (a cost function or profit function) is estimated in the first stage, sometimes jointly with the implied system of input-demand and, perhaps, output-supply equations, and the implications for research benefits are deduced from those estimates in a further step. A comparatively small number of studies have used the third approach (the "directly estimated single-equation" approach) in cases where the purpose of the study was to evaluate research on a particular commodity (e.g., Otto 1981; Zentner 1985; Haque, Fox and Brinkman 1989). Primal Approaches The primal approach to ex post evaluation of agricultural research in­ volves specifying a production function f(.), in which agricultural output in time t (i.e., Q,) depends on the quantities of conventional inputs, X" various "quasi-fixed" factors such as public investment in infrastructure (such as roads, communications, irrigation, and education), Z" the flow of services from the stock of knowledge, K, (which we can represent by a technology index, 't,), and uncontrolled factors such as weather and pests, V,: Q, =f (X, , Z, ' 't, , V, ) (3.1) Econometric Measurement of the Effects of Research 105 Research investments can lead to a change in productivity (output per unit of conventional inputs, QIX) by changing the quality or price of conventional inputs and outputs (i.e., through a change in the technology used to produce those inputs and outputs) or by increasing the stock of knowledge or the use of the existing stock of knowledge. Thus, the state of technology, 'tl' is endoge­ nous, in part because the extent of utilization of available knowledge depends upon relative factor prices, W" the stock of farmers' human capital, HI' and the extent and quality of extension services, E,. The same infrastructural variables, Z" that directly affect output may also indirectly influence Q, through their effects on 't,. Initially we model this relationship as follows: 'tt =' t ( Kt , Wt , Ht , Et , Zt) (3.2) The current stock of useful knowledge, KI' depreciates by an amountD, as it is replaced by better information or when circumstances change to make it less useful, and it increases by an amount I, because of the incorporation of results from past investments in research. Thus, Kt = Kt- l + It-Dt (3.3) With this specification, it can be seen (by repeated substitution for K,.1) that the stock of current knowledge is defined by the entire history of increments and depreciation of knowledge - an infinite lag structure. The dynamics of the relationship between past investments in research, Rt-r> and extension, EH , and current increments to useful knowledge, I" are complicated and uncertain. A general form of the relationship is It = i (Rt, ... , Rt-LR , Et , ... , Et-LE , Kt- b Zt) (3.4) where Z, represents a set of quasi-fixed factors that can affect the performance of a research system, such as the orientation (e.g., commodity, technology, or problem focus) of the research and the institutional and management environ­ ment within which these research resources are deployed. The relationship between research investments and changes in the stock of useful knowledge is sometimes termed a research production function or a knowledge production function. It is a central component in relating agricul­ tural output to research (and extension) inputs.7 Usually, the stock of knowl- 7. This model includes general research invesbnent variables, Rio and extension variables, E indexed for timing, without identifying the nature of the work undertaken (i.e., pretechnology research", applied research, development), or who is undertaking the invesbnent (the public sector, the private sector, or foreigners). In addition, the generation of knowledge involves several interdependent stages of production in which outputs (research results) from one stage are used as inputs into the next. Conceptua1ly, however, the research variable in the model above can include expenditures on any type of research as part of a program or project that leads to new agricultural technology. 106 Econometric Measurement of the Effects of Research edge cannot be observed directly, so the research (knowledge) production function is more a part of the conceptual apparatus than an empirical tool. The empirically useful variant of the research (knowledge) production function is the function that relates output (or productivity) to lagged values of research investments.8 By loosely combining equations 3.1 through 3.4, a relationship between investments in research and output (or productivity) can be defined as Qt = f (Xt , Ut , 'tt [(R t ' ... , Rt- LR ) , (Et, ... , Et- LE ) , Kt-l , Wt , Ht , Zt]) (3.5) = f(Xt, Rt~r , Et-e , Wt , Ht , Zt ,Vt) for r, e =0 to 00 in which current output depends on current flows of conventional inputs, X" indefinitely long lags of past investments in agricultural research, R,_r, and extension, Et-., current values of factor prices, W" and the stock of human capital, H/ The Z vector is commonly taken to include publicly provided "infrastructural" variables that affect the relationship between measured inputs and outputs, like public investments in such things as transportation, communication, education, and health care.1O In equation 3.5, the vector Z subsumes the Z variables from 3.4 that more directly affect the performance of a research and extension system; it may also include variables that reflect otherwise unmeasured changes in the quality of conventional inputs. 11 None of the variables included in Z are fixed in a literal sense; they certainly vary over a sample. But they are properly treated as fixed if they are beyond the direct control of the farmers whose production decisions are being modeled. So, they are the variables that are most unequivocally exogenous:2 Notice that in the transition to the second line of equation 3.5, the lagged value of the knowledge stock Kt-I has been dropped and, as a consequence, 8. Some studies have attempted to estimate the research production function itself (e.g .• Pakes and Griliches 1980; Pardey 1989). 9. Conventional inputs are always included in production functions. Nowadays. human capital variables might be regarded as being conventional too. but here they relate more particularly to the effects of human capital on utilization of knowledge (rather than. perhaps. aIlocative efficiency or management). Prices are not commonly included in production functions. but there is recent precedent for doing so in relation to the induced-innovation hypothesis (e.g .• see Fulginiti and Perrin 1992). The other variables are measures of past research expenditures and current and lagged extension investments such as those that have been incorporated in production functions in several studies (e.g .• Huffman and Evenson 1992). 10. See. for example. Antle (1983). Binswangeret aI. (1987). Lau and Yotopoulos (1989). and Craig, Pardey and Roseboom (1994). 11. Where changes in input quality are fully reflected in measured inputs. it would obviously be double counting to make a further adjustment to account for input quality. 12. In other words. while measured output is a function of Z. the residual part of output unexplained by the X variable is uncorrelated with Z. Econometric Measurement of the Effects of Research 107 the infinite lags of research and extension have replaced their finite counter­ parts. The role of knowledge depreciation may be implicitly reflected in the research lag structure, or some explicit treatment for depreciation may be introduced. Previous studies have not been clear on these points and some errors may have resulted. Many studies have estimated the equivalent of the first two lines of equation 3.5 (i.e., finite lags of research and extension) but excluded the stock of knowledge, Kt_ 1 (e.g., Huffman and Evenson 1989,1992). This amounts to a problem of an omitted variable - an omission of the knowl­ edge-stock variable that can be interpreted, equivalently, as having truncated the research lag. It is appropriate to omit the lagged knowledge stock only if the knowledge stock depreciates completely each year, so that only current increments to knowledge matter. In contrast, Fulginiti and Perrin (1992) effectively included an infinite lag with zero depreciation (i.e., in their model, the useful knowledge stock grew each year, in both gross and net terms, according to a trapezoidal lag of past research investments). In short, Huffman and Evenson (1989, 1992) effectively assumed lOO% depreciation; Fulginiti and Perrin (1992) effectively assumed 0% depreciation. A more general model would allow some research-induced changes in knowledge to have effects that persist indefinitely, while some depreciate away relatively quickly. In practice, in an aggregative model, a relatively flexible specifica­ tion of lags and dynamic relationships may be required. The main im­ plications of omitting the knowledge-stock variable in a reduced-form model are (a) the use of a finite lag is inappropriate, a specification error that will lead to biased coefficients, and (b) the interpretation of the reduced-form coefficients on lagged research and extension variables is unclear. A yield equation variant of this model expresses output per unit of an input (say, land), but the commonly used specifications involve some ~sumptions that may not be warranted. 13 Implementation of this conceptual model requires decisions about which of the variables to include (a choice that is dictated in large part by the availability of data or suitable proxies), which functional form to use, and how to specify the.random part of the model that will be treated as an unexplained residual. In order to measure returns to R&D, it is necessary to have, at a minimum, some data on research investments and conventional inputs. These minimal 13. For example, cross-section production-function estimates in the literature are usually based on an aggregate production function such as 3.5, while the empirical model involves some scaling of the output and input variables (e.g., by hectare for a yield-response function or by number of frums). Craig, Pardey and Roseboom (1994) point out that failure to include the scaling variable then as an explanatory variable amounts to an assumption either that there are constant returns to scale among all scaled inputs or that the "aggregate" production function is appropriately defined in the scaled units. See Dillon and Anderson (1990) for an extensive treatment of the analysis of agricuhural response functions. 108 Econometric Measurement oft he Effects of Research requirements for an econometric evaluation study may be augmented with explicit measures of extension investments, human capital variables, and weather variables. But often such variables are not included in the model and their effects are left as part of the unexplained residual. The implications of this (in particular, the implications of the problem of omitted variables for bias and inconsistency in the parameter estimates) are discussed below in the section on implementation. Decisions about functional form are intimately related to decisions on how to measure and treat the explanatory variables - especially the research and extension variables - and related decisions on estimation procedures.14 Most of the econometric models of returns to research have used a Cobb-Douglas functional form to estimate an explicit form of an equation such as 3.5. The Cobb-Douglas model is extremely restrictive; elasticities offactor substitution are restricted to unity, output elasticities are constant and equal to factor shares under constant returns to scale, and the price elasticities of the implied com­ pensated deri ved demands are invariant to the amount of output. But it has the advantage of being parsimonious with respect to parameters (one per input) and relatively easy to implement econometrically (although it may be difficult to estimate for statistical reasons). In most cases parsimony is judged to be important because a large number of lagged values for research (and extension) expenditures are included and a relatively large number of parameters are used to estimate their effects (Le., so as not to impose too much structure on the form of the lags). Often, however, even in such otherwise relatively inflexible specifications, a great deal of structure is also imposed on the dynamics associated with research in order to conserve degrees of freedom and minimize problems of multicollinearity. The constant elasticity of substitution (CES) production function is somewhat less restrictive than the Cobb-Douglas in that the elasticities of factor substitution may differ from one, but they are still restricted to being positive numbers and constant across the data. Over the past twenty years, a number of locally flexible functional forms have been developed that allow elasticities of substitution to vary over the 14. Some studies have used a two-step estimation procedw-e to estimate the effects of research on production or cast (e.g., Mullen, Cox and Foster 1992).ln the first step, a production-function relationship is estimated in tenns of conventional inputs and some other factors, including time-trend variables but excluding the detenninants of the technology index. Thus, the value of the technology index is, itself, estimated as a parameter given by the changes in productivity attributable to the time-trend variables. Second, the measured technology index (or productivity) is regressed against lagged research variables. This type of two-step estimation procedw-e may be used to handle some of the multicollinearity that may occur when estimating a model with a Imge number of variables characterized by strong trends. But the factors included in the first stage could well be positively correlated with the omitted research variable. The estimated coefficients for these factors will embody part of the effects of research and, as a consequence, be biased upward. We do not recommend this approach for that reason. Econometric Measurement of the Effects of Research 109 sample. These locally flexible models are typically quadratic forms in some transformation of inputs and outputs. For example, the quadratic production function expresses output as a quadratic function of the quantities of inputs. The translog production function has the logarithm of output as a quadratic function of the logarithms of inputs. The generalized Leontief uses square roots of variables. These models require many mor~ parameters than their inflexible counterparts, especially when the number of input categories is large, and this may account for their relative lack of popularity in research evaluation stud­ ies. ls The number of parameters to be estimated may be reduced by imposing restrictions derived from theory. If these restrictions are to be tested in the analysis, as they often are, to "test down" to a more parsimonious specification, the gain in degrees of freedom is forsaken - although the illusion may remain. Globally flexible models, such as Gallant's (1982) use of a Fourier flexible form in a cost function based on a translog model, do not necessarily use many more parameters. However, unless large data sets become available, many studies will continue to use a Cobb-Douglas type of production function, with a relatively small number of input categories to both conserve degrees of freedom and avoid multicollinearity. As a result, the studies will always be vulnerable to the possibility that the estimates of R&D effects are biased because of the use of an inappropriate model specification. This issue will be addressed further in the section on empirical implementation.16 Once an estimate of the impact of research on production has been obtained, the remaining step is to translate that estimate into a measure of research benefits. This step involves, whether implicitly or explicitly, an economic surplus approach. It works as follows. The estimated production function can be used to partition the total output into one part that is explained by conven­ tional inputs (and other nonresearch factors), another part that is explained by research expenditures, and an unexplained residual. One way to proceed is to calculate the value of the additional output due to research for each year in the time series being studied, by multiplying the estimate of the quantity that is attributable to research by the corresponding output price (in the case of aggregated output, a corresponding aggregate price would be used). This procedure is analogous to an economic surplus calculation in which supply is IS. fur instance. with n input categories and assuming constant returns to scale. the Cobb-Douglas requires n-l parameters. while a translog would require n(n-l)t2 parameters; in the case oftive inputs. the fonner requires four parameters and the latter requires 10. It is conunon practice to preaggregate more-detailed input categories into broader classes of inputs to save on degrees of freedom. with the consequence that a number of restrictions are implicitly imposed through this aggregation procedure. 16. In general. whether in the context of primal or dual models. the flexible functional foODS have been used much more often in studies undertaken for reasons other than research evaluation. where most often technological change has been incorporated by the expedient of including time as a technology index (usually either as a linear or quadratic trend). 110 Econometric Measurement of the Effects of Research assumed to be perfectly inelastic and demand is assumed to be perfectly elastic, so that the effect of a research-induced increase in quantity is to shift a vertical supply function to the righ~, in parallel, against a horizontal demand function (see figure 2.6a and the attendant discussion).17 Once a stream of annual benefits has been estimated in this fashion, all that remains is to compare that stream to the corresponding stream of research expenditures, using conven­ tional capital budgeting methods to compute a net present value, cost-benefit ratio, or internal rate of return. A second approach is to compute the value of inputs that would have been required to produce the actual outputs that were produced if there had not been any research expenditure. This computation involves taking the quan­ tity of outputs attributable to research, deducing the quantities of additional inputs that would have been needed to produce those additional outputs in the absence of the research (e.g., under constant returns to scale, all inputs would need to be increased by a proportion equal to the computed, propor­ tional, research-induced increase in outputs), and then evaluating the stream of cost savings using the quantities of inputs saved in each year, multiplied by their corresponding input prices. This approach is analogous to a surplus analysis that involves shifting a perfectly elastic supply function up, in parallel, against a totally inelastic demand function, as a measure of the benefits foregone if the research had not been undertaken (see figure 2.6b and the attendant discussion). Again, translating the stream of benefits into a summary statistic is reasonably straightforward. Whether the research-induced technological change is regarded as output enhancing or (input) cost saving, the primal approach is attractive in that the model directly links the research variable and the production technology. The production function itself is essentially a physical relationship, not a behavioral one (at least in principle), and important explicit behavioral assumptions are not imposed.18 The device of multiplying the additional outputs by the output price, or the inputs saved by the input prices, does not involve any behavioral restrictions either. 17. The functional fonn of the production function and the way in which technical change enters it may imply a particular parallel or nonparallel supply shift due to changing technology (Davis 198\). However, the econometric estimation will typically not pennit strong inferences to be drawn about the nature of the technical change or the nature of the induced supply shift (except, perhaps, locally, say, at the mean of the sample data). Thus, there is usually little basis for holding strong views about the nature of the technical change, nor for insisting on consistency between the fonn of technical change in the econometric model (as a local approximation) and that assumed in the welfare evaluation (as applying globally). 18. Of course observed production-function relations do incorporate some very complex behavioral assumptions concerning the choice of technology (i.e., technology adoption and use), but there are no consequences for the aspects being considered here. There are also behavioral assumptions embedded in the preaggregation of inputs and outputs that are required to estimate any production function. Econometric Measurement of the Effects of Research 111 Dual approaches embed strong behavioral assumptions as restrictions in the analysis. These behavioral assumptions save degrees of freedom by allowing the imposition of parametric restrictions in estimation. The coun­ terpoint to this advantage is that, if the strong behavioral restrictions are inappropriate, we have merely exchanged one potential source of bias for another. The primal approach generally imposes restrictive assumptions about the technology but does not impose any especially strong restrictions on behavior; the dual approach allows a relatively flexible specification of the technology, but the price of this flexibility is the imposition of behavioral assumptions that are of questionable applicability in an agricultural setting characterized by uncertainty and dynamics. Dual Approaches The dual approach involves estimating a cost function (or a profit func­ tion) instead of a production function. As described above, at least for a competitive firm, for every primal representation of production, there is a corresponding dual representation. 19 The dual approach has several potential advantages. First, the use of factor prices, rather than their quantities, as explanatory variables may avoid the problems of simultaneity that arise when input choices are jointly endogenous with output; factor (and product) prices are more likely to be behaviorally, and hence statistically, exogenous to a firm and even to an industry.2o Second, the dual representations, combined with their derivative proper­ ties, permit the estimation of a system of equations comprising the cost function and the system of output-constrained factor-demand functions. Or in the case of the profit function, the dual approach leads to the estimation of Marshallian factor-demand equations and output-supply functions. In a dual analogue to the model in equation 3.1 above, we can write a cost function as Ct =c ( Qt , W t , Zt , 'tt, Ut ) (3.6) where c(.) is the cost function in which C1 is the minimum cost of producing 19. Young et al. (1987) provide an excellent pedagogical treatise on duality theo!)' and applied production economics. More advanced and more comprehensive treatments of the topics can be found in OIambers (1988) and Comes (1992). 20. However, most profit-and-rost function models use aggregate, industl)'-Ievel data - with specialized factors of production for nontraded goods. And for a few traded commodities for which the count!)' can influence world prices (e.g., com and soybeans in the United States, jute in Bangladesh, wool in Australia), measured prices may not be statistically exogenous. In any case, one can always test for . exogeneity (e.g., a Hausman test), and if variables on the right-hand side are not statistically exogenous; alternative estimation procedures (e.g., 3SLS, instrumental variables) can be employed. 112 Econometric Measurement of the Effects of Research output Q, given values of a vector of prices of conventional variable inputs, WI' various quasi-fixed factors (such as publicly provided transportation, irrigation, and education facilities and services), Z" the state oftechnology, t" and uncontrolled factors that are assumed to be uncorrelated with the other variables, U,.21 If equation 3.6 is considered a multi-output cost function, then the argument Q, would be replaced with the vector Q,. As in the case of the primal model, a reduced-form expression for the cost function can be obtained by augmenting the model with other variables, including research and extension variables, as follows. Loosely combining equations 3.6 and 3.2 through 3.4, we can suggest a reduced-form relationship between investments in research and output (or productivity) in which current cost depends upon current prices of conventional inputs, WI' indefmitely long lags of past investments in agricultural research, R,." and extension, E, .• , the stock of human capital, HI' other factors such as fixed inputs, infrastructure, and (when they are not properly accounted for by adjustments in input prices) changes in input quality, Z" and uncontrolled factors, U" so that Ct = c (Qt, W t , Ut , tt [(Rt , ... , Rt- LR ) , (Et, ... , Et-LE ), Ht , ZtD (3.7) = c (Qt ' W t , Rt- r , Et-e ' Ht , Zt ,Ut) for r, e = 0 to 00 Here, all explanatory variables other than W, are quantity variables. A unit­ or average-cost variant of this model expresses cost per unit of output. In order to implement this type of model, a functional form must be chosen. As with the primal approach, both relatively inflexible and parametrically parsimonious forms, such as the Cobb-Douglas, and relatively flexible alterna­ tives, such as the translog, are available. The main issues are comparable to those raised above in relation to the primal model. The trade-off is between flexibility and parsimony, given the constraints of data and the objectives of the analysis. One important consideration is that it is desirable to incorporate the variables that are additional to those normally included in a cost function (i.e., output and the prices of conventional inputs) in ways that yield an augmented modee2 that continues to satisfy the requirements for a well-be- 21. If this model is interpreted as a long-run cost function, then all inputs are, by definition, variable. However, in the shorter run, some factors may be fixed. 1bese can be either included explicitly at the stage of equation 3.6 or left implicit, as, in fact, they are in equation 3.6. Another issue that arises here, that does not arise in the primal approach, is the role of expectations and dynamics. In equation 3.6. the factor prices are treated as if they are known with certainty at the time decisions are made and take effect, whereas decisions about agricultural inputs are often made in an environment of uncertainty about technology, random uncontrollable factors (such as weather and pests), and prices. 22. A cost function must be continuous with respect to input prices, linearly homogeneous in input prices, nondecreasing in input prices, and concave in input prices. See Young et aI. (1987) or Varian (1978). Econometric Measurement of the Effects of Research 113 haved cost function. Obtaining this augmented model is not always easy, and the choice of functional form for the cost function may have strong im­ plications for the nature of the measured impact of research on output-supply and input-demand functions. The derivation of the corresponding output-supply and input-demand equations is straightforward (see section 3.3.2). Output-supply equations are obtained by setting the deri vative of the cost function with respect to output (i.e., marginal cost) equal to price and then inverting to solve for output. The input-demand equations are obtained by the application of Shephard's lem­ ma: Xi.) = ac(.)/aW;". 'iJ'hese output-supply and input-demand equations contain the same para~ters as the cost function itself, and this fact may be imposed as a restriction in the estimation of a joint system comprising the cost function and its derivatives with respect to factor prices (or quantities). The benefit from doing this joint estimation is that, so long as the model is correctly specified, the imposition of these behavioral assumptions, with respect to factor demands and through cross-equation restrictions on param­ eters, means that the parameters are estimated with greater precision than if only the cost function were estimated. Once the model has been estimated, total costs of production for each year may be partitioned into those attributable to conventional inputs (and other nonresearch factors) and those attributable to research (a negative number if research has successfully led to lower costs). Then, the contribution of research, as a cost saving, may be computed and the stream of cost savings may be evaluated.23 Direct Estimation ofS upply The third alternative approach is direct estimation of supply. The literature on single-equation, supply-response models for commodities is large, and the range of issues and approaches is far too great to be dealt with in any detail here. The key issues in supply-response analysis were identified sixty years ago in an article by Cassells (1933) as being how to deal with expectations and dynamics; these issues continue to be difficult. The virtue of single-equation models is that they allow considerable flexibility in the treatment of these ,topics. For estimating the returns to research on a particular commodity (or commodity aggregate), the direct estimation of a supply-response model may well be a better alternative than estimating a production function, precisely 23. One virtue of this approach, relative to the primal approach, is that from a cost function of this type, it is possible to obtain explicit evidence on the distributional impact of research among fixed factors. One of the disadvantages, however, is that the choice of functional form may implicitly dictate the nature oft he factor biases if care is not taken to allow the effects of technological variables to be treated as flexibly as the effects of prices. 114 Econometric Measurement of the Effects of Research because it permits the dynamics of supply response to price to be modeled in some detail along with the dynamics of supply response to research. In general form, a supply-response model for output may be written as Qt=q(Pt , Wt,tt, Ut ) (3.8) where q(.) is the supply function in which Q, is the output produced given values of a vector of expected prices of output, P" and of conventional inputs W" a state of technology, t and other uncontrolled variables, U" as defined above. Loosely combining e" quations 3.8 and 3.2 through 3.4, we can suggest a reduced-form relationship between investments in research and output (or productivity) in which current output depends upon current expected values for prices of output, P" and of conventional inputs, W" indefinitely long lags of past investments in agricultural research, R,_,. and extension, E,_., the stock of human capital, H" other factors (such as fixed inputs, infrastructure, and changes in input quality), Z" that are not captured by the measured input prices, and uncontrolled factors, U so that " Qt = q (Pt , W t , Ut , td(Rt, ., ., Rt-L~' (Et, ... , Et- LE ), Ht , Zt]) (3.9) = q(Pt , Wt , Rt- r , Et-e ' Ht , Zt ,U,) for r, e =0 to 00 Few studies have taken this type of approach in applications to research benefits, even though it dominates the more general supply-response litera­ ture (e.g., Zentner 1982; Fox, Brinkman and Brown-Andison 1987). Once the supply model has been estimated, it can be combined with a model of demand to translate the measured supply shifts into measures of the size of research benefits and their distribution between producers and consumers, using the methods sketched in section 3.3 and described in detail in chapters 4 and 5. Modeling Technical Change As discussed in chapters 2 and 4, the particular form of technical change may have important implications for the size and distribution of benefits (e.g., Duncan and Tisdell 1971; Scobie 1976; Jarrett and Lindner 1977; Lindner and Jarrett 1978; Rose 1980; Norton and Davis 1981). Three alternative approaches to specifying technical change are (a) directly incor­ porating technical-change variables in the function (e.g., Binswanger 1974; Bouchet, Orden and Norton 1989; Kohli 1991), (b) distinguishing between observed and effective quantities and prices, and output- or input-augment­ ing technical change (e.g., Dixon et al. 1982), and (c) using a varying-param­ eter specification in which the coefficients of a static model are themselves functions of technical change (e.g., Fulginiti and Perrin 1992). Econometric Measurement of the Effects of Research J J5 In a profit function, for example, these three specifications may be represented as 1t = g(P, W, Z, 1: I a) (3.10a) 1t =g [P(1:), W, Z I a] (3.10b) 1t = g[P, W, Z I a(1:)] (3.10c) where 1t is variable economic profit (i.e., the return to fixed factors), P is a vector of output prices, W is a vector of prices of variable factors, Z is a vector of quantities of fixed factors, 1: is a vector of technology indexes, and a is a vector of parameters. Martin and Alston (1992) illustrate the effects of these three kinds of technical change using a second-order Taylor-series expansion around a quadratic profit function. The first specification of technical change is well known from the empirical literature on the estimation of flexible functional forms. Technical-change variable(s) enter the function in the same way that a quasi-fixed factor would, except that they receive a zero factor return at the level of the firm. The second specification distinguishes between observed and effective quantities and prices. In this widely used approach, technical change increases the effective quantity of a good associated with a given physical quantity. An important feature of this specification is that there is a corresponding change in the effective price of the good; an increase in the effective quantity of an output provided by each physical unit will lower the effective price relative to that of the physical units. Using this approach, the relationship between physical and effective quantities of a particular good (i.e., input or output), Qi' can be represented by Qi = Q; . 1::, where Qi is the observed quantity of the good, Qi * is the effective quantity of the good, and T; is the level of output-augmenting or input-augmenting technical change for good i. The corresponding relationship between observed and effective prices is Pi = Y; 11::, where P/ is the effective price of the good, Pi is the observed price, and T; is the augmentation factor. When Qi is an input, input-saving technical advance is represented by a decline in 1:~, which reduces the physical quantity of the input required for one effective unit and also lowers the effective price relative to the actual price. When Qi is an output, an increase in 1:~ represents output-augmenting technical change; an increase raises the physical quantity associated with a given effective quantity and raises the effective price for a given actual price.24 Producers are represented 24. For example, consider a fann manager for whom effective outputs are measured in hectares of particular crops. A technical advance that raises yield per acre of one crop without changing its input mix increases the actual output (tonnes) attained per unit of effective output (hectares). While its actual price (in $/tonne) has not changed, its effective price (in $/hectare) has increased. Maximizing over effective 116 Econometric Measurement of the Effects of Research as optimizing over effective quantities and prices rather than observed quantities and prices. The profit function is then defined by replacing the variables with their corresponding effective values. The third way of incorporating technical change is to allow the parameters of the model to be expressed as functions of a scalar technology index. In this approach, it is important for the functions that define the parameters to be chosen so that the desired parametric restrictions hold over the region of interest. This specification makes transparent the need to ensure that any shift remains consistent with theoretical restrictions. It has been the most popular treatment of technical change in both primal and dual approaches. Typically, technology indexes, whether they are time trends or R&D vari­ ables themselves, are included additively - basically as modifiers of only the intercepts of equations that are linear in variables or their logarithms. 3.1.2 Nonparametric Approaches With both the primal and dual approaches, there are a number of concerns about the selection of functional form, the specification of technical change and how research expenditure variables enter the model, and the lag structure. The attraction of the nonp arametric approach to the analysis of production and the evaluation of research is that it avoids the use of functional forms altogether (hence the term "nonparametric"). The foundations for this type of approach are papers by Hanoch and Rothschild (1972) and Varian (1984), and a good exposition can be found in the text by Varian (1992).25 Chavas and Cox (1988) extended the models from Hanoch and Rothschild (1972) and Varian (1984) to include technical change.26 Subsequently, Cox and Chavas (1990) applied that approach to a productivity analysis of U.S. agriculture and Chavas and Cox (1992) analyzed the effects of research on productivity. Conventional approaches to modeling production involve estimating a parametric model and evaluating the properties of the estimated model. A disadvantage of this approach is that the results may be influenced by the functional form chosen for the model. Some such effects are trivially obvious (e.g., the use of a Cobb-Douglas model imposes the restriction that elastici­ ties of substitution are one); some others are more subtle (e.g.; even among quantities and prices. it will be optimal to increase the effective output of this crop by withdrawing resources from other crops. Thus. actual output of this crop will go up both directly through higher output per hectare and indirectly because the higher effective price draws resources from other activities. 25. Recent developments with applications to agriculture and agricultural productivity are reported in papers by Chavas and Cox (1988. 1992). Cox and Chavas (1990). Mullen. Cox and Foster (1992). and Lim and Shumway (1992). Examples outside agriculture include Chalfant and Wallace (1992) and Aacco and Larson (1992). 26. See also Fawson and Shumway (1988). Econometric Measurement of the Effects of Research 117 locally flexible functional forms, there can be substantially different findings about technical change or scale or substitution effects estimated using a particular data set). The problem is that the true functional form cannot be known but a functional form must be imposed as ajoint hypothesis with any other hypothesis test. The nonparametric approach to production analysis avoids the imposition of functional form as a joint hypothesis. Instead, in this approach, the data are checked for consistency with axioms of behavior. The two primary axioms of interest are the Weak Axiom of Cost Minimization (W ACM) and the Weak Axiom of Profit Maximization (W APM) as defined by Varian (1984) and as described by Chavas and Cox (1988). , Consider a competitive firm that maximizes profit and faces the decision problem: 1t (P ,'t ,h) = Max P'X subject to g(X , 't) ~ h (3.11) X where X is a "netput" decision vector (with positive elements corresponding to outputs and negative elements corresponding to inputs), and P is the vector of corresponding prices. The function g(X, 't) represents technology, 't > 0 is a technology index, h is a scalar, and 1t(.) represents the indirect objective function. It is assumed that g(.) is strictly decreasing and concave in X. The firm is observed choosing X T times, Xl' ... , XTt and each observation is associated with a situation t characterized by market prices P, and technology (hI' 't,), t = 1, ... , T. The nonparametric approach to production analysis tests the consistency of the actual decisions X ={ Xl' ... ,Xr} with the optimization problem given by 3.11 (i.e., whether X, could be equal to }(P,. 't,. h,), the solution of the maximization problem above), without ad hoc specification of the functional form for g(X, 't), 1t(X, 't, h), or }(X, 't, h). The parametric tests involve checking a set of inequalities to see whether a production function could exist that would "rationalize" the data in the context of the maximization hypothesis. One such set of inequalities tests for consistency with profit maximization, another for consistency with cost minimization.27 Profit Maximization - WAPM When h is equal to zero, technology g(X, 't) represents the implicit produc­ tion frontier. In the absence of technical change, 't, = 1, for t = 1, ... , T, and g(X,. 1) = O. Profit maximization implies P:(X, - X.) ~ 0 for all sand t. This 27. Nonparametric methods can also be used to investigate many of the characteristics of production technology, such as the nature of returns to scale or various separability restrictions, or production efficiency. Here we are concerned, instead, with the use of these techniques to measure the size and factor bias of technical change in a data set and to evaluate the contribution of research to those changes. 118 Econometric Measurement oft he Effects of Research inequality is WAPM (Varian 1984, p. 584). The interpretation is that if profits are maximized at time t, then it should not be possible to choose any other bundle Xs and obtain greater profit with time t prices. The question compares the time t bundle with the bundles chosen at other times in the data set. Cost Minimization - WACM Cost minimization is obtained from equation 3.11 when h, denotes output (of a single product) so that g(X, '1:) represents the explicit production function and X, is the input vector (redefined here to be positive) with input prices equal to Wr As shown by Varian (1984, p. 581), in the absence of technical change, if h, ~ hs then cost minimization implies that W,'(X, - Xs) ~ 0 for all s and t. This inequality is WA CM. The interpretation is that the inputs chosen in time t must cost less at time t prices than any other bundle that could produce greater output than h,. If costs were minimized at time t, then it should not be possible to choose any other bundle Xs and obtain greater output without incurring greater cost. The question is examined by comparing the time t bundle with the bundles chosen at other times in the data set. Technical change is precluded by assumption in these procedures; one would expect a substantial technical change to lead to a violation of WA PM or WA CM. Chavas and Cox (1988) extended these procedures to incorporate Hicks-n~utral technical change, using '1:,.28 Subsequently Cox and Chavas (1990) allowed for both biased and neutral technical change. When technical change is allowed, the linear programming problem becomes one of solving for the minimum set of technical changes necessary to "rationalize" the data. As described by Mullen, Cox and Foster (1992), Cox and Chavas (1990) showed that the existence of a solution to the following T(T- 1) inequalities is necessary and sufficient for the data to be consistent with profit maximization under the input- and output-additive augmentation (translating) hypothesis (3.12) where Y, denotes a scalar single output with associated price P" X, denotes a vector of inputs with associated prices R" A, denotes output augments (higher values denote higher productivity), and B, denotes input augments (where B > 0 implies factor-saving input bias and B < 0, factor-using input bias). Furthermore, if such a solution exists, then Y(As' X,)/Y(A" X,) = 1 + (As - A,)/Y, is an index of total factor productivity that measures the shift in the production function between time t and time S.29 28. In the notation used by Cox and Chavas (1990) Ads equivalent to the 't, we use in this chapter. 29. More recently, Mullen, Cox and Foster (1992) have shown how to obtain nonparametric measures of total factor productivity, including input-based measures and output-based measures, with an applica- Econometric Measurement of the Effects of Research 119 Chavas and Cox (1992) proposed using their nonparametric approach to analyse the effects of research on productivity. They enumerated five virtues of the approach relative to parametric approaches (p. 584). First, it requires no a priori restrictions on substitution possibilities among inputs (e.g., via parametric restrictions). Second, the method allows joint estimation of the production technology, technical change, and the effects of research on technical progress using very disaggregate inputs. Third, the approach al­ lows considerable flexibility in the investigation of the length and shape of the lag distribution between research and productivity. Fourth, the method permits an investigation of the separate effects of private research and public research on technical progress. Finally, the method is empirically tractable, in that it requires only a standard linear programming algorithm. They illustrate these points with an application to U.S. agriculture. While the nonparametric approach is very attractive in principle, for the reasons outlined by Chavas and Cox (1992), there remain some questions that must be answered in practice before we can countenance a wholesale abandon­ ment of parametric approaches. In the context of applications of nonparametric approaches to consumer demand, there has been concern about the properties of the nonparametric tests. The limited Monte Carlo work that has been done suggests that under commonly observed conditions, it can be difficult to obtain definitive results that can be taken with any confidence.3O Presumably, there is potential for similar questions to arise in the applications to production, so we remain to some extent agnostic on the question of how generally the nonpara­ metric approaches are applicable until further results have been obtained that establish the properties of the measures. In addition, some recent work by Chalfant and Zhang (1994) has shown that the Chavas and Cox methodology (as developed and employed by Chavas and Cox 1988, 1992, and by Cox and Chavas 1990) is seriously flawed: the measures of technical-change bias are not invariant to the scaling (i.e., the choices of units) for prices and quantities?1 Chalfant and Zhang (1994, p. 13) tion to Australian broadacre agriculture. 30. Alston and Chalfant (1992) have reviewed the literature and provide a useful heuristic discussion of the application of tests for consistency of conswnption data sets with revealed preference axioms. They show how to apply the equivalent of Chavas and Cox's (1988, 1992) approach to conswnption data, following Sakong and Hayes (1993) to solve for the set of minimum "taste changes" necessary to rationalize tre data. An important aspect of this work was the issue, raised by various writers, of the "power" of nonparametric tests and the alternative approaches that might be used to impose nonsample information (such as restrictions on elasticities or changes in elasticities from observation to observation) as restrictions on the analysis in orner to increase power. While this work was conducted in relation to consumer data, tre arguments are perfectly symffietrical and apply with equal force to the production side. 31. Chalfant and Zhang also point out that the same criticism applies to application of analogous procedures to measuring the size of shifts in consuiner demands (e.g., by Alston and Chalfant 1992 120 Econometric Measurement of the Effects of Research propose that "Nonparametric methods that use the Chavas and Cox method of minimizing a (weighted) sum of taste change or technical change parameters require weights attached to these adjustments to the data whose scale varies inversely with the units of quantities .... Otherwise, the estimated adjustments are not invariant to scaling." In other words, in measuring technical-change biases, a price vector must be used to weight the adjustments to the input-quan­ tity data required to restore consistency with WAPM or WACM in order for the results to be invariant with units. A disappointing aspect is that a particular price vector (e.g., corresponding to a particular observation in the sample, say, for a particular year's data) must be chosen for the analysis to weight the quantity adjustments for every observation of quantities. The results may depend on the choice of the set of relative prices to be used, and the choice is essentially arbitrary. Further work must be done to establish ways to achieve in variance that are less arbitrary. However, there can be no doubt that poten­ tially, the nonparametric methods have much to offer, as a complement to a parametric analysis at a minimum, and there are grounds for being optimistic about further developments in this relatively new set of techniques.32 3.1.3 Index-Number Approaches A working knowledge of index-number theory and practice is indispensi­ ble for econometric attempts to measure production technologies and the effects of research on those technical structures. Modem index-number procedures enable consistent and economically meaningful measures of input and output aggregates to be formed. These aggregate price and quantity measures can then be used to summarize and describe production-related data, as well as estimating aggregate production, cost, and profit functions. Index-number procedures also enable partial and total factor-productivity measures that provide summary indications of the nature of growth in agricultural output or agricultural productivity to be constructed. These productivity measures can then be used in conjunction with econometric approaches to determine (and subsequently value) the output-enhancing effects of research, extension, and other "unconventional" inputs. Indexes Index numbers are involved in virtually all quantitative economics. In the index-number approach to productivity measurement, this involvement is following Sakong and Hayes 1993). 32. For a recent application of nonparametric teclmiques to an assessment of international develop­ ments in productivity and teclmica1 change, see Fare et a1. (1994). Econometric Measurement of the Effects of Research 121 explicitly obvious, whereas in other approaches to measure the effects of technical change, the importance of the role of index numbers and index­ number theory is not so apparent. Almost always, however, we must use measures of prices and quantities of inputs or outputs that have been aggre­ gated over people, places, or time and across different qualities. The aggre­ gated quantities (and prices) are indexes that mayor may not suffer from index-number problems - in which quantity changes due to relative price changes are properly distinguished from other types of quantity changes. Often the only data that are available are preaggregated, and we might not know whether the aggregation was done in a way that is consistent with economic theory or, instead, in a way that leads to an over- or understatement ofthe actual changes in quantity (or prices) over time. Thus, an index-num­ ber problem may be inherent in the data~ An index-number problem could also be generated by the analyst in making choices about how to aggregate data. In this section, we layout the theory of index numbers that is relevant for constructing the aggregated measures of inputs or outputs that are typically used in empirical work. This theory also provides the foundation for the methods for measuring total factor productivity, discussed below. Aggregating inputs and outputs: An economic approach to constructing index numbers is to choose a method for weighting the quantities that make up the aggregate being indexed - one that is consistent with economic theory and accommodates the optimizing responses of economic agents.33 Inappropriate treatment of the optimizing responses leads to inappropriate weightings and over- or understatement of the aggregate quantity change. To illustrate this possibility, consider panel a in figure 3.2 where Q represents a particular quantity of output with a given state of technology, t =O . When the price of X2 relative to the price of XI is given by Wo, cost-minimizing producers will use the combination of inputs XIO and X20 at a point a to produce output Q. If the price of X2 falls relative to XI (i.e., from Wo to WI)' producers will substitute X2 for XI to minimize the costs of producing the same output, Q, at point b (using input amounts Xli and X21). Now assume that an input aggregate X = g(XI, X2) is formed using a simple linear aggregation of the two inputs weighted by the respective factor prices. If the 33. By contrast, there is an axiomatic approach to index numbers (and implicitly, TFP measurement) that stems from the worlc of FISher (1922) and more recently Eichorn (1976) and Eichorn and Voeller (1976), and reviewed by Diewert (19888 and b). The idea is to "test" the suitability of an index against a number of properties or conditions (e.g., characteristicity, detennination, positivity, and various homoge­ neity and monotonicity properties) that are considered desirable for an index number. These (often statistical or empirical) properties need not bear any relationship to an economic theory of measurement or aggregation. And because no one index number satisfies all known properties, the choice of indexing procedure is based largely on the discretion of the analyst. ' 122 Econometric Measurement of the Effects of Research Figure 3.2: Technical change and input substitution effects (a) Relative input price change; no technical change o (b) Relative input price change; with technical change relative factor prices indicated by the slope of Wo are used as weights, the input aggregate at point a would be valued lower (Le., would cost less) than the point b aggregate (notice xto < xto)' Consequently, an index of produc­ tivity that expresses output per unit of aggregate input would be greater if measured at point a than at b. Econometric Measurement of the Effects of Research 123 In a similar way, a linear aggregator that used relative input prices given by the slope of WI would result in the input aggregate measured at a being valued higher (i.e., more expensive) than if measured at b (notice X'tl > xtl and, equivalently, ~l > ~l)' Thus, the resulting productivity index mea­ sured at a would be greater than that measured at b. In these instances, measured productivity has changed in the move from point a to point b, even though there has been no change in the state of technology as reflected in the shape and position of the isoquant Q. If there is a simultaneous shift in relati ve prices from Wo to WI and a (neutral) productivity improvement so that the same amount of output could be pro­ duced with fewer measured inputs, there is an even more subtle problem wherein the effects of technical change and factor substitution can get con­ founded. In panel b of figure 3.2, the shift of the isoquant from Qo to Q\ (representing the same quantity of output being produced with two states of technology, i.e., 't = 0 and 1, respectively) and the relative price change would lead to a change in input combination from a to d. Alternative measures of cost savings could be made with either price vector if we ever observed producers responding to the new technology at old factor prices (shifting from a to c) or using the old technology with new factor prices (shifting from b to d). The technological change measured at the original prices would be a reduction in cost from X'to to ~o. Since c is not an observed input cost, measuring the change when the input combination shifts from a to d using original prices implies smaller cost savings (a reduction from X'to to xto)' The cost savings are understated because the substitution effect is not factored out. The change in technology measured using new prices (shifting from b to d) would be a cost saving given by the reduction from ~I to ~I' Since b is not observed, measuring the change from a to d using new prices indicates a reduction in costs in moving from ~l to ~l that overstates the cost saving. The longer the time period over which a single set of fixed price weights is used to calculate aggregates, the more likely we are to observe the effects of changes in relative prices that are confounded with changes in technology. We can understate or overstate a drop in inputs (with a corresponding upward or downward bias in a productivity index) depending on the time period over which prices are held fixed in the calculations. Aggregation problems arise with respect to outputs that in many practical situations, are analogous to input-aggregation problem. In numerous cases, Q represents an output aggregate such as "grain crops," "livestock products," or quite commonly, "total agricultural output." The measure is formed over a set of outputs, so a real output aggregate is required. Again, an index that does not confuse movements along an unchanging production possibilities frontier (PPF) due to shifts in relative output prices with shifts in the PPF is desirable. 124 Econometric Measurement oft he Effects ofR esearch Panel a in figure 3.3 illustrates the problem of aggregating across multiple outputs. If an output aggregate Q = h(Q1 • Q2) were formed using a linear aggregation of the two outputs weighted by their relative prices given by the slope of po. the output aggregate at point a (producing output amounts QIO and Q20) would be greater than the point b aggregate (notice eno > Qto). Con­ versely. using the relative output prices PI would result in a lower-valued Figure 3.3: Technical change and output substitution effects (a) Relative output price change; no technical change Q1 Qfo Qto o (b) Relative output price change; with technical change Q1 Qfo eto o Econometric Measurement of the Effects of Research 125 output aggregate at point a compared with point b (notice (!t) < at)~. Using only one set of prices over a long period of time (whether they are Po or PI) is once again likely to indicate a shift in the PPF when in fact there has been none. When improved technology shifts the frontier (from PPFo to PPF\, in panel b of figure 3.3), allowing increased quantities of either output to be produced with the same bundle of inputs, linear output aggregates will confound the effects of substitution and technical change if relative prices also change. If the economy moves from a to d because of technological change and a simulta­ neous increase in the relative price of good 1, movements from a to c or b to d (if observed) could be evaluated with either set of prices and give information on technical change alone. But since neither b nor c would be observed, we must rely on observed output bundles a and d. If evaluated at original prices, Po, the technological change is understated (Q'{o- (!to < (!to - (lto)· If the change in output is evaluated using new prices, PI' the technological change is overstated (Q~) - Q~) > ~) - Q~). Divisia indexing procedures: The method commonly used to minimize the impact of relative price changes when forming aggregate quantity in­ dexes is to use a Divisia indexing procedure. As Richter (1966) and Hulten (1973) show, the Divisia index is desirable because of its invariance prop­ erty; if nothing real has changed (e.g., the only input quantity changes involve movements around an unchanged isoquant) the index itself is un­ changed.34 The formula for an index of aggregate inputs is . .1) D Jt W,' Ms Xl, =Xh exp W'X ds (3.13) b of of where Xlf is the index value of the base period, b, XI~ is the index value in period t, Xs is a vector of input quantities, Ws is a vector of input prices, and M denotes changes in inputs. If the economy of interest - measured at either the sector, industry, or even, farm level - is moving along an unchanged transformation or production surface, the sum of changes in inputs, M, weighted· by current factor prices, W, will be approximately zero; the index will be unchanged. If the economy's transformation surface is shifting, current price-weighted changes will be different from zero, leading to changes in the index value. This invariance property is dependent upon a maintained assumption of optimizing agents. 34. The discussion here deals with constructing indexes of quantity aggregates. An exactly analogous situation pertains to the construction of indexes of price aggregates in which quantities are used as weights. If appropriate aggregation procedures are used, the resulting price index measures the nominal growth of aggregate prices over time. But like its quantity countelJlart. the index can also be considered "real" in the sense that it abstracts from changes in the mix of quantities in the price aggregate being measured. 126 Econometric Measurement of the Effects of Research Unfortunately, the calculation of a chained index such as the Divisia index requires continuous measurement of input prices and quantities. To facilitate calculations in a way that minimizes measurement error, a discrete approxi­ mation can be used to link quantities and prices across adjacent years. In any discrete approximation, some information is lost. The advantage of using a chained index involves the notion that recent quantity changes are weighted by the most recently observed prices. Intuitively, these indexes are attempt­ ing to evaluate current behavior at current prices. In proceeding from the base period to some distant period t, the small steps are chained together to' minimize the measurement error that is possible when only base-period prices and period t prices are used to evaluate real changes in input quantity. Many discrete approximations to the Divisia index are possible. Richter (1966) proposes what others have called a Laspeyres approximation: DL_ DL [ W',_I (X,-Xt-\)]_ DL W',_IX, XI, -XI,_I 1 + W' X -Xlt-\ W', -I X -(3.14) ,-I ,-I ,-I In a similar way, we could define a Paasche approximation: IDP X )] W'X XPP , =X ,-I [ 1 + w'' (X - W',X ,-I =X PP t-\ " , W ~I , 'X (3.15) ~I or a Fisher ideal approximation: 1 1 DF _ DF (W"_I X, J2 ( W/X, J2 XII -XII-I W' X W' X (3.16) I-I I-I I I-I The Tornqvist (1936) or Tornqvist-Theil discrete approximation of the con­ tinuous version of the index given by 3.13 uses both current and lagged cost shares in weighting changes in input quantity, yielding SiJ DT DT m Xi,1 - I XII =XII_I IT (X J where S;,/ =2 (Si,1 + Si,I-I) (3.17a) i=1 i,I-1 and the input cost share for factor i in period t is given by Si,1 = Xi,1 Wi./( .i Xi,1 Wi'IJ (3.17b) 1=1 Growth rates in the input aggregate using the same formulation are calcu­ lated using cost-weighted sums of changes in input quantities: Econometric Measurement of the Effects of Research 127 m In(XIPTI XIP-n =1 : ~ (Si,t + Si,t-l) In (Xi,tIXi,t-l) (3.18) i=l where In(X~T IXe~) is the rate of change in the input Divisia index from period t-l to t. 35 In practice, a series is formed by setting xt{ =1 .0 for any arbitrarily chosen base year b and accumulating the measure forward and, if necessary, backward in time according to equation 3.17.36 Alternatively, the index can be constructed by compounding the growth rates calculated using 3.18 (forward and backward) from the base period. As with input indexes, a Divisia aggregation procedure is used to mini­ mize the impact of relative changes in output prices when real output aggregates are formed. The Tornqvist-Theil output index, Q~T, is n Q S. 1,1 Qi,> T =Q ItD-\T n.. .(:Q= :iL J where -Sj,t )=2 (Sj,t + Sj,t-I) (3.19a) j=l i,t-I where the weight for commodity j in period t is its output revenue share: sj,t =Qj,t pj/( ,i: Qj,t Pj,t) (3.19b) J=I Growth rates in the index are the sum of moving-average-share-weighted relative changes in the individual output quantitites: n In (Qi,>T IQIf-r> = 1: ~ (Sj,t + Sj,t-I) In(Qj,tIQj,t_I) (3.20) j=1 These are the output-index counterparts to the aggregate input indexes defined by equations 3.17 and 3.18. Forming the index follows the equiva­ lent procedure described in the input aggregation discussion. The advantage offered by any of these approximate Di visia indexes is that any substantial drift in relative prices over time is accommodated by rolling weights. In addition, theoretical work on superlative index numbers by Diewert (1976) and Lau (1979) has established that the approximate Divisia indexes are exact for specific aggregator functions. If vectors of inputs are appropriately aggregated with linear functions, the Laspeyres and Paasche 35. This rate-of- change form is the most common representation of the index. To recover the index in level fonn (i.e., equation 3.17), simply take the antilog 00.18 and multiply both sides by XIt-!. 36. Instead of using equation 3.18, some analysts (e.g., Romano 1987) have used In(XIt+/XIb) = 1:Z!) t (Si,t+j + Si,b) In(Xi,t+/Xi,b) for all j = I , ... , T Rather than using quantities and factor shares from adjacent years, in this (inconect) formulation. base period quantities and cost shares are used throughout. 128 Econometric Measurement of the Effects of Research approximations of the Divisia offer exact measures of real quantity changes. The Fisher approximation is exact for quadratic aggregator functions. The Tornqvist Divisia index is exact for the more general class of translog aggregator functions. If disaggregated data are difficult to obtain, we may be forced to use fixed-weight indexes, such as a Laspeyres or a Paasche index, and accept any resulting biases. However, the same amount of information is required to construct the alternative chained indexes, so what basis is there for deciding which of the Divisia approximations to use? The aggregator functions for which the various indexes are exact provide some guidance. For instance, if the translog function is deemed to be the appropriate aggregator, it implicitly takes every input type to be in some sense essential to the aggregate since the translog function is undefined when anyone of the possible inputs is zero. In aggregating national accounts, the categories are typically so broadly defined that the requirement for positive quantities is not a problem. If, however, the aggregate is being formed over finely disaggregated inputs, comer solutions in which some inputs or outputs are not used or produced over part of the sample are quite likely. A linear or quadratic aggregator function, which implicitly allows for partial or complete specialization, is defined as long as at least one input is used and thus may be more appropriate than the translog in applications where some quantities of inputs (or outputs, when an output aggregate is being formed) take values of zero for some observations. Another practical consideration is the degree to which the approximation method provides some smoothing of price weights. When commodities whose prices vary widely from period to period but whose quantity re­ sponses may lag one or more periods are aggregated, there may be less economic sense to employing weighting schemes that make use of only one period's prices. The property of characteristicity (Drechsler 1973) would imply using the price weights most specific to the economic activity being measured. In this respect, the Tornqvist approximation may be more appro­ priate than the Laspeyres for aggregating quantities when there is reason to think that producers are reacting to local prices but cannot do so instantane­ ously. The Tornqvist approximation implicitly smooths prices by averaging current and previous value shares when each value share is calculated using contemporaneous prices and quantities. Direct versus implicit indexes: The discussion so far describes how to calculate direct quantity and price indexes, which are equally feasible if suitably disaggregated data are available. But, in practice, data constraints sometimes require the calculation of implicit indexes instead. Implicit in­ dexes are consistent if the direct index satisfies Fisher's weak- factor-reversal Econometric Measurement of the Effects of Research 129 property, wherein W/X, XI, . WI, = w: 'X (3.21a) ,-I ,-I where Xlt and WIt are, for example, indexes of the input quantity and input price in period t, and the right-hand side of equation 3.21 a represents the ratio of the current to previous period's total cost of production. Given these expenditure ratios and, say, a quantity index, the corresponding price index can be derived, implicitly, by rearranging 3.21a so that W,'X, WI, = I (3.21b) Wt-l X,_I XI, Unfortunately, an implicit Tornqvist-Theil index does not conform ex­ actly to the corresponding direct index because the translog aggregator function, for which a Tornqvist-Theil index is exact in the sense of Diewert (1976), is not self-dual. But, provided the period-to-period variation in relative prices is not "too great," the differences between a direct index and its implicit counterpart are "relatively small," i.e., equation 3.21 b only holds approximately in the case of a Tornqvist-Theil index, but the approximation may be reasonably close (Allen and Diewert 1981). Productivity Measurement Productivity indexes are commonly constructed measures of the relation­ ship between inputs and outputs. The most widely used productivity measure expresses output per unit of a particular input such as land or labor. These partial factor productivities (PFPs) are generally defined as PFP.=.Q I Xi i= 1, ... ,m (3.22) where Q represents output and Xi represents input i. A more careful repre­ sentation is 1_f2l PFPi - JC = Xli i= 1, ... , m (3.23) I where Q* represents aggregate output, x: represents an aggregate of input Xi' and QI and Xli are the corresponding indexes of aggregate output and input Xi' respectively. Equation 3.23 reflects the fact that PFP indexes are usually formed using an output aggregate such as total agricultural output (which includes different crop, livestock, and even, forestry, and fisheries output), or wheat 130 Econometric Measurement of the Effects of Research output (which includes different grades or qualities of wheat). More subtly, but perhaps just as significantly, the denominator in a PFP index often groups together different classes or qualities of the input Xi' Within such a factor grouping, there are aggregation problems arising from heterogeneity that are analogous to the problems of aggregating dissimilar factors such as land, labor, and capital. For example, land inputs may include a mix of different land types such as pastureland and irrigated and rainfed cropland, while the labor input may reflect hours in agriculture that are a combination of hired labor and heterogeneous operator labor of varying ages, accumulated skills, and education. In assessing measured changes in PFP, it would be useful to know whether the total effective labor input in agriculture is increasing over time because more hours of the same type of labor are being employed or because the composition of the workforce has changed to include relatively more highly skilled labor. The finer the classi­ fication scheme is for outputs and inputs, the less likely it is to confuse changes in quantity with changes in quality and, consequently, misinterpret measured changes in PFP indexes. Even if properly constructed, PFPs pose particular problems for distinguish­ ing (research-induced) technical changes from (price-induced) substitution effects. PFPs are affected not only by advances in the state of technology (as indexed, for example, by t in equation 3.10) but also by changes in the effective quantities of other inputs used in production. Additional fertilizer will generally raise yields and, consequently, land productivity, while additional physical capital- be it in the form of an improved hand-held sickle or a bigger combine harvester - can raise labor productivity. A more general concept of produc­ tivity is required to distinguish between changes in output due to technical changes and those arising from changes in the mix of inputs due to shifts in their factor prices. A measure of total factor productivity (TFP) can be defined as TFP=Q (3.24) X where TFP measures (aggregate) output Q, produced per unit of an input aggregate, X. From equation 3.24 it follows directly that the observed proportionate rate of growth of total factor productivity, tjp" is simply equal to the rate of growth of measured output, q" minus the rate of growth of measured inputs, XI: (3.25) where Econometric Measurement of the Effects of Research 131 dTFPt 1 dQt 1 dXt 1 tfpt=~ TFP,' q, = dt Q,' and xt=Tt X, In principle, output and input aggregates could be formed using the Divisia indexing procedure defined by equation 3.13, and from equation 3.25, it would be possible to calculate tfp without explicitly identifying the functional form of the underlying production relationship. The only assumptions interposing between the data and the tfp measures are those that concern optimizing behavior, whereby technically efficient producers substitute around isoquants and production-possibility frontiers in response to changes in relative prices of inputs and outputs. If the underlying technology is input-output separable, to the extent that the output and input aggregates, Qt and XI' can be formed separately, then the 1FP measure follows directly. For relatively small changes in a variable Zt, dZ/2, =d lnZt "" InZt -lnZ1-I (i.e., proportionate rates of change, dZ/ZI' are approximately equal to logarithmic differences, In2, -lnZ1-I)' so a discrete approximation for equation 3.25 is (3.26) Thus, the rate of change in 1FP is obtained simply by taking the difference between the growth rate of the Tornqvist-Theil indexes of aggregate output and input quantities. Work by Diewert (1976), extended for the discrete-vari­ able case by Denny and Fuss (1983), showed that the use of the discrete Tornqvist-Theil approximation to the continuous Divisia index carries with it an implicit assumption that the underlying technology can be represented by a translog model. To the extent that the translog represents a second-order, local approximation to an arbitrary functional form, the Tornqvist-Theil index may be regarded as imposing fewer restrictions between the data and the tfP measure than would the Laspeyres, Paasche, or Fisher ideal index.37 But as Craig, Pardey and Deininger (1993) point out, agreement on the proper index formula does not resolve all measurement issues. Rather than proceeding mechanically in applying the index, important choices must be made about how many distinct inputs and outputs will be used in its construc­ tion. A fundamental insight from the literature on productivity measurement is that, regardless of the index formula used, a high level of disaggregation is required to avoid aggregation bias. Star (1974) showed that one is safe in using preaggregated inputs (e.g., taking all labor to be a single class of input) only if all inputs in the class are growing at the same rate or are perfect substitutes for 37. The approximation potential ofthe translog and other locaJly flexible fonns has been called into question in a number of studies. For example. see Chalfant (1983.1984) and Thompson (1988) (and the references therein for more general evidence) for a discussion ofthe relevant issues and an application to U.S. agriCUlture. 132 Econometric Measurement of the Effects of Research one another. If rates of change in higher-priced inputs exceed rates of change in lower-priced inputs, the rate of growth of the group will be biased downward in any index that fails to treat the inputs as separate components. Disaggregation by itself makes it more likely that inputs and outputs will be measured in units whose quality is constant over adjacent time periods. The finer the distinction among inputs, the more confidence we have that the Xs employed in the index equations 3.23 and 3.24 are truly comparable from year to year. The same holds true for the commodity quantities, i.e., the Qs, used to form a productivity index. Sources ofM easured Productivity Growth38 Measuring TFP growth and identifying the sources of this growth are two distinct but directly related activities. What are the potential sources of measured productivity growth? One possible source is changes in quality of inputs or outputs that have not been accounted for properly in the analysis. Another source is mismeasurement of input or output quantities due to index-number or data problems, for reasons other than unmeasured quality change.39 A third is improved technology as a result of either private- or public-sector R&D (and extension) or technology spill-ins." Input quality: Improvements in input quality could include improved machinery (a "tractor" today is not the same as a "tractor" 30 years ago), improvements in quality of labor with a more educated work force, and improvements in land quality (perhaps through addition of capital or in­ creased water rights), for example. Some of these improvements in input quality result from private-sector agricultural research on agricultural ma­ chinery, chemicals, and for many crops, seeds. Output quality: Si~ilar problems could arise in measuring the quantity of output, which is an aggregate of different types of outputs (as the input index is an aggregate ofthe different types of inputs). If the quality of, say, horticultural products had risen over time, and the measure of the quantity of horticultural products had not been adjusted accordingly, we would be understating the real growth in horticultural output, some of which was in 38. Reflecting on the sources of productivity growth, Schultz (1956, p. 758) wrote that ''The analytical task, as I see it, is to re-establish a strong and satisfactory linkage between input and output over time. In our efforts to do this, we would do well to place before us and keep in mind the characteristics of an ideal input-output formula for this purpose. It would be one where outputs over inputs . .. stayed at or close to one. The closer we come to a one-to-one relationship in our formulation, the more complete would be our (economic) explanation." These and related issues were further discussed by Griliches (I 963b ). 39, Of course to the extent that research-induced quality changes are embodied in outputs and conventional inputs and that these quality changes are measured properly, they should already be captured in the productivity index. Econometric Measurement of the Effects of Research 133 the form of quality improvement. If we measured output without adjusting for quality change, productivity growth would be understated if quality had improved; it would be overstated if quality had declined. Other measurement problems: Even when Divisia indexes are used in conjunction with finely disaggregated input and output data, so that index number problems are minimized, there might be some remaining problems of measurement. Perhaps the most likely one is that there is typically no explicit allowance for the reduction in the stock of environmental and natural resources associated with agricultural production in terms of land degrada­ tion, depletion of soil fertility, chemical pollution of air and groundwater, build-up of resistant pests and diseases with loss of natural predator popula­ tions, and so on. Environmental resources can be thought of as an unmea­ sured input into agricultural production, and part of the additional output may be attributable to consumption of the stock of natural resources.4O Most studies of the benefits and costs of agricultural research do not take this into account. One approach would be to include some natural resource accounting in the analysis, treating the resource base as one of the inputs in agriculture, but suitable data are not usually available for that.41 A second approach that could be useful in the context being considered here would be to adjust the measures of research benefits to reflect a disparity between the private and social costs of production. The latter approach is more feasible, although no one has done it yet. To be done meaningfully, it would require information similar to that used in the direct approach, although data for only for a few recent years may be sufficient. New technology: Putting aside measurement problems, the growth in measured total factor productivity that is attributable to changes in technol­ ogy is taken to represent changes in output for given inputs. In tum, there­ fore, this productivity growth can be attributed to past expenditures on research and extension by public-sector agencies, private research, and the spill-in of technology from elsewhere. Rates of Technical Change Production-function approaches: The most straightforward approach to formally linking notions of technical change with measured rates of produc­ tivity growth is to assume that an index ofthe state oftechnology, 't, can be 40. AjTow of services from the stock of natural resources is always used in production but the concern here is with changes in the stock itself that will imply a reduction in future service flows. 41. For attempts to incorporate the natural-resource-degradation and environmental-extemality ef­ fects of agricultural production directly into productivity indexes, see Archibald (1988), Oskam (1991), Ehui and Spencer (1993), Antle and McGuckin (1993), and Alston, Anderson and Pardey (1994). 134 Econometric Measurement of the Effects of Research incorporated directly in a production function such that42 Qt = f(Xt ' tt) (3.27) Hence, technological progress (i.e., where dt/dt > 0) is perceived as an upward shift of the production function, f(.), or, equivalently, as a downward shift of the isoquant map, as depicted in the shift of an isoquant for given output from Qo to QI in figure 3.2b. Rates of change in output over time can be partitioned into components due to changes in measured input use and to those due to changes in the state of technology. By differentiating Q, with respect to time and dividing throughout by Q, (or, equivalently, taking logarithms and differentiating InQ, with respect to time), we get • dlnQt alnQt dlnt -+It m dlnXi,t --=--- :eQ.-- (3.28) dt alntt dt i=1 ,I dt where EQ,; =d lnQ/dlnXit is the elasticity of output with respect to the quantity of the ith input. If competitive equilibrium is assumed, so that output price equals marginal cost and factors are paid the values of their marginal products, then equation 3.28 may be rewritten, equivalently, as43 dlnQt alnQt dlnt =- t m dlnXi,t -d-t a-ln-t -+dt E I.S·-- .29) t Q,C i=1 I,t dt (3 where S;,t =X ;! W;/I.; X;,t W;,t is the ith factor's share of total cost in time t and EQ•c =( dlnC/dlnQ,r l is the inverse of the elasticity of cost with respect to output, which can be used to define returns to scale. Notice that under constant returns to scale, where EQ•c = 1.0, equivalence between equations 3.28 and 3.29 requires the ith factor share to be equal to the ith factor's output elasticity, i.e., EQ.; = S;.t' Using more compact notation, we can transform equation 3.29 to obtain a measure, g" of the primal rate of technological change, i.e., gt =q t - EQ,cXt (3.30a) where -42. Solow (1957, p. 3l2) used t rather than 'ttin his specification 00.27, noting that ''the variable t for time appears ... to allow for technical change. It will be seen that I am using t~ phrase 'technical change' as a shorthand expression for any kind of shift in t~ production function. Thus, slowdowns, speedups, improvements in the education of the labor force, and all sorts of things will appear as 'technical change ... , 43. To obtain this, we use the results where, under competition, each factor is paid the value of its marginal product (so that Of(X) lax; = W; IP) and price equals marginal cost (P = aC/aQ). Combining these results, and manipulating the expression yields Of(X) lax; = W; IP =W ; I(ac taQ) =W ; (iJQ laC) = W;(QlC)(dlnQ IdlnC) =W ; (Q IC)EQ.c. We can also define dlnf (X) =4 f(.)-1 [Of(X)/ax; 1d X;. Substitut­ ing the first result into this expression, simplifying, and consolidating terms yields dlnf(X) = 4 (W;X; lC)fQ.c dln(X;). Econometric Measurement of the Effects of Research 135 _(alnQt dln'tt] dlnQt -l: s. dlnQj.t gt - aln'tt dt ' qt - dt - :i 'j,t dt and (3.30b) dlnKt dlnK;,t X t --dt- == l:. S· t-- 1 I, dt and q, and x, are the Divisia indexes of growth in output and input, respec­ tively.44 In words, equation 3.30a expresses the primal rate of technological change as the rate of change of output, minus a scale-adjusted index of the rate of change in input. Hence, under the assumption of constant returns to scale, input-output separability, efficient and optimizing producers, and disembodied technical change of the extended Hicks-neutral type, the rate of change in TFP given by equation 3.25 also measures the rate of technological change or shift of the production function. ' Cost-function approaches: It is also possible to use the cost or profit function to derive a dual rate of technological change that is a counterpart to the primal rate. Assuming that technically efficient producers act to minimize production costs at any given level of output, a minimum cost function can be written as Ct = c(Qt, W t , 'tt) (3.31) where cO is the cost function that defines C, as the minimum cost of producing any output of Q" given a vector of input prices W, = (WIJ ' ••• , Wm ,,) and the state of technology indexed by 't,. Differentiating equation 3.31 with respect to time and dividing throughout by C, (or, equivalently, differentiating InC, with respect to time) yields dlnCt alnCt dlnQt m alnCt dlnW;,t alnCt dln'tt --=----+l: ----+-- -- (3.32) dt alnQt dt ;=1 aln Wi,t dt aln'tt dt By Shephard's lemma, the optimal (cost-minimizing) factor-demand equa­ tions are given by the derivatives of the cost function with respect to the factor prices: X~, =a c/aw~, =a c(Q" WI' 't,)/aWi". An equivalent representation is that factor cost-share equations are given by Si" =a lnc/alnW i" = (WjC,)ac(Q" W" 't,)/aWi". Thus, equation 3.32 may be rewritten as _ alnCt dln'tt = alnCt dlnQt + ~ Si t dIn Wi,t _dlnCt (3.33) aln'tt dt alnQt dt i=l 'dt dt Using more compact notation, we can transform equation 3.33 to obtain 44. Most analysts use time, t, as the technology index, implicitly assuming that dlnt,/d t = 1.0. 136 Econometric Measurement of the Effects of Research (3.34a) where h _(_ olne, dInt,] _ dlnQ, C _ dIne, , - oInt, dt ' q, - dt ' , - dt and (3.34b) dIn W, dIn Wi,t W,=~"'1:iSi,'~ Thus, the dual rate of technological change, h" may be computed as a scale effect (actually the rate of growth of output over time, q" weighted by the elasticity of cost with respect to output, Ec,Q = olnC/olnQ,) plus a Divisia index of the rate of growth of factor prices, W,' minus the rate of growth of total costs, Ct' To establish tile relationship between the dual and primal rates of techno­ logical change, the first step is to logarithmically differentiate total cost (e, = W,'X,) with respect to time so that dIne, m dIn Wi" m dlnXi" dt = 1: Si" ~ + 1: Si" dt (3.35) i=1 i=1 Combining equations 3.30 and 3.34 allows us to respecify 3.35 as h - - olnc(.) dInt _ olne, (olnQ, dInt, ]_ E , - oInt, dt - olnQ, oInt, dt - C,Q g, (3.36) Primal and dual rates of technical change are equal (but opposite in sign) if an!=l only if the elasticity of cost with respect to output equals one (i.e., Ec,Q = 1.0), or in other words, if the technology exhibits constant returns to scale. But constant returns to scale was required to ensure that the primal rate of technological change, g" equalled the rate of change of total factor produc­ tivity, tfp. Under this assumption, the dual rate of technological change, h" is also equal to tfp.45 Because of scale effects, direct estimates of the primal versus dual rates of technical progress from 3.30 and 3.34, respectively, generally differ. As Antle and McGuckin (1993, p. 182) observe, ''this happens because the primal rate is computed with input levels that are held constant, whereas the dual rate is computed with input levels adjusting optimally to technological change." 45. According to Berndt and Khaled (1 fJ79) it was Ohta (1994) who first demonstrated that the primal rate of technological change is equal to the dual rate of technological change times the dual rate of returns to scale. Econometric Measurement of the Effects of Research 137 Factor Bias and Scale Effects The focus of the discussion so far has been on measuring both the levels and rates of growth of total factor productivity and the link between changes in productivity and shifts in production and cost functions over time, com­ monly called technological change. But technological change (whether in­ duced by research or other factors) can have differential effects on the productivity and, hence, utilization of specific inputs in a multiple-input production process. And these effects may be interesting and of relevance to policy-making. Indeed, one of the main reasons for constructing PFP mea­ sures is to assess the impact of technical change on the productivity of specific factors of production. An increase in labor productivity, for instance, gives some (upper-bound) indication of the increased returns to labor. Unfortunately, the extent _to which (research-induced) technical change con­ tributes to measured gains in PFP is not readily apparent because, as we have discussed, changes in the PFP of a particular input can occur simply in response to the increased use of other inputs, even in the absence of the substitution consequences of technical change. This is so even when care is taken to form the quantity aggregates used to calculate a PFP in a way that abstracts from the substitution consequences of relative price changes.46 Homothetic technology: To address these limitations, various measures of technological bias have been proposed that utilize some notion offa ctor­ neutral or factor-biased technical change.47 This aspect of technical change was initially discussed by Hicks (1948), who defined bias in terms of the impact of technical change on the ratio of the marginal products of the factors of production or, equivalently, the marginal rate of technical substitution (MRTS) between two factors of production.48 From this primal perspective, neutral technical change involves a shift in the production function that leaves the MRTS unchanged (at, say, a particular point in input space - which is equivalent to holding total costs of production constant); factor-bi­ ased technical change causes the MRTS to change.49 46. In practice, there is also usually an aggregation problem to confront with the Xi variable, which is often an aggregate across different classes of X;. For instance, when Xi represents total labor input in agriculture, this entails aggregating across hired workers, unpaid family workers, and farm operators, all of whom embody different degrees of human capital. . 47. Technological change can affect the optimal quantity and mix of outputs.as well as inputs, but the related concepts of output-neutral and output-biased technical change are not dealt with here. For a discussion of this topic, see Hullen (1978) and Antle and Capalbo (1988). 48. For a more complete development of the factor-bias aspects of technical change, see Binswanger (l974b, 1975, 1977) Blackorby, Lovell and Thursby (1976), Stevenson (1980), Antle (1984), and Antle and Capalbo (1988). 49. Technical change causes isoquants to shift toward the origin in a parallel (i.e., factor-neuttaI) way or to "twist" or, equivalently, to shift toward the origin in a nonparallel (i.e., factor-biased) fashion. 138 Econometric Measurement of the Effects of Research Figure 3.4 illustrates the impact of technical change on a homothetic tech­ nology (e.g., a Cobb-Douglas technology) where the expansion paths, e, are linear and the optimal factor proportions (and thus optimal factor cost shares, given fixed prices) are invariant to the scale of production. If the quantity of output were held constant at Qo, the technical change from 'to to 't l would cause cost-minimizing producers facing constant factor prices to adjust their input mix and produce at point b instead of point a on the invariant expansion path e (figure 3.4). This is a Hicks-neutral change in technology because the MRTS between XI and X2 remains constant for optimizing producers. This is so irrespective of whether output is held constant at Qo (so that technical change enables the same quantity of output to be produced for lower total cost, i.e., C I < Co) or if more output is produced with the same total cost, in which case the quantities of inputs are held constant at point a and the isoquant Qo('to) is relabelled as Q*('tl)' where Qo('to) < Q*('tI)'SO In contrast, the technical change from 'to to 't2 causes the MRTS (i.e., the slope of the isoquant Qo['to]) at point a to change and so is biased in the Hicksian sense. Optimizing producers facing constant factor prices would produce the same output quantity, Qo, at point c instead of a and would thereby lower their Figure 3.4: Biased technical change, homothetic technologies o so. This discussion refers to technical progress in which the change in technology either increases output for given inputs or reduces costs for given outputs; but the same argwnents apply equally well to a technological regression or immiserizing technical change. Econometric Measurement of the Effects of Research 139 costs of production from Co to C,. Similarly, if the total costs of production were held constant at Co, optimizing producers would alter their factor mix and produce Q,('t2) at point d instead of Qotto) at point a. Technical change from 'to to 't2 induces a shift in the input ratio X/X2 (or, equivalently, relative factor shares, P,X/P2X2) that entails less of input X, to be use~ relative to X2 ; in this sense, the technical change is factor-X2 using and factor-X, saving. For homothetic technologies such as this, the direction and magnitude of the bias, as indicated by the change in the input ratio or the change in relative_ factor shares, is the same whether it is evaluated holding total costs constant (i.e., the shift from point a to point d) or the quantity of output constant (i.e., the shift from point a to point c) because the input ratio is independent of the scale of output. Nonhomothetic technologies:s, The Hicksian notion of bias, measured in terms of changes in the MRTS, does not readily carry over to the case of nonhomothetic technologies, which have nonlinear expansion paths along which the MRTS changes with the scale of production. Hence, the factor bias measured in terms of observed changes in input ratios will be sensitive to the pre- and post-technical change scale of production, and it matters whether the factor bias is measured holding total costs or the quantity of output or something else constant. Figure 3.5 illustrates the difficulties. In panel a of figure 3.5, a change in technology from 'to to 't, enables optimizing producers to economize on inputs X, and X2 and to produce the same quantity of output, Qo, at lower {:ost (i.e., total costs are reduced in going from 'to to 't,) at point b rather than point a. Because the MRTS between X, and X2 is unaffected, the technological change is Hicks-neutral (or, by Blackorby, Lovell and Thursby's (1976) interpretation of Hicks neutrality, the technical change is expansion-path preserving) even though the input ratios and corresponding factor shares have changed. In this case, the overall bias effect of the technical change, wherein X/X2 has decreased so that the technical change is factor-X2 using and factor-X, saving, is entirely due to a scale effect. In panel b, the move from a to c also results in the same quantity of output being produced at a lower cost (i.e., C, < Co) and with a different factor intensity (i.e., X/X2 has again decreased). But in contrastto the case in panel a, the overall effect of factor bias can be decomposed into a Hicksian effect (i.e., a move from point a to point b, holding the total cost of production constant) and a scale effect (i.e., a move from b to c). Alternatively, the overall effect can be partitioned into a scale effect involving a move from a to d and a Hicksian effect SI. The discussion in this section and the next draws heavily on Antle and Capalbo (1988) and Karagiannis and Furtan (1993). 140 Econometric Measurement oft he Effects ofR esearch Figure 3.5: Biased technical change, nonhomothetic technologies a) Xl o b) c) Xl o Econometric Measurement of the Effects ofR esearch 141 represented by a move from d to c, wherein the quantity of output is held constant at Qo. In this case, the scale and Hicksian bias effects of the technical change are in the same direction; they are both factor-X2 using and factor-Xl saving. The technical change in panel c of figure 3.5 involves an overall move from a to c that can be decomposed into a Hicksian effect, represented by a move from a to b (or, alternatively, from d to c) and a scale effect, involving a move from bto c (or, alternatively, from a to d). In this case, the scale effect is factor-X2 using but is dominated by the Hicksian effect, which moves in the opposite direction and is factor-Xl using, causing the overall bias effect to be factor-Xl using. Measuring bias: Both pairwise and overall measures of input bias are pOssible and commonly calculated. Following Stevenson (1980), Antle and Capalbo (1988), and Karagiannis and Furtan (1993), it is useful to categorize the factor-bias effects of technical change in terms of its effects on factor shares.52 The overall bias relates to technology-induced changes in the cost share of a particular factor relative to the shares of all the other factors of production; pairwise bias concerns the change in the cost share of a particular factor relative to the cost share of another particular factor. From this factor-cost perspective, pairwise bias for any pair of inputs can be formally defined as _ alnS; (Q , W, t) alnSj (Q, W, t) . . Bij (Q , W, t) - at - at ' , '¢ J (3.37) where Sj = alne/aln W; is the cost share of factor i. Pairwise comparisons of all inputs can be made in this fashion, with the technology being classified as factor-i using relative to factor j if Bjj > 0 and vice versa. But there is obviously a difficulty of interpretation when technical change uses input i relative to inputj but saves input i relative to some other input, k; it is unclear whether the technical change is saving or using factor i overall. To assess this aspect of technical change, it is useful, following Binswanger (1974 a, 1974b), to define the overall bias effect as m alnSlQ , W, t) B;(Q , W , t) =~ Sj (Q , W, t) Bij (Q , W, t) = :"l (3.38a) j'14 at where Sj is the cost share and Bij is the pairwise measure of bias given by 52. To complement this dual perspective on bias, it is also possible to calculate primal measures of factor bias in tenns of the effects of technical change on the MRI'S or, equivalently, relative marginal physical products. For a good discussion of these primal ~ures of factor bias, see Antle and Capalbo (1988) and Antle and McGuckin (1993). 142 Econometric Measurement of the Effects of Research 3.37.53 If B; > 0, then the technical change is factor-i using overall, and vice versa; B; =0 for all i is necessary (but not sufficient) for Hicks neutrality. To assess the degree of overall Hicks biasedness requires decomposing B; into its Hicks (i.e., H;[Q, W, t]) and scale (i.e., A;[Q, W, t]) components. Antle and Capalbo (1988) show that B j (Q , W, t) = H j (Q , W, t) + Aj (Q , W, t) (3.38b) where the Hicksian bias in technical change is measured as a shift in the expansion path, holding the total cost of production constant so that alnSj (Q , W , t) H j (Q , W , t) = at I dC =0 (3.39) and the scale effect of biased technical change due to the movement along the expansion path is given by aalnC (alnC)-1 alnSj alnC ( 1 ) aSj Ai (Q , W , t) = :t alnQ alnQ = a:t Sj EC,Q alnQ (3.40) where EC,Q =a lnClalnQ is the elasticity of cost with respect to output. Notice that if the technology is homothetic, alnS/alnQ =0 so that the scale effect is zero (i.e., A; = 0) and the overall bias effect equals the Hicksian bias effect. It is also useful to recall that the Hicksian bias effect and the scale effect do not necessarily run in the same direction so that both the overall direction of bias and the magnitude of bias depend on both the directions and relative magnitudes of the two effects. 3.2 Specification and Measurement Issues Apart from the choice of general approach (i.e., primal versus dual or supply-function), a number of other specification and measurement issues arise in any econometric study of research benefits. Issues of the choice of variables to include, proxy measures for those variables, the representation of technical change, and the functional form for the empirical model are all intimately related and are best resolved jointly. 53. Time, t, is commonly used as the index of technology, 't, so that the measure of bias in technical change is also a measure of the factor share's technology-induced growth rate over time, ceteris paribus. Otherwise, to achieve the same result, it might be necessary to scale the measures of bias by the rate of change ofthe technology index, dtIdt. Econometric Measurement of the Effects of Research 143 3.2.1 Primal Models Choices about functional form are dictated to a great extent by the availability of data and the purpose of the analysis. It is desirable to choose a specific functional form consistent with maintained hypotheses (such as positi ve marginal products that decline over the relevant range), which does not impose a priori the degree of economies or diseconomies of scale and which allows for both complementary and substitute relationships among inputs. These considerations favor the use of flexible functional forms. 54 Simplicity for its own sake and computational ease may also be relevant considerations. Complicated functions may contain implausible implications that are hard to detect, and they may make it difficult to compute economic effects or relationships such as the elasticity of substitution. By far the dominant criterion, however, has been data availability. This means that relatively simple and parametrically parsimonious models, which are tech­ nically and empirically undemanding, have been chosen most often. The Cobb-Douglas Production Function The Cobb-Douglas production function has been the basis for most econometric studies of agricultural research benefits. This function is written in standard, geometric form as m k z Q, = 50 fIXfj fI~~ fI.t,r,? ell, (3.41) i=l g=l h=l and for econometric estimation, it is transformed to a function that is linear in the logarithms of the variables: m k z InQ, = iQ In Wi" In Q, + I ~gQ In Kg" In Q, i=1 g=1 m k + I I 'Yig In Wi" In Kg" i=1 g=1 where C, is the cost of producing output Q, in period t; all other variables are defined as before. This function can be regarded as a quadratic approximation to the unknown, ''true'' cost function when iQ =0 for all i ; (d) I'Yig =0 for all i j=1 g=1 This cost function does not constrain the structure of production to be homothetic, nor does it impose restrictions on the elasticities of substitution or economies of scale. A system of m, output-constant demand functions for the factors of produc- 61. Shumway, Saez and Gottret (1988) and Ball (1988) estimated multioutput versions of this type of model for U.S. agriculture, while Zhang et aI. (1993) did likewise for Indonesian rice and soybean production. Econometric Measurement of the Effects of Research 149 tion can be derived from the factor-share equations. These input (cost) share equations are defined using the logarithmic form of Shephard's lemma: Xi,tWi,t alnCt Si,t =- C- = aln W- t I,t (3.48) m k = ai + iQ InQt + I:.aih In Wh,t + I:. rig InKg,t h=1 g=1 where Si.t is the cost share of the ith input in period t. Conventionally for estimation, it is assumed that the error terms in the cost function (3.46) and the associated cost-share equations (3.48) have inter­ temporally independent distributions with zero mean and nonzero contem­ poraneous covariance. In this framework, it is not sufficient to estimate the m-share equations (3.48) since the parameters needed to derive the impact of knowledge-related variables such as research (i.e., the 00, Pg, PgQ, and Pgh parameters) appear only in the cost function. The cost function and the corresponding share equations can be ~stimated jointly as a system of equations using maximum-likelihood methods, with cross-equation restric­ tions imposed on the parameters that appear in more than one equation. The benefit from using this systems approach is that as long as the model is correctly specified, the imposition of these behavioral assumptions through cross-equation restrictions on the parameters means that the parameters are estimated with greater precision than if only the cost function were used.62 Because cost shares must all sum to one, an input-share equation is dropped from estimation to avoid singularity ofthe covariance matrix in the actual estimation. Barten (1969) showed that estimation is invariant with respect to the equation dropped if maximum-likelihood estimation methods are used.63 The parameters of the deleted cost-share equation are easily recovered using the maintained hypotheses of symmetry and linear homoge- 62. Typically, in practice, it would appear that little serious thought is given to where the error terms come from or how they ought to be specified. It seems they are simply tacked on to a static model to expedite statistical estimation. As Berndt (1991, p. 471) notes, the usual rationale for including additive error terms in the individual factor-demand (or -share) equations is that "firms make random errors in choosing cost-minimizing bundles." He also cites McElroy (1987), who proposes an alternative view in which the "errors" are in the eyes of the econometrician (i.e., specification or measurement errors rather than optimization errors). Given the direct relationship between the share equations and the cost function, it seems inappropriate to simply add random terms to the share equations without paying specific attention to the corresponding modifications in the cost function, itself, when the full system is being estimated. The appropriate modification to the cost function may depend on the rationale being offered. 63. ln practice an iterative version ofZ ellner's seemingly unrelated regression procedure is often used to estimate the set of cost- and input-share equations. While the estimates obtained using Zellner's procedure are sensitive to the equation dropped, Kmenta and Gilbert (1986) used Monte Carlo methods 150 Econometric Measurement of the Effects of Research neity (3.47) in input prices. Allen partial elasticities of input substitution can be computed from the cost function using C t Ci,h,t (J 'ht = -----=--'- (3.49) I Ci,t Ch,t where Cj and Ch represent partial derivatives of cost with respect to inputs i and h, respectively, and (Jjhl is calculated using the available data (i.e., the predicted values of the factor shares) and parameter estimates as (Uzawa 1964; Binswanger 1974a) 2 a'h fall *' h a,,+S't-S't I + s:-I s ,and II I, I, or i (JUt = S~ (3.50) ~ ~ ~ The (output-constrained) own- and cross-price elasticities of factor demand can be obtained by multiplying the Allen elasticities by the respective (predicted) factor shares so that lliht = (Jiht Si,t for all i and h (3.51) Notice that these elasticities of factor substitution and factor demand can change over time as factor shares change, but they are often calculated at the sample means ofthe data.64 3.2.3 Single-Equation Supply Models Supply functions have been estimated for research evaluation in several recent studies. Otto (1981) estimated supply functions for corn, wheat, soybeans, and sorghum in the United States including a variable for public agricultural research in each equation. The estimated coefficients were used to calculate internal rates of return to research. Zentner (1982, 1985) esti­ mated supply equations for Canadian wheat and rapeseed including a pub­ lic-research variable. He then used the shifts in the supply equations, implied by the estimated research coefficients, to calculate the change in economic and Ruble (1968) used fonnal proof to show that an iteration of this procedure will converge to results of maximum likelihood and thereby the invariance property of Barten holds in this case, 64, Mundlak (1968) and Blackoroy and Russell (1989) discuss the alternative elasticity-of-substitu­ tion measures, such as that developed by Morishima wherein olI:t = Sh,l «(Jiltl - (JMI) = 'I1iltl - 'I1hht, Allen's elasticity of substitution corresponds to measuring how one input adjusts to a factor price change, assuming constant output, and the Morishima measure corresponds to measuring the responsiveness of two factors to a change in one price, assuming constant costs, Two factors classified as substitutes according to Allen are also substitutes according to Morishima, but not the converse, Moreover the Morishima measure is not symmetric whereas the Allen measure is, Econometric Measurement of the Effects of Research 151 surplus due to research. Finally, he used the surplus changes to calculate rates of return to research. A similar approach was used by Fox, Brinkman and Brown-Andison (1987) to calculate the rates of return to swine, beef, egg, dairy, broiler, and sheep research in Canada. They also included government policy variables in the econometric model and sought to account for the effects of trade and government policy on the rates of return to research.6s The supply function used to estimate the effects of agricultural research can be represented in general form by Q = q (P , W , t, Z) (3.52) where Q is quantity of the commodity produced, P is expected output price, W is a vector of expected input prices, t represents technology-related variables such as research and extension expenditures, and Z is a vector of other supply-shift variables. A number of conceptual and practical problems may be encountered in making the transition from this general form to a specific, empirical supply function. Several of the conceptual issues associated with specifying the supply function in equation (3.52) for use in research priority setting are similar to those discussed for estimating a production function. The choice of a func­ tional form often imposes a particular type of supply-function shift when a research variable is included (e.g., parallel or pivotal). As discussed in chapter 2 and below in chapter 4, the nature of the supply shift from agricultural research is difficult to know a priori and yet the type allowed has an important effect on the calculated research benefits. Fox, Brinkman and Brown-Andison (1987) used a Box-Cox transformation together with an assessment of the signs and significance of individual coeffi­ cients to choose the functional form for the supply equation and, at the same time, to choose the form of the research-induced supply shift. For some commodities, they chose a partial logarithmic function and in others, a linear function. The partial logarithmic form imposed a pivotal proportional shift while the linear form imposed a parallel shift. Several cautions are in order with respect to tests for choosing a functional form and the implications of those tests for the nature of the research-induced supply shift. First. unless sufficient data are available, including data near the price axis, the results of the tests should be viewed as providing at best only an indication of the true functional form. The relationships that seem to hold at the point of approximation may not hold at all when one extrapolates away to the price axis, as is necessary in research evaluation. Further, the tests that were applied are adequate only for comparing alternatives within the Box-Cox class but not for a more general set 65. Details on using the economic swplus approach for evaluating agricultural research, with or without market -distorting policies, are provided in chapters 4 and 5. 152 Econometric Measurement of the Effects of Research of alternatives. Finally, and for similar reasons, it would be incorrect to assume that these tests have also provided a convincing test of the nature of the supply shift, even though a particular type of shift was selected.66 In addition to the specification issues that arise in applying the primal and dual methods discussed above, there are others that are specific to the supply­ function approach. They arise from the flexibility of the ad hoc single-equa­ tion-modeling approach that is both its greatest strength and greatest drawback. In the primal and dual approaches discussed above, no attention was paid to the dynamics of responses over time, expectations, or uncertainty, which we know are important in agricultural decisions. Those aspects are less relevant in the primal approaches, but they are often ignored in the duality­ based models simply for convenience. As discussed by Cassels (1933), Col­ man (1983), and Just (1991, 1993) in the context of supply models, these aspects of specification choices (along with risk and government policy vari­ abIes) are important and difficult and have occupied much of the literature. Expected prices rather than actual prices are typically used for the output price variables and sometimes for input prices. Adaptive expectations, ratio­ nal expectations, or futures prices are often used to represent expected output prices. Because of its simplicity, the adaptive-expectations model has been widely used; the lagged values of the dependent variable and the output price are included in the group of right-hand-side variables.67 Prices of substitutes and inputs, as well as other shift factors, are included to enable the estimation to identify a supply function rather than a demand function or a combination of the two. Research, extension, and education variables can be defined just as they were for the production function described earlier. Weather variables, relevant infrastructural variables, and government policy variables may also be included as shift-type variables in the model. Another conceptual problem with specifying and estimating commodity supply functions concerns the nature of supply dynamics. One example arises in the use of time series of annual observations for commodities for which biological cycles are not annual. For example, poultry supply response may be modeled more accurately with quarterly observations (e.g., Chavas and Johnson 1982). Alternatively, some livestock industries and perennial crops involve production "cycles" of much longer than one year, and differ­ ent explicit treatments to handle the dynamics of supply response to prices 66. The evidence of goodness of fit provided by the functional-fonn tests is based on more than the research shifter and, indeed, may be based on very few data points near the axis. The latter is a concern because a functional form that provides a good local approximation may not globally satisfy regularity conditions. 67. Nerlove (1956) discusses the adaptive-expectations model. FISher (1982) discusses rational expectation fonnulations. Gardner (1976) discusses futures prices as expectation variables. Econometric Measurement of the Effects of Research 153 are indicated.68 These issues of dynamics are also present, albeit slightly less obviously, in the supply-response functions of annual crops such as wheat because the dynamics relate to the use of durable specialized capital (e.g., see Burt and Worthington 1988). 3.2.4 Output and Input Data Translating available statistics into meaningful measures of the input and output data required to study production processes, and the impact of R&D on those production processes, is often a challenging task and one that requires considerable care.69 Aside from fundamental constraints on the statistics themselves in terms of their quantity and quality, analysts must make numerous decisions when processing and transforming the data, and these decisions can have significant implications for the eventual results. This section and the one to follow discuss these data issues, paying particular attention to the problems that arise when the available statistics are sparse and fall short of what is required to produce precise measures of the variables they are meant to represent. Output Quantities and Prices Quantities: Output quantity measures are required to estimate production yield, supply, and cost functions. For analyses involving only a single, homogeneous commodity, an output variable measured in quantity terms is relati VelY straightforward to construct. For any work using aggregate output, some index or quantity aggregate must be formed. Agricultural data are usually reported on a seasonal or annual basis. When secondary data sources compiled by local statistical agencies or by interna­ tional agencies such as the FAO or the World Bank are used, care must be taken to ensure that the reported output measure represents actual quantities pro­ duced in a given year or season and not quantities available. This is of particular concern when the output variable is derived from farm sales or marketing data that incorporate stock carryovers from one period to the next. A more serious concern in developing countries characterized by subsistence agriculture is that sales or marketing data exclude quantities consumed in the household where they are produced. For some commodities and countries, 68. For livestock, see Jarvis (1974); for perennial crops see French and Matthews (1974), Alston, Freebairn and Quilkey (1980), and Dorfman and Heien (1989). 69. This point has been stressed repeatedly by Griliches (1960, 1986b, 1994) and Gardner (1992), among others. For additional discussions of the measurement issues dealt with in this section, see TImmer (1970), Kennedy and Thirlwall (1972), Yotopoulos and Nugent (1976), AAEA (1980), Ball (1985), Shumway (1988), and Craig, Pardey and Roseboom (1994). 154 Econometric Measurement of the Effects of Research these quantities can represent more than half the production. Statistical agen­ cies in many countries attempt to estimate the amount of consumption of products produced by farm households, although these estimates are frequently fairly crude. It may be necessary for analysts to derive their own estimates of home consumption in order to generate plausible output measures. As described in section 3.1.3, additional aggregation problems arise when dealing with output aggregates such as "grain crops," "livestock products," or "total agricultural output." In these cases, some notion of a real quantity index of aggregate output is required. Unweighted and perhaps undesirable pre­ aggregation may be implicit in the data-collection process even for single-com­ modity measures of output like tons of wheat, barley, or rice. A simple addition of high-protein (baking-quality) wheat and feed-grade wheat or malting- and feed-grade barley or an unweighted sum across various qualities of rice can result in distorted real output measures when the mix of commodities compris­ ing the aggregrate varies over time. The potential for bias is particularly serious if the relative prices of the various commodity classes vary much across observations.7o To form a real output aggregate in such cases requires data on the quantities and corresponding prices for each component of the aggregate and the construction of a Divisia output-quantity index. This aggregate is real in the sense that it abstracts from changes in measured output that are due solely to substitution effects as producers respond to changes in the relative prices of the components of the aggregate. In other words, a real output aggregate attempts to distinguish between changes in the size of the real commodity basket and changes in the composition of the basket. When constructing an output index, analysts need to ensure that the form and units of the quantity data match the corresponding price data and that these aspects of the data are held constant over time. Sometimes a quantity series, say for rice, will be reported in units of "dry-stalk paddy before milling," while the corresponding output price series is reported in "milled-rice" terms. Appro­ priate conversion factors are generally available from local statistical agencies. Unfortunately, often only preaggregated data on the value of production are available. If a corresponding Divisia price index were also available, then an implicit Divisia quantity index could be recovered from the value aggregate. In practice, Divisia price indexes are usually unavailable and some fixed-weight (Laspeyres) price index must be used to deflate the value aggregate. The longer 70. A related problem is when an output measure such as tons of sorghum fails to include by-products such as crop residues that are used for feeding, housing, and so 00. Overlooking by-products could be an especially telling omission in the present context if the principal objective of a crop breeding program were to increase the quantity and quality of crop residues rather than just increase grain yields per se. Similarly, it may be important to incorporate manure in, say, a sheep output aggregate that also includes meat and wool. Econometric Measurement of the Effects of Research 155 the time horizon of the study, the more likely are fixed-weight indexes of output prices to understate the rate of change in output prices by failing to account for substitution effects. Consequently, using them to deflate output values will overstate the real changes in output. In choosing a price deflator, one should use the price index that most nearly reflects the composition of the aggregate value to be deflated.71 To the extent that the basket of goods represented by general price indexes differs from the aggregate of interest, their use will introduce additional sources of bias into the analysis. Prices: In addition to their use in forming real output-quantity aggregates, output prices also enter into supply, cost, and profit functions either directly or in the form of expectations. The issues of measurement and aggregation discussed above for output quantity variables carry over to the construction of output price variables. The only substantive difference is that appropriate quantity measures are used as weights in forming direct (Divisia) price aggregates. If fixed-weight quantity indexes must be used to deflate nominal value aggregates to derive implicit fixed-weight price indexes, the likely bias is to overstate price increases. This method should only be used if data limitations make that the only feasible option. When expected prices are used instead of actual prices, the analyst must choose how to model expectations. That choice will have unpredictable im­ plications for any subsequent use of the estimated model in evaluating the welfare impact of technological change. For the most part, however, the choice of expectation model can appropriately be based solely on econometric con­ siderations with the aim of obtaining consistent parameter estimates, leaving the issue of interpretation to subsequent stages. One complication that ought to be anticipated is that expectation structures, such as adaptive expectations, that involve the use of lagged dependent variables involve dynamic supply re­ sponses over future time periods to a price change in the current period. These dynamics will complicate the estimation of research benefits in a model with downward sloping demand and, therefore, endogenous prices. In such a model, a one-shot research-induced supply shift might involve dynamic adjustments in quantities and prices over an indefinitely long future. An even more com­ plicated picture emerges when the dynamic exogenous impact of research (through research lags, the adoption curve, and so on) generate endogenous dynamic responses.72 71. See Pardey, Roseboom and Craig (1992) for more discussion on this matter. 72. Care is required in the construction of such models. Indeed, some such models (such as the adaptive-expectations model) can involve dynamics applying symmetrically across the explanatory variables in the model so that the dynamics of output response to research (and other explanatory variables) would be constrained to be of the same form as the response to price changes. 156 Econometric Measurement oft he Effects ofR esearch Input Quantities and Prices Labor: Although labor is one of the key inputs into agricultural production, it is notoriously difficult to obtain accurate measures of the labor actually used and the price paid for that labor. A large proportion of agricultural labor consists of operator and family labor for which observable market transactions are sparse. It is often difficult to establish the extent to which farmers and their families work full or part time at farming. Typically, the development process is characterized by a substantial but not necessarily uniform increase in the rate of part-time farming in addition to an eventual decline in the number of farms. Even in the case of hired labor, complicated sharecropping arrangements and remuneration contracts, particularly in less-developed countries, often make it difficult to price the hired-labor component. In addition to data issues, there are the familiar conceptual and measurement problems associated with accounting for variations in quality (in this case, quality of the human agent) over space and time. A difference from other inputs and outputs is that, with labor, some of the input is provided by the individuals whose decisions and actions are being analyzed, and this may add some special problems of measurement and interpretation of prices and quantities. Further issues arise in interpreting information about labor markets, given the dynam­ ics of education and human capital formation and their interplay with agricul­ tural R&D (e.g., Welch 1970; Schultz 1975). Of course, if complete data were available on the prices and quantities of different categories of labor over time, the ideal index-number procedures described in this chapter could be applied in a routine fashion to obtain quantity and price indexes for the aggregate labor input. But such data never exist. The issues, then, concern appropriate approx­ imation procedures to be used in dealing with incomplete data. Typically, appropriate measures of wages, or the opportunity cost of time, are not readily available for any type of labor used in agriCUlture, especially unpaid operator and family labor, which in developing countries, constitutes most of the labor input, indeed of all inputs. When market prices are not available, they can sometimes be imputed using information from parallel labor markets. For instance, Craig and Pardey (199O) used data on the earnings of rural workers, differentiated by classes of age and education, as a measure of the opportunity cost of time for farm operators in the same classes. Then, using information on the characteristics of farm operators, they were able to construct state-level indexes of the aggregate quantity and price of labor used in u.S. agriculture. An alternative approach is to use a hedonic approach wherein a wage function is estimated by regressing wages against human-cap­ ital variables (e.g., age, education, and experience), other demographic vari­ ables (e.g., gender and ethnicity), and employment characteristics (e.g., full- Econometric Measurement of the Effects of Research 157 versus part-time employment, operators versus hired or unpaid family), and then using the estimated model to predict the wages of farm operators in various demographic classes and by region prior to aggregation. Capital: Capital stocks and their service flows pose a number of critical and difficult measurement problems in studies of production and productivity.13 As development proceeds, durable inputs typically account for an increasingly large share of total inputs.74 They also embody substantial (often privately funded) R&D output, and for these reasons, their accurate measurement and treatment are of particular importance when the productivity effects of publicly funded R&D are being assessed. The durable factors of production that are often treated as capital goods include physical inputs such as tractors, trucks, auto­ mobiles, combines, forage harvesting equipment, farm buildings, the farm plant, and other equipment, as well as biological inputs that are used for periods exceeding the frequency of the output measures. When annual data are used, this biological capital would certainly include breeding stock for cattle, sheep, pigs, goats, and chickens; milking stock such as dairy cattle and goats; and animals used for traction such as horses, buffaloes, and mules.75 The data commonly available are stocks of capital that are (a) either unweighted sums or value-weighted sums across different classes and/or (b) heterogeneous types of capital such as tractors of different horsepower ratings, combines of different widths or capacities, and trucks of different sizes. The service flow from capital needed for production studies or the rental rate needed for cost studies can only be inferred from information on capital stocks and their values. So measurement of service flows typically involves assumptions about the time path of the marginal physical product of capital, the relationship between the physical product and its market value, and the age and quality composition of the existing capital stock. In competitive capital markets, the value of a unit of capital is equal to its expected, discounted real service flow. To aggregate distinct types of capital or infer actual service flows from the value of capital stocks, several character- 73. See Griliches (1960, 1963c), Yotopoulos (1967), and Jorgenson (1974). Craig, Pardey and Deininger (1993) give a much more detailed treatment of the issues dealt with briefly here, including coping with data limitations, forming informed guesses about key parameters, and handling quality change. For a discussion of emodical technical change cum quality-of- capital issues, see also Jorgenson (1966) and Hulten (1992). 74. For example, in 1985 capital services accounted for around 14.4% of the total production costs of u.s. agriculture (Craig, Pardeyand Deininger 1993). Although comparable data are difficult to obtain, in many LDCs, the capital share is substantially lower. 75. Those biological factors of production with service lives of less than one year do not need to be treated as capital inputs. They are more appropriately viewed as outside inputs, as defined by Star (1974), to the extent that they are produced outside the current production period and carried over to the following (but not any future) production period. An appropriate way to deal with these factors is to define the corresponding output measure in net terms and also to exclude them from the measured inputs. 158 Econometric Measurement of the Effects of Research istics of the stock and flow relationship must be understood: (a) the expected lifespan of a machine (in order to incorporate likely exhaustion of a particular type of capital), (b) the pattern of physical deterioration (to incorporate eco­ nomically meaningful measures of deterioration), and (c) the impact of quality differences or obsolesence on the market value of preexisting capital stock. The assumption of competitive capital markets implies that the purchase price of a unit of capital at time t, P"~ is equal to the expected, discounted flow of current and future real rents, PI' from that same unit of capital over its service life of length L, i.e., PI = E (PI + P;1 + ;;2 + ... +L _;I+L 1 (3.53a) I t 1+1 IT D +k k=O ' where the discount factor is D, =( I + r t), and r t is the rate of discount in period t. The discount rate represents the opportunity cost of funds, and it is typically assumed to be constant over time - i.e., rt+k = rt = r for all k.76 Then the formula reduces to L PI = L Pr+k (3.53b) k=O (I+d To make use of this relationship, we must make further assumptions about its parameters. The real rent in period t corresponds to the real service flow from capital in period t. The relationship between values and service flows is based on expectations about the nature of current and future service flows, so purchasers must take expected usage into account. In practice, planned usage is unknown, so analysts typically assume that market data reflect the relationship between values and service flows under planned normal use and maintenance. Two common assumptions about the likely profile of service flows from capital are the lightbulb (or one-hoss-shay) assumption and the declining-bal­ ance assumption. Under the lightbulb assumption, with normal use, the real service flow from capital is expected to remain constant throughout the capital's service life. When its service life is over, the flow stops. With a declining-balance assumption, normal use results in a linear or geometric decay 76. In practice a long-run, real interest rate for a portfolio of government and private bonds outside the agricultural sector is an appropriate discount rate. For the United States, this rate has historically averaged around 4%. Using national account data to calculate a residual return on capital measure for the economy as a whole (e.g., Jorgenson and Griliches 1967) or the agriculturaI sector (e.g., Ball 1985) may be inappropriate because it includes the returns to a number of factors that are being estimated rere - in particular, research. Econometric Measurement of the Effects of Research 159 in real service flow over time at a constant rate, O. Service life is assumed to end when current real service flow drops below a critical threshold. Each assumption about the service-flow profile implies a distinctive pattern of changing market valuation over the service life of the unit of capital. Craig, Pardey and Deininger (1993) show that the value of a machine or animal in year k = I, ... , L of its service life is given by _ Pk Pk - (3.54) /..{L,k,r,o) Under the lightbulb assumption, Pk = P (so by construction, the rate of deterioration, 0, equals zero) for all years k = 1, ... , L of a machine's working life, and Pk =( P +.2.D +-u&- + ... + d-gl-kJ = '}.. (LP,k ,r) (3.55a) where l '}..(L,k,r) = l:r [1_D-r for r > 0 and (3.55b) '}..(L,k,r) = (L-kr l for r = 0 This service-profile assumption implies that the time path of the real value of a unit of capital is concave to the origin for positive interest rates and declines linearly when there is no discounting of future service flows. Under a declining-balance assumption with a geometric rate of deprecia­ tion,o, (3.56a) Pk - '}..(L,k,r,o) where ).(L,k,r,S) = ~[ 1 - (::y ·. .+ f (3.56b) which collapses to Pk = p';'}..(r,o) where '}..(r,o) = (r+O)/(l+r) for a unit of capital with an infinite life span so that the market value of the capital declines over time at exactly the same rate as the service flow. For those units 160 Econometric Measurement of the Effects of Research of capital with a finite service life, the time path of the real value of capital with a decaying service flow will be convex to the origin even when there is no discounting of future service flows. Inferring real service flows from the market value of a single unit of capital is straightforward once the service-flow profile is parameterized. The current real service flow from a unit of capital that is k years old is a particular fraction, A(L, k, r, 0), of its current market value. This factor ofp roportion­ ality is described in equations 3.55b and 3.56b, above. It is more complicated to infer service flows from stocks that contain different classes of capital, more than one machine type within a capital class, or more than one vintage of machines of a given type. For the purposes of this discussion, a capital class consists of machines that have identical service-flow profiles, i.e., identical depreciation rates and life spans, 0 and L. Nevertheless, the real service flows will vary with age and type of machine, so it is crucial to know something about the age and quality distribution of types within a class. Capital types within a class, e.g., 50-, 100-, and 200-horsepower tractors, are not equally effective, but it may be tolerable to assume that the shapes of their service-flow profiles are identical. With complete information on the numbers, ages, and productive qualities of different types of capital, figures on the aggregate value for each class could be constructed and then simply multiplied by the appropriate factor of propor­ tionality to infer aggregate service flows for that class. However, to do so requires the solution of two distinct aggregation problems. First, capital must be aggregated over dissimilar types of machines within a class of capital, assuming identical service-flow profiles. This is fairly easy to do since the relative prices of machines of the same age will accurately reflect relative productive qualities, even if the machines embody different technologies. A more difficult problem arises in aggregating over different vintages of capital whose productive qualities differ because of decay, exhaustion, and obsolescence. In this case there are problems in inferring service flows using relative prices because the relative prices of used and new machines are not accurate reflections of the current relati ve service flow from these machines, even ifthey have the same service profile. This arises because the decline in market valuation with age captures more than just the decay in the real service flow from capital; it also reflects the time to exhaustion of the capital stock so that A(L,l,r,O) < A(L,k,r,O) for all k > 1. The appropriate measure of service flow would apply a different factor of proportionality to the value of each different age group. Unfortunately, the data commonly available concerning capital are either estimates of the aggregate market value of the stock of capital of a particular class (e.g., the value of tractors on farms) or unadjusted counts of capital Econometric Measurement of the Effects of Research 161 within a class (e.g., the number of tractors on farms). As an approximation, aggregate service flows, SF, can be inferred from aggregate market-value data, MV, by applying the factor of proportionality, 'A(L,a,r,O), appropriate for the "typical" machine. Taking the typical machine to be one whose age is the class average, a, yields N L N L SF = 1: 1:Pk,i Xk,i "" 'A(L,a,r,O) 1: 1: Pk,i Xk,i = 'A(L,a,r,O) MV (3.57) i=! k=! i=! k=! where i is an index running over the N types of capital in the class, k indexes theL possible vintages in the class, andXk,; and Pt,; are the number and market value of vintage-k, type-i machines, respectively. If the available figure for total capital stock for a class of capital is simply an unweighted sum of the units of all types and vintages, we must somehow account for the likely composition of the total in order to calculate service flows. In the absence of other information, the total stock figure may be considered to be an accurate count of the "typical" unit within each class. Here, typical is defined by the unit within the class of the most likely type, x, and age, a. Taking the reported total, ee, to be counts of typical machines, the total capital stock figure can be adjusted by the assumed rate of deterioration under normal use to reflect total effective new machines of type x, using N L PI .Xk . N L xi,x = 1: 1: (l-ol-1 ,I ,I"" (1_0)0-1 1: 1: Xk,i i=1 IFI Pl,x i=1 IFI (3.58) = (1_0)a-l ee Aggregate service flows, SF, can then be inferred from this undifferentiated stock by taking the approximate machine counts, ~,x, expressed in units of new machines of type x, and employing the market value of these new machines and their corresponding factor of proportionality. N L SF = 1: 1:Pk,i Xk,i "" 'A(L,l,r,O) p!,x~,x (3.59) i=! k=! When a quantity measure of the physical stock is desired and the only available data are the undifferentiated counts, ce, then the approximation in equation 3.58 can be used to express this stock in terms of new machines of type x. If the market value of capital in place is available, informed guesses about the age distribution must be made to derive a quantity measure. Dividing the total market rental by the rental of a new machine of type x would give exact counts of type-x machines, but we can only approximate 162 Econometric Measurement of the Effects of Research the rental value of this undifferentiated group by taking each unit within it to be of age a. Using only the factor of proportionality for new machines of type x and its market price, PI,x, the market value for the capital class, MV, can be converted to approximate counts of new machines of x, using N L N L ~ ~ A(L,k,r,O) Pk,i Xk,i xi,x =~ ~ Pk ,I· XIc, i = i_=I _k-=--l :-__- =-___ i=l k=l PI,x A(L,I,r,O)PI,x (3.60) N L A(L,a,r,O) ~ ~ Pk,i Xk,i ==_----:-_---..:...--=i-=-:l: 'ck..=..:l.. . ___ _ A(L,a,r,O) MV A(L,l,r,O) P1,x - A(L,l,r,O) P1,x An index of the real service flows from capital over time can be con­ structed by using the measures of service flow and physical stock described above to aggregate over M different classes of capital. For instance, to estimate the aggregate real service flow summed across multiple classes of capital such as tractors, combines, and trucks requires the actual or approx­ imate quantities expressed in units of the numeraire of each capital class along with its corresponding service flow. Letting AI,e =A (Le,l,rA) represent the factor of proportionality for new machines of class c and letting SFe represent the corresponding service flow, the formula for the Tornqvist­ Theil Divisia index of the aggregate quantity of capital services is a special case of equation 3.17, namely, where (3.61) As with any other Tornqvist-Theil quantity indexes, the changes in real quantities of each capital class are weighted by its share in total capital costs. The class share, in this instance, is the two-period average of its share of total Econometric Measurement of the Effects of Research 163 capital rental costs. It might be tempting to use market-value shares for weights in this index, but they differ from rental-cost shares if, as assumed here, service-life profiles differ across capital classes. Land: The problems with measuring quantities and prices of land fall into two types. First, as with other capital, the relevant measure for many purposes is an annual flow of services from the asset rather than the market price of the asset, as described above. In many cases, data on market transactions are not available for annual land rental, so rental rates must be inferred from information on asset prices. Such imputation requires informa­ tion on variables such as the discount rate, tax rates, and the expected rates of growth in rents and capital gains (e.g., Alston 1986). One virtue of land compared with other capital is that it is often reasonable to assume zero depreciation so that the problems of imputing types and rates of depreciation from sparse data may be nonexistent. However, the other variables (espe­ cially expected rates of capital appreciation over an indefinite horizon) are typically difficult to estimate, even when reliable data are available on past price movements. Second, and unlike most other capital items, the data on asset prices of land are often questionable: they are usually based on either limited numbers of observations of transactions in very thin markets or surveys of expert opinions. Additional conceptual and measurement questions arise when land is used in rotation or in other systems that persist for several years (e.g., as in perennial crops such as coffee, cocoa, or rubber) or in production systems that allow for multiple crops within a year (e.g., irrigated rice and wheat systems in Asia where two or more crops are grown in sequence or intercropping systems where two or more crops are grown simultaneously). One conceptual problem is to match quantities with correponding prices (e.g., hectare plantings per year, rather than simply hectares, to deal with multicropping). Like other capital, land is heterogeneous, and this heterogeneity leads to potential problems in constructing meaningful aggregates. The recommended general approach to dealing with land quality, as with other capital, is to disaggregate as much as practicable when developing indexes of land aggre­ gates. Relevant quality attributes for land, which affect its productive potential and value, include natural characteristics (e.g., climate, topography, arability, and soil type) and human modifications of natural characteristics (e.g., fertility, disease resistance, and pest populations affected by past production practices; erosion gullies; groundwater and salinity; capital "improvements" such as terracing, roads, irrigation equipment, and infrastructure).77 Ideally, the value of land itself should be distinguished from the value of 77. Peterson (1986) used an internationalland-quality index to scale total hectares of land. It was an interesting attempt to get at the problem of heterogeneity, but the index has some problems: it was built on 164 Econometric Measurement of the Effects of Research fixed improvements to land (such as buildings, water storage, irrigation infrastructure, fences and roads, and in the case of perennials, the standing stock of trees and plants). Once the nonland component has been separated, it can be treated as part of the capital stock, as described more generally above. When such separation is not possible because data are incomplete, it may be necessary to attempt to treat capital improvements as an element of land quality. For instance, different classes of land (e.g., pasture versus crop, rainfed versus irrigated) can be distinguished and aggregated according to their different rental rates (e.g., Craig and Pardey 1990). An additional set of problems can be encountered when property rights are not clear or where common-property or open-access rules apply. In such settings, it can be difficult to impute a value for the quantity of land being used in the particular crop and livestock enterprises being studied. Judging the appropriate imputation procedures is likely to require detailed informa­ tion on actual production practices. Other conventional inputs: Remaining conventional inputs include such things as irrigation inputs, fertilizers, herbicides. pesticides. seed. electricity, fuel, oil, and veterinary services. In principle, the measurement of many of these inputs is relatively straightforward, since they are typically purchased (at least to a great extent) in markets, making it possible to obtain reasonably good information on prices according to qualities. Also, they are typically nondura­ ble. so that all of the problems associated with dynamics and imputing flows of services (which were discussed under land, labor, and capital) are unimport­ ant, although many of these inputs are storable, and distinguishing between time of purchase and time of use may be important. In practice, however, the statistics that are available on these "other" inputs are often aggregates that have not properly accounted for quality differentials, and the measures of quantities and prices are biased (i.e., the index-number problem). Agricultural chemicals may be measured in physical quantities (e.g., tons) rather than more relevant effective units (e.g., pounds of active ingredient for herbicides or pesticides). This is particularly a problem for pesticides where, within preaggregated categories, the mix and concentration of active ingre­ dients has changed significantly over time. Similarly, organic and inorganic fertilizers must be distinguished, and within each of these classes, there are analogous problems arising from a varying mix and concentration of active ingredients. In inorganic fertilizer premixes. not only does the mix and concentration of, say, nitrogen, phosphorous. and potassium (N-P-K) matter, the chemical forms ofthe N-P-K also matter. Once again, the simple rule is to disaggregate. Additional problems arise with organic fertilizers such as an hedonic approach using only U.S. land values that may not be representative of, nor extrapolate meaningfully to, relative prices in other parts of the world. Econometric Measurement of the Effects of Research 165 livestock manure, green manures, human waste, and crop residues. Clearly there are quality issues similar to those for inorganic fertilizers. In addition, since organic fertilizers are often not traded at all, but rather are an interme­ diate good, both produced and consumed on the same farm or group of farms (i.e., an "inside" input), data are typically woefully inadequate. The "inside" nature of these inputs means that there may be problems in imputing costs to particular production processes, although for broad output aggregates, such problems may not arise. Care must also be taken to ensure that double-count­ ing problems do not arise (e.g., livestock feed consumed on the farms where it is produced embodies the land, labor, and other inputs used to produce it). Weather indexes: Many production studies ignore weather as an input. As a result, typical indexes of output quantity and productivity fluctuate from year to year as a consequence of unmeasured weather influences, so the interpretation of the indexes is obscured by weather-related measurement error. In econometric production studies, the effects of weather are typically relegated to the error term. This practice will not lead to bias or problems of interpreting econometric estimates of production relationships if the omitted weather variables are uncorrelated with the included explanatory variables. In many situations, it is implausible to assume that farmers do not respond to weather within the production period (for instance, harvesting and pest­ control inputs are surely affected by weather), and when that is so, some effort to explicitly account for weather effects may be worthwhile. Serious difficulties arise in identifying meaningful measures of weather that are relevant for explaining the production of individual crops, even those for which the agronomic relationships are well understood. The problem is magnified considerably when crop aggregates are being dealt with. For instance, dryland wheat yields depend on both the timing and quantity of rainfall (evapotranspiration), both immediately prior to planting and at harvest. And other cereal crops depend on similar rainfall variables. The main problem is that different crops depend on different weather variables (rainfall, temperature, humidity) at different times within the year (e.g., Geigel and Sundquist 1984). Among the u.s. studies attempting to account for weather effects, wide­ spread use has been made of a procedure suggested by Stallings (1960). To construct a "weather index," Stallings regressed experimental yields of seven crops from various locations against a linear time trend; the amounts of non weather inputs were presumed constant over time and space. The index for each location and for each crop was defined as the ratio of actual yield to the yield predicted from the regression. Aggregate indexes across regions were formed by weighting the indexes for the individual locations according to corresponding regional production. These, in tum, were aggre- 166 Econometric Measurement of the Effects of Research gated across crops to obtain an overall index. Proxies, policy variables, and other problems: In practical econometric work, it is customary to replace the economic constructs in a theoretical model with more readily available variables that are intended to proxy for the "true" variable. For instance, all of the indexes discussed in this chapter are, necessarily, approximations to the true quantities and prices that are included in theoretical models (because they use discrete data and so on, as discussed above). But some proxies are much cruder, involving much greater leaps of faith than that involved in using a discrete approximation to a Divisia index. For instance, wholesale prices may be used as proxies for their farm-level counterparts, or the price of a particular fertilizer may be used as a proxy for all fertilizers or all agricultural chemicals (or, at least, as a proxy to compute a rate of change for the broader price aggregate). The main statistical problem with using proxies is that they are invariably imperfectly correlated with the "true" variable, and the resulting "errors-in­ the-variables" problem means that the estimates are biased and inefficient. The statistical problems extend to parameters on variables that are measured precisely, not just the one being proxied. In addition, problems of interpre­ tation may arise when the units of proxies for prices or quantities differ from those of the corresponding "true" variables. This problem is so pervasive (and inescapable) in practical econometric work that it is almost never mentioned. Most practitioners seem reconciled to accepting whatever biases are involved and hoping that they are not too important. Policy proxies are particularly problematic. Here, the question is not so much whether a proxy is a close enough approximation to the true variable but, rather, whether it belongs in the model at all. In many cases, the effects of policies are already (or ought to be) reflected in the measures of prices and quantities of inputs and outputs. The additional inclusion of explicit policy proxies is a form of double counting if the other variables are measured properly. Meaningful inferences based on the results from such models are unlikely. On the other hand, certain policies, whose effects are not reflected in appropriate measures of quantities and prices (e.g., the provision of infra­ structure in the form of public goods such as rural roads, education, police, and hospitals), might well warrant the inclusion of an appropriate proxy (Antle 1983; Binswanger et al. 1987; Lau and Yotopoulos 1989; Hu and Antle 1993). Drawing the distinction between which variables should be included in principle and interpreting the estimated coefficients on proxies for those variables in practice is bound to be difficult. Craig, Pardey and Roseboom (1994) suggest that such variables may be providing indirect information on the roles played by conventional inputs - particularly Econometric Measurement of the Effects of Research 167 physical and human capital that have been mismeasured - as well as the direct effects they would be supposed to capture. A final set of measurement issues that arises, particularly in less-developed countries, concerns the differentiation between the productive and consump­ tive uses of inputs and, relatedly, the household use versus marketed quantity of output. Home consumption of farm output remains significant, even in more-developed countries, and is a major aspect of peasant farming systems. Typically, it is difficult to distinguish the consumptive use of land and build­ ings from their uses as productive assets. The same is partly true of vehicles and fuel. Attribution of inputs, especially operator and family labor, among activities is particularly troublesome in the context of part-time farming, which is becoming a prevalent form of farming in more-developed countries, moreso where farming is regarded in part as a hobby or leisure pursuit. 3.2.5 Research and Extension Variables Most of the difficulties associated with including variables representing research, R, and extension, E, in primal or dual models of production arise because research affects agricultural production neither directly nor instan­ taneously. There are considerable time lags between investment in research and the generation of usable technologies, and there are lengthy lags in the uptake of technologies. As new technologies depreciate or become obsolete, their output-enhancing or cost-saving effects may eventually wane. Specifi­ cation of the length and shape of the lag relationship and, relatedly, the depreciation of the existing stock of research-induced knowledge, has been largely ad hoc. Past attempts to estimate rather than impose these parameters on the analysis have been inconclusive. Indeed, recent, and essentially exploratory, studies by Pardey and Craig (1989) and Chavas and Cox (1992) raise more fundamental questions about the common parameterizations used to estimate the productivity effects of research, and they suggest there may be much to gain from more work in this area. In addition to current and past local research, research by other states, regions, and countries on similar production problems, commodities, or factors of production would be expected to have a positive effect on local production technologies. However, environmental factors, among other things, place natural constraints on the degree to which increments to the agricultural knowledge stock made in (or for) one locale can be transferred to others.78 Attempts to identify the magnitude and significance of these research-spillover 78. This new knowledge may be embodied in new plant varieties, agriCUltural chemicals, or agricul­ tura1 machinery, or it may consist of pretechnology material, such as new breeding lines for crops and 168 Econometric Measurement of the Effects of Research effects have produced mixed results and have usually been derived under particularly strong and, in many respects, unrealistic hypotheses. Because the modeling decisions made with regard to these research-lag and spillover issues have a crucial and direct bearing on the measured effects of research, they are discussed here in some detail. Measuring Research79 Constructing the time-series data on research expenditures required for the ex post methods described in this chapter, as well as the ex ante approaches described elsewhere in this book, can be difficult and time consuming. Often there are no consolidated research budgets or expense reports that present data in sufficient detail over a long enough period to enable a lengthy series of aggregate research expenditures to be readily assembled. These data difficul­ ties are compounded if research expenditures are required for a single com­ modity or clearly defined group of commodities, such as food crops, cereals, or small ruminants. There are problems in apportioning aggregate expenditures among specific commodities, particularly for institutes that share research facilities and equipment across a number of commodities or lines of research. It is also common to find a multiplicity of executing, funding, and reporting agencies that provide incompatible, incomplete, and even conflicting data. When data are being compiled, it is helpful to first develop a clear understanding of the institutional history of the local research system, paying special attention to details of institutional mergers and divisions, as well as mandate changes, so that a consistent time series is developed. Reporting standards and disbursement practices differ, so it is often useful to compile data on the research funds available or, more appropriately, on research expenditures according to their source and type. On this basis, three broad expenditure classes can be identified. Core funding: The first category includes funds from state, provincial, and national governments in support of routine or core expenditures. These core expenditures typically cover basic and on-going operational costs such as salaries, utilities, maintenance, consumables, administration, and essential travel and communications. If these data are not available from "performer­ based" records (i.e., by summing across the funds received or spent by all the agencies actually performing the research), it will be necessary to compile them from "source-based" reports. Public funds for research may livestock, as weD as the new research know-how, techniques, and so on that are reported in joumals, books, or symposia 79. For some additional guidelines on compiling data of this type see OECD (1981), UNESCO (1984), PanJey and Roseboom (1989), and PanJey et aI. (1992, appendix 2). Econometric Measurement of the Effects of Research 169 come from a disparate set of government agencies, such as the ministries of agriculture, science and technology, and education, so care is needed to ensure that all relevant funding sources are included. For state and provincial research systems, a significant portion of their funds may also come from the national government, as in the case of the U.S. state agricultural experiment stations, where about 26% of the funds come directly from federal govern­ ment sources (Huffman and Evenson 1993, p. 223). Data obtained from performer-based records could well understate the resources used for research to the extent that administrative overhead and even the salaries of research-related staff appear in ministerial budgets separately from the reported allocations to agricultural research. Conversely, source­ based records may overstate the resources actually used in support of research by including all the administrative and salary costs associated with managing agricultural development and support programs (or teaching activities), where research represents but one, and possibly a minor, component. 80 An additional concern when research expenditure data are being assembled from either source- or performer-based records is the need to capture actual expenditures and not amounts budgeted; the differences can often be substantial . . Other government funds: The second expenditure category is publicly provided funds from domestic governments used for development or proj­ ect-related purposes. These include the costs of executing specific research activities such as additional support for construction, equipment, experimen­ tation and related travel, research data collection, analysis and reporting, project administration, and dissemination of research results. While these expenditures are generally more volatile than core expenditures, they can usually be allocated to particular research programs more readily than is the case with core expenditures. Not all research programs relate to a specific commodity or narrowly defined group of commodities. Research on factors of production (e.g., soils and water) or research of a more basic nature (e.g., development of modem biotechnology techniques, cytology, or fundamental studies of growth hormones) can have productivity-enhancing effects across a range of commodities. This makes it difficult to apportion these types of expenditures to a particular commodity in any meaningful way. But to the extent that these aspects of a research program do or should ultimately have a commodity impact, albeit often a multicommodity impact, it is reasonable to consider them as a component of the relevant commodity 80. If government bureaucracies are bloated because of public-sector job-creation practices, it might be difficult to justify charging the extraordinarily large "administrative overheads" of the central govern­ ment bureaucracies that come with this overstaffing against the research agency being evaluated. In fact, it has been argued that overly bureaucratic systems stifle research ingenuity and actually reduce the stream of benefits coming from research. 170 Econometric Measurement of the Effects of Research program( s) and apportion them accordingly. 81 One practical option for doing so is to assume that the share of project-related expenses that can be readily identified with a specific commodity program is representative of the overall commodity orientation of the research so that (3.62) where Rj " is the estimated total expenditure on commodity j in year t, R, is total research expenditure in year t, rj" is project-related expenses in year t that are readily allocated to commodity j, and l:j rj , is the sum of project-re­ lated expenses in year t that can readily be allocated to one of n commodities. Allocating preaggregated core expenditures to specific commodities is problematic. Given that a substantial fraction (often 60% to 70%) of core budgets is used for salaries, a plausible approach is to compile data on the time researchers spend working on particular commodities (preferably mea­ sured in full-time equivalents of actual time spent doing research) and use this information to partition total core expenditures. An alternative approach is to use the share of project-related or development costs going to each commodity as the basis for apportioning the core budget. The human capital component that typically dominates core budgets gives rise to highetdegrees of "fixities" relative to project-related budgets (e.g., it may be difficult to convert a rice breeder rapidly into a com breeder). Thus, it may be useful to smooth the fluctuations in development expenditures when using them to prorate pre aggregated core expenditures by adopting some variant of the following equation: n-I Rtt = R; x;; :0 (rj,t-kiRt-k ) (3.63) where R;,I =t he imputed core budget expenditures on commodity j in year t, R; =t otal (or unallocated portion of) core budget in year t, rj,,-Ic =d evelopment or project-related expenditures on commodity j in year t~k, and Rt-k = total project-related expenses in year t-k. Donor funds and grants: The third class of expenditures includes funds received from other sources, such as loans or grants from donor govern­ ments, public agencies, and private-sector organizations; funds recouped from fee-for-service activities; sales of farm produce; funds from the sale of new technologies; user fees, such as patent royalties or licence charges; and 81. Presumably, research on improved soil management practices, for example, is targeted toward a particular, and usually representative, site and/or production system, for which the likely commodity mix is readily apparent. Econometric Measurement of the Effects of Research 171 so on. Each of these raises its own particular set of measurement difficulties. In measuring the flow of donor-sourced funds going to research, particu­ lar care is needed to distinguish between actual expenditures and funds allocated or made available. Nontrivial and unpredictable lags between the commitment and the actual disbursement of donor funds mean that in many cases, a sizable portion of the funds allocated to a multi-year project is never actually spent. The mismatch between funds available and expenditures is compounded by the fact that donor support is often biased toward lumpy capital items such as new buildings, pl~nt, and equipment. Unfortunately, in many instances, the available data are restricted to information on budget totals and project starting and ending dates. In this case, the practical option is to assume that the appropriated or, preferably, the expended budget total was disbursed in-equal annual amounts for the duration of the project. It is common for donor-funded projects to cover a number of commodities (e.g., a research project on food crops, on upland agriculture, or on farming systems) and to include nonresearch activities (e.g., development of irriga­ tion infrastructure or joint research-extension activities). These budget data are often reported along functional rather than programmatic lines. In the absence of a detailed accounting of the use of the funds, the judgment of knowledgeable research scientists and administrators can be used to estimate the portions of project funds going to research on particular commodities. There can also be difficulties in identifying the research component of farm operations that may be undertaken in support of agricultural research. To the extent that such farm operations are necessary to execute a program of research, it is appropriate for them to be -included as part of the cost of doing research. But some systems engage in farming operations well beyond what is required to support research. In some cases, the surplus earnings from farm sales are used not only to support research but also to supp~rt a whole host of nonresearch activities. In these instances, including all of the re­ sources devoted to farm operations can substantially overstate the costs of doing research. Research expenditures are a more accurate measure of the funds used to do research than simple budget or funds-available figures where, for exam­ ple, the year-to-year carryover- of unspent funds can lead to a significant mismatch between the temporal pattern of funds allocated versus funds actually spent on research. The capital budgeting exercises described in section 5.4.2 make it clear that the pattern of research expenditures (as well as the patterns of research benefits flowing from these expenditures) can significantly alter the estimates of discounted research costs and benefits, or rates of return to research, coming from such an exercise. Research expenditures versus service flows: A research expenditure 172 Econometric Measurement of the Effects of Research series might not be a particularly accurate measure of the real resources actually used to do research. The major problem in this regard arises when a large share of total expenditures is invested in physical capital inputs such as new or refurbished buildings and research facilities, equipment and autos, or the upgrading of human capital.82 This is especially common in newly established or rapidly growing systems. Such an expenditure aggregate overstates the flow of services coming from these research inputs in years of high capital expenditures and correspondingly understates them in subse­ quent years. This is because the resources spent in constructing new build­ ings and training scientists can produce a stream of research services for many years after the investment has been made - assuming adequate maintenance, repair, and in the case of scientists, continued training and opportunities for professional development. It is the service stream, not the corresponding expenditure series, that is the best measure of the resources actually used to do research. To express a research input aggregate in terms of service flow, it would be necessary to identify expenditures on durable inputs, such as buildings and equipment, autos, and training of personnel, apply the procedures described in section 3.2.5 to derive service-flow esti­ mates for these capital inputs, then combine these estimates with recurrent costs such as salaries and consumables to give a measure of aggregate service flows (pardey, Craig and Hallaway 1989). \/A basic distinction must be drawn between research expenditures (a rrteasure of the costs of research) ~nd an index of the quantity of research inputs. The expenditure series simply adds up expenditures across cost categories, giving them equal weights. In contrast, a quantity index aggre­ gates across different categories - flows from research capital stocks, such as buildings, land, and equipment (as measured by aggregate service flows), as well as from current consumption of nondurables and labor - as de­ scribed above, using different weights for the different categories. These two ways of measuring research are used for different, but related, purposes. Consider an analysis of a time series of annual investments in research. An economic evaluation of the stream of research must use the expenditure series, while the economic explanation of the consequences of research, in an econometric model, is best served by using the quantity index for each year. This is because the evaluation involves comparing the stream of actual costs to measured benefits, whereas the explanation attempts to aggregate across different categories of expenditure, accounting for changes in the composition of that expenditure, since different types of expenditure 82. During its formative years, the state agricultural experiment station system in the United States spent upwards of 29% of its total expenditures on capital inputs. For some states in some years the share was around 60% (Pardey, Craig and Hallaway 1989). Econometric Measurement of the Effects of Research 173 have a different impact on the production of new knowledge, and a constant relationship is posited between the production of new knowledge and the quantity of research. In a practical evaluation study, the quantity index may be used in the first stage to estimate the parameters of the research production function. Then, in a subsequent step, alternatives are simulated. In order to evaluate the alternatives, it is necessary to translate the series of simulated counterfactual quantity indexes into a research-expenditure counterpart. This step is usually left implicit, and we suspect the issue has been largely overlooked. To our knowledge, all previous studies have used research expenditures directly in the econometrics, ignoring the index-number problems. That omission has involved an implicit assumption that the composition of total research expenditure is constant or, at least, that changes in the composition of expenditures are econometrically unimportant. Such an assumption is con­ venient in that data to enable a better index to be constructed are often unavailable.83 Also, work remains to be done on how best to design counter­ factual simulations of the streams of component expenditures. Measuring Extension Throughout this book, _we have regarded and treated extension as a component of a continuum of types of R&D activities that interact and together determine the impact of R&D expenditures. On this view, extension is the same, conceptually, as any other R&D input: the measurement issues are no different and the approaches are the same as for agricultural research, as discussed above. And the typical distinction between research and exten­ sion is somewhat arbitrary; similar distinctions might also be drawn, with equal or greater justification, among other categories within the R&D aggre­ gate (e.g., between pretechnology, applied, and development research; or public, private, and foreign research, as discussed in chapter 2). Like different types of research, different types of extension have different types of effects. Some research conducted in u.s. agricultural experiment stations, for instance, bears no perceptible relationship to production agricul­ ture, including some applied work on nonagricultural topics as well as basic research. Similarly, some extension work is directed toward urban issues. In some cases where there might be a connection to production agriculture, its nature may be vague (e.g., 4-H extension activities, home economics, etc.). Some previous· studies (e.g., Huffman 1978) have attempted to distinguish 83. See Pardey, Craig and Hallaway (1989) for an example of decomposing research costs into various categories. See also Mansfield (1987) and Bengsten (1989a and b) for a related discussion of R&D price indexes. 174 Econometric Measurement oftke Effects ofR esearch between extension work that affects productive efficiency and that which affects allocative efficiency. Such distinctions are difficult to draw concep­ tually, let alone in the data that are typically available. Typically, for both research and extension, we must aggregate across types of expenditures (as well as over time), and more often than not, extension is aggregated with research in a single R&D variable (e.g., Gril­ iches 1964; Evenson 1968; Cline 1975; Mullen, Cox and Foster 1992). Such aggregation is undesirable to the extent that research and extension have different kinds of effects on knowledge or output. It is also undesirable where there is some interest in differential effects or where the mix of research and extension has been changing over time so that aggregation bias is likely to result from aggregating research and extension.84 In any of these cases, it would be preferable to include separate research and extension variables (perhaps allowing for interaction effects), but in practice, multicollinearity is likely to be a problem. The use of preaggregated research and extension variables, as by Huffman and Evenson (1992), might be necessary to circum­ vent the statistical problems that can arise when both research and extension variables are included in an econometric model. Like research, while much of the extension effort is carried out in the public sector, a significant and increasing amount is being done privately by agribusiness, either as part of their marketing effort or on a fee-for-service basis. And farmers themselves invest heavily in search and screen activities among private and public information sources, which are increasingly acces­ sible through electronic and print media. Data on either agribusiness or farmer investments in extension-related activities are typically not available in a form suitable for econometric analysis, if at all, so the main implication of this is that one must be aware of the potential biases from excluded variables when interpreting estimates from an incomplete model. A similar caveat applies with equal or greater force to the very common (almost universal) practice of estimating models including public- but not private­ sector research. 84. In particular. it is commonly suggested that extension lags are shorter than research lags. and spillover effects may be important for research (for which the model might need to include research spill-in variables as well as I~ research) but not for extension (for which it might be sufficient to include only local expenditures). Econometric Measurement of the Effects of Research 175 Aggregating Research and Extension - Temporal Aggregation The capital-theory approach to modeling the generation and use of knowl­ edge in agriculture defines a stock-of-knowledge variable, K" in terms of the (depreciated) stock of existing knowledge, K,_t, and net increments to that stock, I" as in section 3.1.1. Assuming that a geometric, declining-balance, depreciation process applies to the knowledge stock, such that Dt =( 1 - OK)K" equation 3.3 becomes (3.64) where OK is the rate at which existing knowledge becomes obsolete because it is replaced by better information or circumstances change to make it less useful. Crop management practices developed and used in the 1950s may be of limited use today given the availability of genetically superior crop varieties, improved agricultural chemicals, improved farm machinery, and so on. Such obsolescence is reinforced by biological deterioration through, for example, the evolution of resistant diseases and pests that lower yield potential and may increase the yield variability of existing crop varieties over time.8S Changing economic circumstances can shift relative prices in a particular locale and lead to changes in local output and factor mixes that lower the productive value of existing agricultural knowledge targeted for that locale. The notion of a stable relationship between research expenditures and increments to the stock of knowledge follows naturally from the perception that in general, science progresses by a sequence of marginal improvements rather than through a series of discrete, essentially sporadic, breakthroughs.86 The systematic aspects of this input-output relationship in research are directly related to the degree of aggregation being employed. For instance, the annual publication count or knowledge output of individual scientists may vary quite capriciously, but this random variation is less important at higher levels of aggregation (e.g., at the project, program, or system level).87 A general form for the knowledge or research production function was defined in section 3.1.1 (equation 3.4) as I, = i (R,. ... ,R,-L , K,_I, Z,) (3.65) R where current increments to knowledge, It, are determined by the historical pattern of research expenditures, the existing stock of knowledge (here 85. See Anderson and Hazell (1989) and the references therein. 86. See, for example, Minasian (1969), Rosenbelg (1976), Griliches (1979), Pakes and Griliches (1980), Kamien and Schwartz (1982), and Pardey (1986). 87. For this reason, among others, it does not make much sense to use these procedures to evaluate the effects of a narrowly defined field of research. 176 Econometric Measurement of the Effects of Research broadly defined to include fundamental and pretechnology types of knowl­ edge) and a set of institutional variables, Z" related to the management and deployment of research staff and the resources with which they work. The mix of research may matter. When budget cuts curtail the fundamental sciences, the marginal gains from continued investments in the more applied agricultural sciences may fall. 88 Similarly, the efficacy of agricultural re­ search is likely to vary with the commodity, site, and problem focus of the research. Some production problems for some commodities (e.g., rust in wheat or black sigatoka in bananas) are simply harder and, consequently, more costly and time-consuming to solve than others . . Taking equations 3.64 and 3.65 together, it is convenient to view mainte­ nance research - which seeks to replace or replenish the (research-induced) productivity gains from past research investments - as leading to new knowl­ edge that substitutes for declines in the stock of usable knowledge. Ruttan (1982, p. 60) speculated that the rate of knowledge decay, OK' is likely to increase as agricultural production systems become more knowledge intensive so that a larger share of research resources is needed simply to maintain the exisiting knowledge stock. This decay rate is an additional source of "instabil­ ity" in the relationship between research and knowledge stocks. If a particular crop in a specific locale is analyzed, then the relevant OK may well be quite volatile over time and not constant as depicted here. For instance, in the early 1970s, outbreaks of brown planthopper in Indonesia inflicted a great deal of damage on improved but nonresistant varieties of rice, implying a significant and sudden increase in OK for genotypic-based rice technologies used in parts of Indonesia. As the commodity and locational focus of the analysis broadens, one would expect a lower and more stable rate of knowledge decay.89 The extent of utilization of the existing stock of knowledge (i.e., the service flows, F, arising from the existing knowledge stock) depends upon the relative prices of products, P" and factors, WI' the stock of farmers' human capital, H" and the extent and quality of extension services, E" among other things, so that Ft = f (Kt , Pt .. Wt , H t , Et) (3.66) Combining equations 3.64, 3.65, and 3.66, it is possible to accommodate these conceptual ideas in an empirically tractable way by defining the stock of knowledge in use as 88. Evenson and Kislev (1975, ch. 8) develop this idea in more detail. Recent examples of this phenomenon are modem technologies of gene manipulation and control, which had their origins during the early 1970s in the health sciences (Persley 1990) and have given rise to the "biotechnology revoution" that offers the promise of significant technology advances within the agricultural sciences. 89. Although in a shift to more intensive, monocropping systems, a consequent narrowing of the genetic base in use may offset this tendency. Econometric Measurement of the Effects of Research 177 (3.67a) where the effects of prices and human capital variables are suppressed, for simplicity. Assuming a constant-elasticity functional form for the lag rela­ tionship, the stock of knowledge in use may be defined as90 Kt = IT R}~r IT E1~e (3.67b) r=l e=l where, in this instance, A.T and ell. are the lag coefficients that specify the shape of the lag profile linking local research expenditures in period t-r, R'_T' and extension expenditures in period t-e, E, .• , to the stock of knowledge being used in period t. The process by which investments in research and extension lead to changes in technology is complex.91 Quantifying the process usually entails aggregating across a portfolio of research activities (e.g., across different technology types, commodities, or institutions) that vary in many respects, not least in their lag profiles. The lag between the inception and completion of a line of research (i.e., the research gestation lag) can be around two to three years for some crop-management types of research (e.g., developing improved fertilizer rec­ ommendations). The lag may be even shorter if the research is highly adaptive or of a search-and-screen nature that builds directly on earlier work. In contrast, conventional breeding programs for cereals usually take six to ten years to­ develop a new variety, while similar work on perennial crops such as coconuts or bananas can take up to fifteen years. All of these lags are influenced by the specific research problems being addressed and by the experience and talent of the researchers doing the work. For example, a barley breeding program may seek to increase yield potentials, improve nutritional or malting qualities, incorporate some degree of pest or disease resistance, or involve some combination of all of these traits. Lags are also affected by the institutional environment in which the work is carried out. For example, persistent problems of insufficient or untimely disbursement of 90. It is important to note that the choice of functional form for the construction of the index of the stock of knowledge in use is not dictated by theory and may have implications that are undesirable. Here. for instance. the choice of a geometric (Iinear-in-Iogarithms) form means that every past value of research or extension is necessary in the sense that if the value of anyone of them is zero. the value of the stock will fall to zero. 1be linear-in-Iogarithms form is econometrically convenient when combined with a Cobb­ Douglas model and has been popular for that reason. A linear-in-leveIs form would not have the problem of all past values being necessary but it could imply increasing returns to scale in some specifications. This is another example of the general problem that analytical convenience and tractability. which often dominate specification c1X1ices. may not always fit well with the other requirements of the analysis. 91. Nelson and Winter (1982). for example. develop an evolutionary model of technical change that tries to incorporate some of these complexities. 178 Econometric Measurement of the Effects of Research operational funds may result in missed growing seasons and considerably longer research cycles. Research programs with limited experience or expertise in a particular line of research may incur substantial start- up costs or delays that do not apply to on-going, "mature" research programs. Once new technologies are available, there are further, and often substantial, lags in their adoption. While many of the green-revolution technologies have been developed and extended in package form (e.g., new plant varieties plus recommended fertilizer, pesticide, and herbicide use, along with water control measures) many of the components of these technological packages are taken up in a piecemeal, often stepwise, fashion (Herdt and Capule 1983; Byerlee and de Polanco 1986). Eventually, new technologies are given up as they are replaced by superior or substitute technologies. Natural, economic, and politi­ cal forces all have a direct bearing on adoption and dis adoption processes.92 The role of public and private extension services - broadly defined to include the technology-transfer activities of the input-supply sector and the search-and­ screen activities of farmers themselves - are particularly pertinent in this regard. At the same time, some countries have unduly cumbersome or ineffi­ cient seed certification and distribution systems that cause additional delays in getting new technologies into farmers' fields. Such complications make it difficult, both conceptua!}yand empirically, to isolate the effects of research from those of extension.!The specification provided in equation 3.67 implies that it is possible to estimate separate effects of research and extension on agricultural output or productivity growth. To the extent that the effects of research are not independent of expenditures on extension, it might be appro­ priate to incorporate a research-extension interaction term (or terms) into this specification (Evenson 1988; Huffman and Evenson 1993). In practice, the temporal weights A.T and and yare the two Pascal parameters. Estimates can be made by searching over values of 9 or cI> and y for a specification that minimizes the sum-of-squares residuals after imposing LR• If 9 = 0.5, the binomial lag structure approximates a symmetric, normal distribution; with small values of 9 the distribution is skewed right (i.e., with the peak near zero so that the mode is always less than the mean); for very small values a geometric (e.g., Koyck) distribution is approximated; for 9 > 0.5 the distribution is skewed left. A Pascal distribution is skewed to the right if Y is large, with smaller cI> leading to greater skewness. If y = 1, the Pascal is simply a geometric distribution; it approaches a Poisson (of mean m) as y -t 0 and 1$ -t m, and if m is large, the limiting form is a Gaussian distribution. Form-free lag structures: While these types of distributions are quite flexible, the choice of anyone distributional form, a priori, still imposes a definite (and not necessarily appropriate) shape on the individual lag coeffi­ cients. An altogether different approach is to use the form-free distributed lag technique developed by Hatanaka and Wallace (1980) and apparently first used by Silver and Wallace (1980). In the Hatanaka and Wallace approach, it is not the lag structure per se but the moments (Le., the total lag effect or sum of coefficients, 111>, the mean of the distribution, Ill, the variance of the distribution, 112, and higher-order moments)98 of an arbitrary lag distribution that are estimated directly. Applying the Hatanaka and Wallace (1980) form-free approach to estimat­ ing R&D lag structures would mean including current and lagged research expenditures as separate regressors in the function being estimated. While the A., coefficients on the individual research expenditure variables (measuring the short-run effects of research on output, variable cost, or profit) are generally estimated with low precision, linear combinations of these coefficients are likely to have lower variances than the individual coefficients. This is because positively autocorrelated variables, such as most time series of research (and extension) expenditures, give rise to estimated regression coefficients that by 98. With this notation, the i subscript on Iti indicates the order of the moment, so that i = 0 is a zero-order moment, i = 1 is a first-order moment, and so on. Econometric Measurement of the Effects of Research 185 and large have negative covariances. Therefore, the sum of the Ar coefficients has a variance smaller than the sum of their variances. Moreover, Hatanaka and Wallace (1980) argue that in their method, the lower-order moments (i.e., the total effect, 110 = L).,., the mean III = r.rA,., and the variance J.12 = r/Ar), which are of greater practical interest, are generally estimated with greater precision. Moreover, there is a hierarchy of precision: V(Ilo) < V(IlI) < V(J.12). Directly estimating the moments of a lag distribution using a procedure that imPoses no priors on the shape - but which does require the fairly plausible notion of smoothness and non-negativity concerning the lag weights and also presumes a given lag length - affords the analyst various options.99 One option is to employ the method-of-moments technique to estimate the relevant param­ eters of a flexible probability density function using the moments obtained from a first pass through the data, and then to impose shape restrictions that are not at variance with the data when research expenditures are preaggregated for a second round. This approach is described by Silver and Wallace (1980), for example. Spatial Aggregation The new knowledge generated from research by other states, regions, or countries on similar outputs, inputs, or production problems can have a positive effect on the performance and payoff to local research. The impor­ tance of spillover effects has been studied for a number of different industries (Jaffe 1986; Bernstein and Nadiri 1988, 1989) in addition to agriculture (Evenson and Kislev 1975; Evenson 1989) and was recently reviewed by Griliches (1992). Empirical estimates for spillover-intensive industries in the United States attribute up to two-thirds of the research-induced productivity gains to spillovers (Bernstein and Nadiri 1989). A form-free approach to assessing the spillover effects of nonlocal re­ search, which is analogous to the form-free approach described above with respect to temporally aggregated research expenditures, would be to include temporally preaggregated research expenditure variables in the analysis for each of the nonlocal research sites deemed relevant in terms of their potential to generate research spill-ins to the site of interest. For example, if the units of analysis were states or provinces within a country, then separate research expenditure variables would be included for all other states, regions or countries doing research that is potentially relevant to the "home state." This approach is usually not feasible because of problems with multicollinearity 99. Truncating the fitted structure so that it is short relative to the ''true'' (but unknown) lag structure will cause estimates to be biased; including lagged tenns whose coefficients are zero does not bias the estimates but does increase the variances. 186 Econometric Measurement of the Effects of Research oulegrees of freedom. In a manner analogous to the temporal aggregation procedures discussed above, the most tractable approach is to define a plausible set of spatial weights, 'YhI" which enable research conducted in a number of other locations, R~.p to be combined into a research spillover aggregate, S:." that reflects the pool of nonlocal research that may spill in to the home region, k, such that H S;,t = n( R~,t) 1h1; (3.74) h#< Then an augmented version of 3.67a can be formed so that Kk,t = k (RZ,t, S;,t, EZ,t ) (3.75) where Kk., represents the current (period t) stock of knowledge in use in the kth locale, and R:", s:", and EZ,t respectively represent the preaggregated local research, nonlocal research, and local extension variables that pertain to the kth locale. In constructing S:,,, one option is simply to sum all the nonlocal research (i.e., 'Yhk = 1.0 for all hand k). This is not a particularly fruitful approach. It does not address the question of how to decide what constitutes relevant research, and in so doing, it inappropriately treats all nonlocal research as being equally relevant from the perspective of its spill-in potential to a particular locale. In practice, most country studies exclude all research done abroad (i.e, the 'Yh/cs are implicitly set to zero when h indexes research carried out by other countries), even though there is clear evidence, for at least some commodities, that substantial international technology transfers are pos­ sible. 100 And for those studies where the unit of analysis has been states or regions within a country, there has been little uniformity in the way out-of­ state research has been treated. For several reasons, distance from the source of information plays a significant role in shaping research spillover potentials, particularly in agri­ culture, (e.g., Lindner, Pardey and Jarrett 1982), but spatial closeness is an incomplete indicator of spillover potential. Evenson (1980, 1989), Otto (1981), and others have used the concept of an agroecological zone and the relationship between geopolitically defined regions (e.g., states within the United States) and these zones when defining the 'Yh~' The idea is that the potential for agricultural research to spill over is higher between locations that are in some sense agroecologically similar than between locations that are not. While this is intuitively appealing, it is often difficult to implement. First, there are problems in defining the technical criteria (e.g., aspects of 100. See, for example, Brennan (1986) in the case of wheat and Pardey et aI. (1992) for rice. Econometric Measurement of the Effects of Research 187 soil, climate, and topography) by which to construct relevant agroecological zones (Wood and Pardey 1993). Second, depending on the scale at which they are defined, the agroecological zones may not correspond closely to the geopolitical regions by which research expenditures and other economic variables are usually reported. Deciding which zone (or part thereof) and hence which Yhk is pertinent for any two regions hand k is problematic. Studies that aggregate across a number of commodities or technologies that vary in their site-specificity face further problems. Some subsectors (e.g., intensive livestock operations or glasshouse horticulture) may be less sensitive to variations in the natural environment, so the potential for spillovers to occur over greater physical distances is higher. In addition, economic, institutional, and other factors exacerbate or ameliorate the ecological constraints to tech­ nology transfers. Given the conceptual and practical difficulties of formally using agro­ ecological zones to measure region-to-region spillover potentials (i.e., they~), some less-demanding "reduced-form" procedures may be in order. It is reason­ able to expect that the economic, ecological, and social characteristics of a region are reflected in the commodity orientation and input use within the region. On the presumption that local research is congruent with local com­ modity (and input) mixes, a measure of technological proximity, yl:Jc, can be used as a proxy for spillover potential. Given an n-dimensional vector of outputs and an m-dimensional vector of inputs, the "position" of region s in input-output space can be characterized by a vector F. =< / , ... ,/,,/,+1 , ... , r+n) where each of the elements of F. denotes the region's or state's quantity of outputs,j =1 , ... , n, or quantity of inputs i= 1, ... , m, respectively. Following Jaffe (1986, 1989), a measure of the technological proximity, Yi:k, between any two regions h and k in input-output space can be defined as FhF: yft = ~(FhFh')(FkFk') (3.76) Clearly, Ykk equals one if both input and output quantities - up to a factor of proportionality - are identical and will approach zero the more dissimilar the input and output mix is between any two regions. If the output bundles of two regions are dominated by corn, for example, then the research portfolios of both regions will be similarly biased toward com. In this case, the spillover potential would be higher than if the comparison were between a predominantly corn-producing region and, say, a coconut-producing re­ gion. Differences in the factor mix across regions - for instance region 1 produces corn using "low-input" rainfed technologies while region 2 uses "high-input" irrigation technologies - means that the spillover potential of 188 Econometric Measurement of the Effects of Research com-related research is lower than if patterns of factor use and technology structures were similar across the regions. 3.2.6 Statistical and Econometric Issues In addition to the conceptual and measurement issues discussed above, several econometric problems can arise in models used for research evalua­ tion. The most severe problems typically includej a) specification error, (b) multicollinearity, and (c) simultaneity problems. SpecJication Error If relevant variables are omitted from a model, then the included variables must explain the effects of other omitted variables as well as their own effects. As Griliches (1957) and others have shown, in a linear model, E(h l ) = ~1+~2PI where ~I is the "true" coefficient, hi is the OLS estimate of that coefficient when another relevant variable has been omitted (and E(h l ) is the expected value of the OLS estimate of ~I)' ~2 is the coefficient that would have been found on the omitted variable, and PI is the regression coefficient that would be found if the omitted variable were regressed on the included variable. If the omitted variable is not correlated with the included variable (i.e., PI =0 ), the estimated coefficients on the included variables will be unbiased. If PI '# 0, the bias depends on the signs and magnitudes of ~2 and PI' In the examples discussed earlier, if extension, education, and private research are omitted, and these variables are expected to positively influence production and are positively correlated with research, then the research coefficient will be biased upward. When a variable is included but mismeasured, the coefficients of included variables may be biased as a result. For example, if the quality of an input has changed over time but a quality adjustment has not been made, the coefficient on the research variable will be biased if research is correlated with the quality of the input. There has been a debate in the literature over whether inputs should be left unadjusted for quality changes that result from research so that a more complete assessment of the impact of research can be obtained from the research variable. 101 Multicollinearity Multicollinearity can be a serious problem, particularly with time-series data, when production functions are used to evaluate agricultural research. Many variables move together over time, and research, extension, and 101. See Denison (1961, 1969) and Jorgenson and Griliches (1967). Econometric Measurement of the Effects of Research 189 education are often highly correlated, particularly if the formulation of the lag structures calls for the inclusion of more than one variable for each factor. When multicollinearity is serious, the large variances on regression coeffi­ cients for the collinear variables mean that limited confidence can be placed in the parameter estimates. 102 The model becomes highly sensitive to speci­ fication and to sample coverage. Parameters exhibiting large standard errors - instability and lack of conformity to prior expectations - may indicate multicollinearity. Sometimes, however, problems arising from specification error are attributed to multicollinearity. \03 There are no easy solutions. More data, more prior information, or some other means of making fewer demands on the data are the only options. Sometimes prior beliefs can be imposed as restrictions on the coefficients on nonresearch variables (e.g., making use of mixed estimations in which coefficients on conventional inputs in a Cobb-Douglas production function are constrained to be equal to their factor shares). This procedure assumes profit-maximizing behavior and equilibrium. \04 Biased estimation proce­ dures such as ridge regression and principal-components regression are also feasible. These procedures trade off variance for bias, but selection of a final regression must be based on highly subjective tests. IOS Ridge regression can be expected to perform best in situations where the data are highly collinear and all coefficients are expected to be of the same sign and roughly the same size, as may arise in production functions estimated with time-series data. \06 Multicollinearity among nonresearch, extension, or education variables is likely to be less of a problem with cost, profit, or supply functions than with production functions because most real input and output prices are less highly correlated than production inputs. On the other hand, serial correla­ tion may be more serious. Simultaneity and Causality Simultaneity is often a problem when a production function is estimated to evaluate agricultural research.l(17 The problem arises for two major reasons. \02 For example, see Belsley, Kuh and Welsch (1980). \03. Diagnostic tests based on variance inflation factors, condition indexes, and variance proportions, which are available with many regression packages, can be used to detect some such problems. \04. The use of productivity functions may be thought of as a further example of this type of approach, where the structure of the technology is effectively imposed as a prior restriction in order to focus the estimation on the response to research. lOS. For examples of ridge regression applied in an evaluation of agricultural research, see Norton, Coffey and Frye (1984) and Leiby and Adams (1991). 106. See Brown and Beattie (1975). 107. See Marschak and Andrews (1944). 190 Econometric Measurement of the Effects of Research First, conventional inputs may not be behaviorally and, hence, statistically exogenous variables. Uncontrollable factors during a production period (e.g., weather or pests) commonly affect both inputs and outputs (affecting the input-output relationship), so the measured effects of inputs become biased because the input quantities are correlated with the error terms in the model. Second, future output and its profitability may depend on past research, but research expenditures in tum may depend on both past output and expectations about its future. 108 These two sources of simultaneity are related but arise in somewhat different ways. Simultaneity between the quantity, price, or value of production of a commodity and the value of research spending can arise from funding arrangements that use output levies or taxes, for example. In these cases, a rise in output (say, in response to an increase in demand) will lead to an automatic increase in the total funds available for research. Similar forces might also be at work in situations without explicit tax arrangements if, for instance, research resources are allocated according to a congruence rule - in proportion to the value of output. Also, while farmers do not directly choose a level of research in the same way as they choose conventional inputs, changes in output levels can lead producers to pressure governments to adjust support for research in general or for particular commodities. These forces are not likely to result in serious statistical problems of simultaneity because of timing: current research does not affect current output but current (or recent) output can affect current research. The relationship is more likely to be recursive than simultaneous. Using a set of causality tests, Pardey and Craig (1989) provide evidence that while research expenditures do . cause output changes, output indexes also carry information that helps predict future research expenditures. Their results suggest that part of what is measured as an effect of research on output may in fact reflect the effect of output changes on research. The empirical significance of simultaneity problems is difficult to assess and there is no truly satisfactory solution to the problem in the produc­ tion-function framework. While an instrumental-variables approach may be tried, in a study that involves estimating a large number of commodity produc­ tion functions, it might be preferable simply to acknowledge the possible bias. It is often argued that simultaneity is less of a problem with profit- or cost-function models because many of the independent variables are prices. In a competitive market, prices are determined by the market and not by individ­ ual farmers. Hence, price variables and error terms should be uncorrelated (Varian 1978). However, while such reasoning may be valid for models estimated with farm-level data, it may be less valid when aggregate national, 108. See Griliches (1979), Pardey and Craig (1989), and Schirnmelpfennig and Thirtle (1994). Econometric Measurement of the Effects of Research 191 regional, or state-level data are used. For example, if national time-series data are used, observed output prices may be influenced to some extent by aggregate output levels. Furthermore, just as with the production function, research expenditures may be influenced by aggregate output levels. With either of these situations, the independent variables might be correlated with the error term. Simultaneity between output and price may be reduced by use of lagged or futures prices, but it probably will not be completely eliminated, particu­ larly with futures prices. There is little that can be done to reduce the problem associated with output causing research expenditures, although the problem is mitigated somewhat by the fact that research expenditures are included in a lagged fashion. An instrumental-variables approach could be used in which the instrument for the research variables would be fitted values from regres­ sion of research expenditures on their lagged values, lagged expected output prices, and other variables hypothesized to influence research expenditures. Simultaneity may also be a problem with single-equation supply models. I09 3.3 Calculating the Effects of Research Econometrically estimated production, productivity, cost, and, profit func­ tions can provide useful summary measures of the structure of production technologies and their change over time. Aspects of these production technol­ ogies are represented by the size and sign of the marginal product of inputs and their associated output (cost or profit) elasticities, economies of scale and economies of scope, elasticities of input substitution, and so on. Our primary interest is in estimating the effects of agricultural R&D on output, cost, profit, and input use, with a particular emphasis on developing measures of the benefit streams flowing from past investments in research. To the extent that changes in the more familiar production parameters for quantifying the effects of R&D matter, they will be given due attention, but mostly we refer readers to texts on production economics for a review of the specific details on measurement and interpretation concerning these production parameters.110 3.3.1 Growth Accounting The rates of growth in aggregate agricultural output and its components differ markedly over time and across regions and countries. While the 109. See Huffman and Miranowski (1981) for an example of combining a demand function with a supply function that includes a research variable. 110. See, for example, Ferguson (1975), Fuss, McFadden and Mundlak (1978), Beattie and Taylor (1985), and Chambers (1988). 192 Econometric Measurement of the Effects of Research increased use of conventional inputs such as land, labor, and capital, as well as purchased inputs such as fertilizers, pesticides, and energy, often accounts for a sizable share of the measured growth in output, it is by no means the only important source of growth. Changes in the quality of conventional inputs - such as improvements in the human capital aspects of labor and new and improved agricultural machinery, fertilizers, and pesticides - can also be significant sources of growth. For our purposes, it is desirable to distinguish the growth consequences of the new technologies and know-how attributable to investments in public agricultural research and extension from other growth-promoting influences that stem more directly from pri vate R&D investments and public spending on education, health, and rural infrastructu­ ral services. Accounting for these various sources of growth in order to get quantitative indications of their importance and to identify whether their relative contribution has changed over time is a useful way of summarizing the agricultural sector's historical pattern of development and the contribu­ tion of research to that development process. A simple yet informative first step in this direction is to quantify the relative importance of yield versus area effects on the measured growth in agricultural output. Because Qt == At (Qtl At) (3.77a) where Q, is output in period t, and yield, Y, = Q/A, is output divided by area, it follows directly from taking the logarithmic differential of equation 3.77a that q,= a,+Y, (3.77b) where dQ, 1 dA, 1 dY, 1 q,= dt Q,' a'=dt A,' and Y'=dt Y, Dividing through equation 3.77b by q, enables the rate of change in output to be partitioned into that proportion due to changes in area and that due to changes in yield. III Using the econometric procedures described in this chapter, a more complete accounting of the sources of growth is possible. Taking the deriva- 111. Notice that if area is measured in tenns of stock (e.g., hectares in rice) rather than in tenns of flow (e.g., hectare plantings or area sown to rice), then multiple-cropping effects will be captured as part of the measured increase in yield rather than an increase in area. To the extent that these multiple the inverse of the commonly calculated research-intensity ratio) was constant for all t and r =0 , ... , LR• In fact, it is unlikely that the total marginal products of research derived from equations 3.85a and 3.85b would be equal. Such outcomes are attributable to the use of the Cobb-Douglas functional form, which involves constant elasticities rather than constant marginal products. The distinction between the total marginal products of research obtained from 3.85a versus 3.85b is rarely made in practice. A common approach is to use an averaging or approximating procedure to calculate the total mar- Econometric Measurement of the Effects of Research 197 ginal product of research, wherein sample average values of output and research expenditures are used to form Qili., or even the sample average of the output-to-research expenditure ratio (QIR), which is then used in place of Q,IRt-r or Q/+JR, to calculate -CD -CD ~R'\Q QLR Q TMPQR = TMPQ R =.LJPK"'r-== PK-= LAr= PK-= (3.87) f"/-r /+1'" r=O R R r=O R if the temporal weights Ar are normalized so that L.Ar = 1. 114 The research "benefit" associated with these shortcut procedures is obtained by valuing the approximate total marginal product of research at the sample average output price P so that -..cD -CD Q- VMrQ/l,_r = VMPQ /+,!l, = PK R P (3.88) Because this aggregation does not take explicit account of the time-dated nature of the benefit stream, the economic interpretation of such total marginal products, or how they may be used in a benefit-cost analysis, remain unclear. The commonly employed procedure of using the sample average price of output, P, to value the marginal product of research means that this assump­ tion is embedded in the estimated benefits from research. lIS Of course a variable output price could easily be incorporated into a research-benefit calculation using a production-function approach if P, were used instead of P when valuing the marginal product of research. Effects of a one-shot increase in research: Alternatively, the model can be used to consider, or evaluate, the impact over time of an increase in research spending in a particular year. Equation 3.84 expresses the impact on current output of a unit change in past research, expressed as marginal products. The equivalent representation for the impact of current research on future output is given by (3.89) The total benefit through time involves aggregating these marginal benefits. They can be expressed in value terms by multiplying each marginal physical 114. Using sample values of average output and research expenditures may be appropriate given that the estimated regression coefficient used in these calculations to measure the elasticity of output with respect to the stock of research is itself "averaged" over the sample data. 115. Recall that no explicit assumptions are made about output price when a production function approach is used to estimate the marginal product of research. 198 Econometric Measurement of the Effects of Research product by the corresponding price, and these benefit streams can, in tum, be used in the capital budgeting procedures described in section 5.4.2 to provide summary measures of the effects of research, such as the net present value of research or the internal rate of return to research, for the case of a "one-shot" unit change in research spending. For instance, looking forward from the current year to evaluate a one-shot increase by one dollar in the flow of research spending for the current year t, the formula for the net present value of that change is NPvfi{J(t) = ~ [ VMPQ~rR,(t) ]_ 1 r=0 (1 +pl (3.90) = PK ~ [ A.r Pt+r Qt+r ] _ 1 Rt r=O (1 + p)r where p is the interest rate. This formula can be simplified considerably if we can presume, as above, that the price, quantity, and research variables will be projected as unchanging over the relevant future period using either historical averages or current values. Then (projecting current values, for example) the net present value is simply equal to the present value of the future lag weights, 1:)., /( 1+ p)', multiplied by the constant term PKP,Q,IR,. Simulating alternative scenarios: Algebraic manipulations of the sort shown above yield useful results only when simplifying assumptions, such as constant prices and output and constant research spending (or constant changes in it), are imposed. More generally, when one desires to evaluate the benefits of particular research alternatives, the easiest method is to use the estimated model to project forward or backwards under alternative assump­ tions and then to evaluate the difference in output under the alternative assumptions. For instance, to evaluate a permanent increase of research spending by one unit in every future year, production could be simulated over the indefinite future under two scenarios: first, assuming a base level of spending (e.g., the most recent value) and, second, assuming the base value plus one dollar. The additional output could be valued at the current price or some other projection of price, and the present value of the additional output could be compared with the present value of one dollar per year in perpetuity. Scobie and Eveleens (1987) provide an example of this approach. Two virtues of this method are that (a) it permits projection under a consistent set of ceteris paribus conditions that may vary over time (e.g., prices or fixed factors may vary or the impact of changing the ceteris paribus conditions on measured benefits may be examined) and (b) any set of alternatives that is of Econometric Measurement of the Effects of Research 199 interest (e.g., a one-shot increase or decrease by any amount or letting research be reduced to zero to deduce an average value of research benefits) can be simulated. A translog production function: If a translog production function (such as 3.43) is used to assess the output-enhancing effects of research, then the elasticity of output with respect to the stock of research is simply TL a In Qt m I ~Q.K,= -aKl = ~K+~KKlnKt+L2('Y;K + 'YK;)lnX;,t (3.91) n t ;=1 It follows that the marginal product of Rt-r on output in year t when the technology takes a translog form is MpT.Q'L R = ~TL Qt , '-r Q,K'tR-r (3.92) Summing over the lag distribution (i.e., r =0 , ... ,LR) yields the total marginal product of research. Multiplying the marginal product by the current price gives the current-value marginal product of research, which may be aggregated similarly using the lag weights. Productivity Functions In many studies, a productivity function is estimated instead of a production function. This is usually done for statistical reasons such as a shortage of data or in response to multicollinearity problems in an initial attempt to estimate a production function directly. The principle in adopting the productivity-func­ tion approach is to impose restrictions on the estimation (and by implication on the parameters and the nature of the technology being estimated) in order to obtain better estimates of the research-related coefficients. General approach: Consider the agricultural production function from equation (3.5): (3.93) where, in year t, Q, is output, X, is a vector of conventional inputs, V, is a measure of uncontrolled, random factors (such as weather), and Rt-r and Et-e represent infinite streams of past investments in research and extension (other influences, such as prices and human capital, are left implicit for now). It may be ambitious to attempt to estimate all of the parameters of a 200 Econometric Measurement of the Effects of Research relationship such as equation 3.93 jointly with a single time-series data set, especially in light of the long lags between research and its effects on output. When the main interest is in the relationship between research and extension and output (or when the focus is on productivity per se), degrees of freedom can be saved, and the odds of a successful estimation may be improved by imposing structure on the relationship between conventional inputs and outputs. One option is to assume, say, a Cobb-Douglas relationship between inputs and outputs and use factor shares to deduce the component of output attributable to conventional inputs (as done by Griliches 1963b). A similar principle, but a less restrictive set of implicit assumptions, is involved when an index-number approach is used to approximate an unknown technology. Assuming weak separability, equation 3.93 may be represented as Qt= g(Xt)·h(Rt-r, Et- e, Vt), for r,e, = 0 to 00 (3.94) Then, itis a small step to transform this to a model of total factor productivity (TFP) Qt g(X ) = TFPt = h(Rt- n Et- e, Vt), for r,e, = 0 to 00 (3.95) t Since a desirable (and typical) feature of an input-quantity index is that it is homogeneous of degree one in its component inputs, using a conventional TFP index in an equation such as 3.95 involves an implicit assumption that the technology being approximated is characterized by con~tant returns to scale, as well as separability, in the conventional inputs, X. Deducing research impact: Thus, a TFP index could be constructed, as described earlier in this chapter, and it could be regressed against a distrib­ uted lag of past investments in research and extension. Then, the resulting estimates could be used to deduce the effects of changes in research invest­ ments (marginal or total and one-shot or permanent) on the time pattern of productivity. This could be done algebraically (as described above in the context of production functions) or by numerical simulation. The results could then be evaluated using the capital budgeting methods outlined above. Alston, Pardey and Carter (1994) - hereafter referred to as "APC" - provide an example of applying the numerical simulation version of this approach to estimate the rate of return to public-sector agricultural research in California. Their econometric approach for estimating rates of return to research involves three steps. First, total factor productivity is regressed against measures of research and extension. Next, the estimated parameters are used to simulate th~ stream of total factor productivity (and, hence, output) that would be associated with (a) the actual past stream of research Econometric Measurement of the Effects of Research 201 expenditures and (b) different (hypothetical) past streams of research expen­ ditures. These streams may be used to derive marginal effects (for small decreases in research), or total effects (when simulating output streams with research fixed at zero). In the final step, the simulated differences in the value of output and corresponding changes in the cost of research are used to deduce rates of return. To simplify the estimation problem, APC constructed a preaggregated re­ search variable using the finite trapezoidal lag structure of Huffman and Evenson (1992). In this specification, the research stock variable, K" in year t, is a weighted sum of research expenditures, Rt , over the past 35 years. That is, (3.96) The lag weights are zero for the first three years, they increase linearly up to year nine, then they are constant until year 15, after which they decline linearly to year 35. The weights are normalized so that they sum to one. APe estimated various regression models, with total factor productivity (TFPt in year t) as the dependent variable, and in the preferred model, TFP was a quadratic function of the research stock variable. TFP 2 t = CXo + UK K, + UKK K, (3.97) To analyze the effects of research on productivity, they compared actual production in California with predictions from the model if the research variable had been equal to zero beginning in 1914 (the earliest year of research that affects output in 1949, given a 35-year lag). The value of the additional output attributable to research-induced productivity growth was computed for each year from 1949-1985 using the parameters from the statistical model as A A 2 !l.TFP, (UK K, + U KK K,) GARB, =G VP, TFP, = GVP, TFP, (3.98) That is, the gross annual research benefit in year t, GARBt , is equal to the proportional change in total factor productivity in year t attributable to research, multiplied by the actual value of output (Le., gross value of production, GVP) in year t. To compute a rate of return to research, the stream of benefits, GARB, from 1949 to 1985 is compared with the stream of research expenditures, R, from 1914 to 1985. Since these streams of benefits and costs are in nominal (undeflated) terms, the corresponding internal rate of return is comparable to nominal (Le., observed) interest rates without requiring any adjustment for inflation. The comparison of GARB and R is biased in that research after 1914 had some effects on output between 1914 and 1949 that is not being valued; 202 Econometric Measurement of the Effects of Research also, research up to 1985 has effects that persist for up to 35 years into the future, and benefits beyond 1985 are left out as well. In this sense, the estimated annual internal rate of return - 21.4 percent - is conservatively low. On the other hand, although APC left out benefits for certain years, they also left out costs of extension and private research that might be responsible for some of the measured benefits. A second internal rate of return was computed by adding the costs of extension, E, in every year since 1914 to the costs of research and computing an internal rate of return comparing those total costs (R+E) to the stream of GARB from 1949 to 1985. Adding extension costs reduced the rate of return to 19.1 percent. This still excludes private research costs and spill-in effects. As a crude adjustment, APC also computed rates of return to research ·and to research and extension by repeating these calculations but using half the values for measured benefits (i.e., using GARBI2). The estimated annual internal rates of return were reduced to 19.5 percent (including research costs alone) and 17.1 percent (when the sum of research and extension costs were used). Thus, the rate of return to research was quite insensitive (varying from 17.1 to 21.4 percent) to whether the stream of estimated benefits was cut in half or extension costs were included along with research costs. The explanation for this insensitivity is that timing is all-important. The annual research cost is small relative to the maximum annual benefits, but it is spent over many more years. APC pointed out that a number of additional caveats should be kept in mind when these estimates are interpreted and used, and similar caveats would apply to other studies using similar approaches. First, the analysis involved a large extrapolation (to zero research) from the historical experience reflected in the data used in the statistical model. Such extrapolations are hazardous in that the statistical confidence we can attach to them may be low. The "average" rate of return computed in this fashion is valuable for some comparisons, but it is likely to differ from the "marginal" rate of return that should be used for considering marginal changes in research budgets (as opposed to the effects of elimination of public-sector research for which the average return is pertinent). A marginal analysis was made difficult (if not precluded) by APC's use of a preaggregated research-stock variable with presumed lag weights and a pre­ sumed lag length. This introduces a second major caveat. The assumptions about the lag structure were based on work by Huffman and Evenson (1992) that might not be applicable to California. 116 116. On the other hand, the WOIK by Pardey and Craig (1989) supported the use of long lag lengths­ greater than 30 years - as subsequently asswned by Huffman and Evenson (1992). Econometric Measurement of the Effects of Research 203 Cost Functions The measured effects of research on the cost of production can also be used to value the stream of benefits coming from past investments in research. In this case, the research-benefit stream reflects the marginal, cost-reducing effects of research given a fixed level of output(s), fixed quantities of ''fixed'' factors, and fixed prices of "variable" factors. This contrasts with the produc­ tion-function approach to estimating the benefits from research, where the marginal, output-enhancing effect of research was obtained holding the quan­ tities (not necessarily the prices) of all inputs constant. For the Cobb-Douglas cost function (3.45), the elasticity of cost with respect to the research stock is ):CD OCI KI oln CI A ""C,K,= oK C= :"lInK = I-'K (3.99) I lOt where K" the stock-of-research variable, represents the full set of knowledge­ related variables in equation 3.67. Taking the marginal, cost-saving benefits from changes in the stock of research to be the cost-function counterpart to the research-benefit measure obtained using primal procedures (i.e., using equa­ tion 3.82), then for Cobb-Douglas cost functions, it follows that CD ocfD olncfD CI CD CI CI MBc,x,(t) = - oK = - oinK K =- ~C,K, K = - ~K K (3.100) I I I I I If the translog cost function (3.46) were used, the cost saving from changes in the research stock would be TL o:"I CtT L :"II 0 nCIT L C - t TL Ct MBC,KP) = - oK, = olnK K = - ~c, K, K t t t (3.101) = - ( ~K + ~K Kin Kt + ~ Yi Kin Wi,t + ~KQ InQt J~ ' FI t where symmetry has been imposed on the cross-partial derivatives of the cost function with respect to the knowledge-stock variable, K, and for simplic­ ity, nonresearch contributors to the knowledge stock have been suppressed. Dual analogues to the primal procedures can be used to derive estimates ofthe stream of marginal benefits arising from changes in research expendi­ tures, rather than from changes in the corresponding stock of research. To do this, we substitute 204 Econometric Measurement of the Effects of Research nLRR ?:rr for K, r=() in the respective cost function and develop corresponding measures of the marginal benefits (cost savings) from research investments. For a Cobb­ Douglas cost function, this stream of research benefits is given by dcfD dlncfD dlnKt C MB~~ (t) = - --- - ----=--- ~-- t ,-A: dRt-k - dlnKt dlnRt_r Rt- r (3.102) = _~CD ~ __ A A. Ct -'C,R,_r R t-r - PK r R t-r If the translog cost function (3.46) were used, the cost saving from changes in research would be m MB'[}~,(t) = - ( ~K + ~K Kin Kt + + 1: 'Yi Kin Wi,t i=l (3.103) Ct + ~KQ InQt ) A.r R t-r The cost-saving, research-benefit streams given by 3.102 or 3.103 can then be incorporated into the capital budgeting exercises described in section 5.4.2. The calculations may be simplified by the use of sample average values of production costs, C, and research expenditures, N, to derive approx­ imate measures of the stream of cost-saving benefits from research. As with the production-function approaches, however, only a limited range of ques­ tions can be addressed in that fashion. Greater flexibility is available when the estimated model is used to simulate streams of costs of production under alternative research investment scenarios and then the streams of costs and benefits are evaluated in a second step. Supply Functions The last approach we consider for estimating research impact and evalu­ ating benefits is directly estimated supply functions. The general form of such equations is as set out in equation (3.8): Q,= q(Pt. Wt. Rt- ro Et-e, Zt, Vt), for r,e, = 0 to 00 (3.104) where, in year t, Q, is output of a commodity of interest, P, is a vector of (expected) output prices (of the commodity of interest and related commod- Econometric Measurement of the Effects of Research 205 ities), W, is a vector of (expected) prices of conventional inputs, R'_r and E,_. represent indefinite streams of past investments in research and extension, Z/ represents fixed factors (including human capital for now), and U, repre­ sents uncontrolled factors. In the literature on estimating commodity supply functions, attention has focused to a great extent (following Cassels 1933) on the problems of specifying dynamics and expectations variables. Choices about these aspects of model specification have implications for the use of the estimated supply function for evaluating research benefits. For example, in a model with endogenous prices due to downward-sloping demand, the use of a Nerlovian distributed lag (to represent either dynamics of supply in a partial adjustment framework or adaptive expectations) implies that a research-induced supply shift today will have an impact on production and prices (and research benefits) over the indefinite future. In contrast, in a static model with expected prices equal to actual prices, the same supply shift would have an effect only in the current period. One virtue of the approach of using directly estimated supply functions to measure research benefits, on the other hand, is that it makes the closest possible connection between the statistical model and our conceptual sup­ ply-and-demand model of research benefits. As a consequence, the output from this type of model can be used relatively directly in an economic-sur­ plus model of research benefits. Two alternative approaches suggest themselves. One approach is to use the econometrically estimated supply function to deduce a measure of K for each year to be simulated (to be used in an all-or-nothing evaluation of past research, perhaps); the other is to use the estimated supply function to simulate prices and quantities under alternative scenarios (e.g., under the actual pattern of research investments and under counterfactual alternatives - marginal or total, one-shot or permanent changes from the actual values). Then, the values of the streams of producer (and consumer and, perhaps, taxpayer) surpluses under the actual past stream of investments and the counterfactual alternatives may be evaluated and compared. Finally, as discussed above in relation to production functions, the simulated change in the stream of benefits (the difference between the actual and counterfactual surplus measures) can be compared with the change in the stream of research costs used to define the counterfactual experiment in a benefit-cost analysis. Consider the stylized case of a linear supply function with only the own-price, P" and the research stock, K, = PI Rt_ 1 + ... + PL Rt - L , as its R R arguments, with exogenous prices and no market distortions (notice that this model assumes that actual prices coincide with expected prices). The esti­ mated supply function is given by 206 Econometric Measurement oft he Effects ofR esearch bt = Bo + BI Pt + PI Rt- I + P2 Rt- 2 + ... + PL Rt- L (3.105) R R Assuming that supply is inelastic (recall that the formulas for producer research benefits with a linear supply function depend on whether supply is elastic or inelastic), the producer (and, in this case, total) gain from research in year t is equal to the research-induced change in quantity, AQ" multiplied by the price, Pt. That is, APS, = P ,AQr Thus, all that is necessary for this analysis is to simulate the quantities under the actual research expenditures, R, and some counterfactual alternative of interest (e.g., no research in a particular year or in all years, or a marginal change in a particular year or in all years), R*, and then the stream of changes in producer surplus can be compared and evaluated against the stream of changes in research expendi­ ture (M = R* - R) associated with it. One advantage of the linear model, in this simplest of cases, is that this stream is easy to compute, as shown below, without the intermediate step of simulating quantities having to be gone through: (3.106) Then, the net present value of the change in research spending could be computed using LR LR NPV = ~ APSt_n (1 + rt - ~ M t- n (1 + rt (3.107) n=-LR n=1 It would be a trivial extension to this analysis to accommodate endogenous output prices or other supply shifters, although endogenous prices would require computing changes "in quantities explicitly, as well as a different calculation for producer benefits. And there would be no problem to calcu­ late research benefits in the case of elastic supply, using the formulas presented in appendix A5.2. More serious challenges arise when one goes from the static model to a model with dynamics and expectations. In such a case, it would seem sensible, as a matter of course, to use the estimated model to simulate prices and quantities with and without research, to simulate costs of production as the integral beneath the estimated supply function, and to compute producer surplus for each period as the simulated revenue minus simulated costs of production. While all of this would seem straightforward in principle, it may well be that the effort involved in putting such approaches into practice (compared, say, with a productivity function) can account for the limited use that has been made to date of directly estimated supply functions, with dynam­ ics and expectations embedded in them, for evaluating research benefits. 4 Economic Surplus Methods The concept of economic surplus underlies most of the methods used by economists to estimate the benefits and costs of agricultural research or to assess agricultural research priorities. In this chapter, the introductory mate­ rial presented in chapter 2 is extended to demonstrate how variations on the basic economic surplus approach can be used to model and measure the economic effects of research-induced technical changes in the market set­ tings that commonly confront practitioners. We consider the size and distri­ bution of research benefits in the context of mUltiple factors, multiple product markets, and market distortions. This chapter begins with the basic economic surplus model that considers a single market in a closed economy. Then the model is extended to consider various multi market settings, mainly to disaggregate the measures of bene­ fits that are obtained from the basic model (to allocate the "producer surplus" among individual productive factors as quasi-rents and to allocate consumer surplus among different groups of consumers).l First, a horizontal disaggre- I. If some of the fixed factors are only fixed in the short or intennediate tean, the rents accruing to them are usually called quasi-rents, following a tradition begun by Alfred Marshall. If they are fixed in the long run they are called rents, following a tradition begun by David Ricardo. The tean producer surplus is an unfortunate one because producer surplus represents either quasi-rents or rents to owners of fixed factors and usually are not returns to producers per se except as those producers are owners of the fixed or quasi-fixed factors. The supply curve used for economic surplus analysis in aggregate agriculture is often said to be the long-run supply curve, which slopes upward ,because of variable inputs being applied to a fixed supply of land. In fact, the curve often used is estimated from annual observations on prices and quantities. Therefore, the curve is really an intermediate-run curve and the measured producer surplus is quasi-rent (returns to quasi-fixed factors such as producers' and input suppliers' own labor and fixed capital) as well as rent (returns to land). More important, the supply curves used in research evaluation and priority setting are for individual commodities. Therefore, land is also a quasi-fixed rather than a fixed factor. 207 208 Economic Surplus Methods gation (across different markets for a product or for different producing and consuming groups and across different products) is presented. Next. vertical disaggregation of research benefits (among factors of production or across stages of a multistage production system) is discussed. Then. the effects of market distortions on the size and distribution of research benefits are considered. These distortions include a range of commodity policies and programs (such as trade-distorting policies or domestic programs affecting factors and products). exchange-rate distortions. and finally. externalities. We present this analytical framework as a set of principles that can be applied to a range of situations beyond those specifically considered here. How to collect and use the data and information required to make these models operational is discussed in chapter 5 with specific reference to developing summary measures of the effects of research. Some important simplifying assumptions are retained throughout. First, supply-and-demand curves are assumed to be linear and to shift in parallel as a result of research-induced technical changes.2 Second, a static (single­ period) model is used and dynamic issues are put aside. Third, competitive market clearing is imposed. Fourth, as discussed in chapter 2, Harberger's (1971) "three postulates" are invoked so that standard surplus measures may be used as measures of welfare change. Under these assumptions, for a range of situations, comparative static models of the effects of a research-induced supply shift, are presented. The qualitative effects are shown using supply­ and-demand diagrams, and formulas to compute the effects are presented. The formulas express research-induced surplus changes as functions of technical, market, and policy parameters. 4.1 The Basic Model 4.1.1 Surplus Distribution in the Basic Model The basic model of research benefits in a closed economy is shown in figure 4.1. In this model D represents the demand for a homogeneous 2. Much has been written about the implications of functional forms ofs upply and demand, elasticities of supply and demand, and the natw-e of research-induced supply shifts for the size and distribution of research benefits (e.g., Lindner and Jarrett 1978; Norton and Davis 1981). These arguments are summarized in chapter 2. In relation to total benefits, functional forms and elasticities are relatively unimportant compared with the nature of the supply shift. In relation to the distribution of benefits, functional forms are relatively unimportant compared with the sizes of elasticities and the nature of the supply shift. The assumption ofa parallel shift is very important. Alston and Wohlgenant (1990) have shown that with parallel shifts, the choice of functional form has little effect on either the size or distribution of benefits. See also Voon and Edwards (I991b). Economic SurpLus Methods 209 Figure 4.1: SurpLus distribution in the basic model of research benefits !'lice 51 Po PI d 10 D I I 0 Qo Q Quanlily I product, and So and S, represent, respectively, the supply of the product before and after a research-induced technical change. All curves are defined as flows per unit time, typically annually, as are the economic surplus measures. The initial equilibrium price and quantity are Po and Qo; after the supply shift they are P, and Q,. The total (annual) benefit from the research-induced supply shift is equal to the area beneath the demand curve and between the two supply curves (tlTS = area loab/,). This area can be viewed as the sum of two parts: (a) the cost saving on the original quantity (the area between the two supply curves to the left of QIJ - area loael,) and (b) the economic surplus due to the increment to production and consumption (the triangular area abc, the total value of the increment to consumption - area QoabQ, - less the total cost of the increment to production - area QuebQ,) . Alternatively, we can partition the total benefit into benefits to consumers in the form of the change in consumer surplus (!lCS = area P("pbP,) and benefits to producers in the form ofthe change in producer surplus (!lPS = area P,b/, minus area Poalo). Under the special assumption of a parallel supply shift (where the vertical difference between the two curves is constant), area del, =a rea Poa1o and the change in producer surplus is equal to the net benefit on current production (area P ,ecd) plus the gain on the increment to production from Qo to Q, (area bce) for a total producer surplus gain of area P,bcd. As shown in box 4.1, these effects can be expressed algebraically as follows: 3 3. The text equations are for the case of a parallel supply shift. For a pivotal shift in the basic c1osed-economy case, the formulas for change in total surplus, 6.TS, change in consumer surplus, 6.CS, and 210 Economic Surplus Methods ~CS = Po Qo Z (l + O.5Zrt) (4.1a) ~PS = Po Qo (K - Z)(l + O.5Zrt) (4.1b) !:J.TS = !:J.CS + ~PS = Po Qo K (1 + O.5Zrt) (4.1c) where K is the vertical shift of the supply function expressed as a proportion of the initial price, rt is the absolute value of the elasticity of demand, E is the elasticity of supply, and Z =K tI(E+rt) is the reduction in price, relative to its initial (i.e., preresearch) value, due to the supply shift. 4.1.2 Disaggregating Benefits and Costs The model in figure 4.1 may be used to measure research benefits in terms of supply and demand defined at the farm, retail, or some intermediate stage of the marketing system. The measurement of total benefits is not affected by the choice of where to measure benefits in the marketing chain - the total producer and consumer surplus (or total change in surplus) is the same at all market levels. What is affected by this choice is whose benefits are included in producer surplus and whose are included in consumer surplus.4 When research benefits are measured at the retail level, producer surplus includes quasi-rents to all factors employed in producing the retail product (including marketing, distribution, and processing that takes place beyond the farm level) as well as quasi-rents to farming inputs; consumer surplus measures the surplus of consumers who buy at retail. When research benefits are measured at the farm level, producer surplus includes only the quasi-rents accruing to inputs used in farming; quasi-rents accruing to off-farm processing and market­ ing inputs are included along with final consumer surplus in "consumer surplus" measured at the farm level. Thus, choosing the market level for the analysis implies a choice about the aggregation of owners of producti ve factors and final consumers in the welfare analysis - i.e., the vertical aggregation. Implicit choices about horizontal aggregation of surpluses are also made in the basic model. At a given market level we have aggregated all suppliers together and all demanders together.5 For some purposes, it may be desirable change in producer surplus, MS, are /'"TS =0 .5KPoQo (I + Zl1) /'"cs = ZPoQo (I + 0.5 Z11) MS=/'"TS -/'"CS where K is the proportionate vertical shift down in the supply curve due to a cost reduction. 4. See Just and Hueth (1979), Just, Hueth and Schmitz (1982), and chapter 2. 5. In addition to these choices, some more subtle questions of aggregation arise from choices about whether to use general-<:quilibrium or partial-equilibrium definitions of the supply-and-demand curves. When a general-<:quilibrium definition (allowing forthe feedback effects when other product prices adjust) Economic Surplus Methods 211 BOX 4.1: Algebra/or Research-Benefit Calculation/or the Closed-Economy Case in Figure 4.1 The model: The relative reduction in price is defined as Z = KEI(E+Tl) = - (P, - Po)! Po, where Po and Qo are equilibrium price and quantity before the supply shift, E is the supply elasticity, and Tl is the absolute value of the price elasticity of demand. The equation for Z is obtained by solving linear supply-and-demand equa­ tions for price as a function of slope and intercept parameters, treating a research-in­ duced supply shift as an intercept change, and converting to elasticities: Supply: Qs = a + ~ (P + k) = (a + ~k) + ~P Demand: QD =Y-'6 P where k is the shift down of supply due to a cost saving induced by research. In figure 4.1, k = (Po - d), and the supply shift relative to the initial equilibrium price is K = klPo = (Po-d)IPo' Equilibrium price change: Setting Qs = QD = Q yields the eqUilibrium price P = (y - a - ~k)/(~+'6). When k = 0, Po = (y - a)/(~ + '6); when k = KPo, P, = (y - a - ~KPo) I(~ + '6). The research-induced change in price is (P, - Po) = - ~KP(/(~ + '6) and the absolute value of the relative change in price is given by - (P, - Po)IPo = ~K I(~ + '6). Converting the slopes to elasticities (multiplying through the numerator and denominator by PoIQo) yields Z = KEI(E+Tl) = - (p, - Po)/Po. Consumer surplus: In figure 4.1, the consumer surplus change is given by !1CS = PoabP, = rectangle PoaeP, +triangle abe = (Po - P,)Qo + 0.5(Po - P,)(Q,- Qo) or !1CS = (Po- P,)Qo [I + 0.5(Q,- Qo) IQol Using the definition above that Z =- (P, - Po) I Po so that (Q, - Qo) IQo = ZTl yields !1CS = PoQoZ (I + 0.5ZTl) Producer surplus: The producer surplus change is !1PS = P,b11 - PrP10 = P,bcd + del, - Poalo = Plbcd given that del, = Prplo under the assumptions of a parallel supply shift and linear supply and demand. !1PS = P,bcd = rectangle Plecd + triangle bce = (P I- d )Qo + 0.5(P,- d )(Q, - Qo). Thus, !1PS = (P, - d) Qo [I + 0.5(QI - Qo)1 Qo]. We may define (P,- d) = (Po - d) - (Po - PI) = KPo - ZPo and (QI - Qo) IQo = ZTl. Thus,!1PS = (K - Z )PoQo (I + 0.5ZTl)· Total surplus: Note also that !1TS = !1PS + !1CS = PoabcPI (= Poacd + abc), which in this instance equals Irpbll (= loael, + abc) given that Poacd = loael, under the "laws of parallelograms." to disaggregate suppliers or demanders into subcategories according to geopolitical boundaries (e.g., domestic or foreign), according to income classes, or according to their business characteristics (e.g., small or large farms, adopters or nonadopters of the new technology). A further complica­ tion is that the total research benefit and its distribution may be affected by is used, the surplus measures reflect welfare effects in other corrunodity markets as well as the one being studied. When a partial-equilibrium definition (assuming that other corrunodity prices are constant) is used, some further analysis may be needed to compute the effects of any induced price changes in related markets that feed back into the market of interest. 212 Economic Surplus Methods price-distorting policies and externalities. When government revenues are involved in commodity policy, it can be important to distinguish the effects of new technology on government revenues (or taxpayers) from the effects on consumers, producers, or factor owners. 4.2 Horizontal Market Relationships The basic model refers strictly to the case of a homogeneous product being sold in a single market. Now we consider multiple markets for a single product, multiple products, and possible shifts in demand arising from research-induced quality changes. In section 4.3 we consider "vertical" market relationships - in particular the distribution of benefits among factors of production (or across stages in a multistage production process). 4.2.1 Multiple Markets for a Single Product One type of horizontal market relationship that is often important is the case of an internationally traded good. While some commodities, particu­ larly root crops and several of the main fruits and vegetables, are produced and consumed almost exclusively domestically, most commodities are also either exported or imported. When a country is a large enough producer or consumer of a commodity in the world market that its production or con­ sumption affects world prices, part of the gains or losses from a shift in its supply curve will be realized in other countries. In addition, when research conducted by one country is transferable to other countries, technology spillovers can cause further reductions in the world price, which in turn affect the initial country conducting the research. Even when there is no international trade, there may be significant inter­ regional trade within a country, and therefore, price or technology spillovers among regions within a country may matter. The approaches gi ven below for considering the national implications of research on an internationally traded good are applicable to the regional implications of research on an in­ tranationally traded good or to the intranational implications of research on an internationally traded good. Several studies have developed models for evaluating agricultural re­ search in the context of trade. Some of these studies have assumed that the country conducting the research is a small exporter or importer of the commodity in the world market and thus does not influence world price.6 6. Examples of studies in which the small-country case with international trade has been considered include Alcino and Hayami (1975), Nguyen (1977), de Castro and Schuh (1977), Hertford and Schmitz Economic Surplus Methods 213 Some studies have allowed for price spillovers. Some have allowed for technology spillovers in which research results from one country or region are adopted in another. Others have allowed for both output price effects and spillovers of research results to other countries.7 There are two primary means of modeling technology and price spill­ overs. The first is to develop a commodity model with equations to represent the home country (country A) and equations to represent the rest of the world (ROW) or region B as a group (e.g., Edwards and Freebairn 1981, 1982, 1984). The second is to develop a commodity model with equations for country A and for each of the other major countries in the world - a less aggregative treatment of the ROW (e.g., Davis, Oram and Ryan 1987). Yet another approach is to treat the ROW as an aggregate but to disaggregate the domestic economy into two or more regions. The rationale for the first approach is that an individual country is often concerned with spillovers only to the extent that they affect the total domestic research benefit and, perhaps, its distribution between consumers and produc­ ers, but not the spatial distribution of domestic benefits. It may be easier to obtain an accurate estimate of the aggregate excess-demand curve (elasticity) facing the country than it is to obtain accurate estimates of all the individual supply-and-demand curves in other countries, which are influenced by their domestic and international trade policies and other factors. However, Davis, Oram and Ryan (1987) used secondary and synthetic estimates of demand­ and-supply elasticities for most countries of the world and attempted to measure the extent of technology spillovers by commodity by country. This approach is justified when there is a specific interest in the disaggregated cross-country effects. We begin here with the two-market case and then extend the analysis to the n-market case. The general representation can be interpreted in terms of any geopolitical aggregation - of regions within countries or among countries - or aggregation according to other criteria. Two Markets with No Technology Spillovers Price spillovers occur when a technical change in one country or region has an effect in other countries or regions through effects on the prices of goods traded between the countries or regions. Such price spillovers arise only when the innovating country is a large country in trade (i.e., able to influence international prices for the commodity). To analyze research-in­ duced price spillovers in an excess-supply, excess-demand framework, we (1977), Flores-Moya, Evenson and Hayami (1978), and Norton, Ganoza and Pomareda (1987). 7. Examples include, Martin and Havlicek (1977), Edwards and Freebairn (1982, 1984), Davis, Oram and Ryan (1987), Mullen, Alston and Wohlgenant (1989), and Mullen and Alston (1990). 214 Economic Surplus Methods model the worldwide market in terms of trade between the home country (country A) and all other countries (ROW) so that market clearing is enforced by equating excess supply (the difference between domestic demand and supply) and excess demand (the difference between ROW demand and supply). This is represented in figure 4.2 in which panel a represents supply and demand in the home country (country A) and panel c represents aggre­ gated supply and demand in the ROW (i.e., region B). In the case shown here, the home country is a "large-country" exporter and the ROW is a "large-coun­ try" importer. All of the supply-and-demand curves are assumed to be linear. The excess (export) supply in country A is shown as ESA.o in panel b - given by the horizontal difference between domestic supply (initially SA.O) and demand (initially D A•o). The initial excess (or import) demand from the ROW is shown as EDs.o in panel b - given by the horizontal difference between ROW demand (initially Ds.o) and supply (initially Ss,o)' International market equilibrium is established by the intersection of excess supply and demand at a price Po. The corresponding domestic quantities are shown as consumption, CA .O, production, QA.O' and exports, QTo; the ROW quantities are shown as consumption, Cs.!» production, QB.O, and imports, QTo. Research in the home country causes a parallel shift of domestic supply from SA!.) to SA,l' and in consequence, the excess supply shifts from ESA.o to ESA.I' The new equilibrium price is PI' The corresponding domestic quantities are shown as consumption, CA.I, production, QA.I' and exports, QT,; the ROW quantities are shown as consumption, Cs.I> production, QB.I, and imports, QT,. The research-induced supply shift in country A causes the world price to fall. Consumers in both countries and producers in country A gain, while ROW producers lose. From the domestic standpoint (in panel a), consumer benefits (as measured by the change in consumer surplus) are given by area PoaePI behind the demand curve and the benefits to producers are given by the area P1bcdbehind the supply curve. This area corresponds exactly to the same area (also P,bcd) in figure 4.1 for producer benefits in the case of a nontraded good. From the standpoint of domestic producers, the relevant measure of surplus is unaffected by whether the consumers are domestic or overseas. The determinants of producer benefits in both cases are the size of the research-induced supply shift, the resulting decline in price and the initial output. Indeed, the formula for domestic producer surplus continues to be equation 4.1 b with the interpretation that the relevant demand elasticity is that for total demand (i.e., domestic plus ROW) rather than simply domestic demand. From the ROW perspective (in panel c), consumer benefits are given by the area behind the demand curve, PJgP I , and producer losses are represented by the area behind the supply curve, PohiP,. Box 4.2 shows the algebra for these areas. Figure 4 .2: Size and distribution of research benefits for a traded good (exporter innovates, no technology spillovers, I large country) I v...V..,. (a) Country A production, (b) Excess supply, demand, (c) ROW (region B) production, consumption & trade & trade consumption & trade Price Price Ss,O SA,O ,ESA,O Po PI d EDS , 0 Ds,O I I I I I~QTo~1 I I I 'E I QT. ')1 ~QT o ---+t I *-QT I )' I , I I I o CA,O CA,I QA,O QA,I 0 QTO QTI o QB,O QS, l Ca,O Ca,l Country A Quantity Traded Quantity ROW Quantity 216 Economic Surplus Methods Since both consumers and producers benefit, in the home country national research benefits are unambiguously positive. In the ROW, there are two effects in opposite directions - producers lose but consumers gain. How­ ever, we can see that the consumer gain must be greater than the producer loss. The change in "consumer surplus" measured off the ROW excess demand (area PokmP I in panel b) measures the net ROW benefit (i.e., con­ sumer benefit less producer loss) and is equal to the areafgih in panel c. This diagram, therefore, shows that both countries must benefit from a research­ induced technical change in the exporting country. If the innovating country, country A, is an importer (as shown in figure 4.3), consumers worldwide benefit from the research-induced price decrease BOX 4.2: A 1gebra for Research-Benefit Calculationfor the Open-Economy Case in Figure 4.2 The model: The percentage reduction in price in the case of an open economy is defined as Z = fAK/[fA + SA TlA + (I-SA) Tl~] = - (P I- Po)/Po where SA is the fraction of production consumed domestically (i.e., in country A), fA is the domestic supply elasticity, TlA is the absolute value of the elasticity of domestic demand, and Tl~ is the absolute value of the elasticity of export demand (i.e., the ROW, or region B, excess demand), all defined at the initial equilibrium, and other variables are as previously defined. This price reduction is obtained by solving linear supply-and-demand equations for price as a function of slope and intercept parameters, treating a domestic research­ induced supply shift as an intercept shift, and converting the following to elasticities: Domestic supply: QA = a A + ~A (P+k) = (aA + ~Ak) + ~AP Domestic demand: CA = YA - 0AP ROW supply: QB = a B + ~BP ROW demand: CB = YB - 0BP where k is the shift down of supply due to a cost-saving technological change. Equilibrium prices: We can solve for the equilibrium price by setting QA + QB = C A + CB to obtain P = (YA + YB - aA - a B- ~Ak)/(~A + YA + ~B + 0B) When k = 0, P = Po = (YA + YB - a A - aB)/(~A + 0A + ~B + 0B)' When k = KPo' P = PI = (YA +YB - a A - a B - ~A KPO)/(~A + 0A + ~B + 0B)' Thus, the change in price is PI - Po = - ~AKP(/(~A + 0A + ~B + 0B)' and the absolute value of the relative change in price is Z=-(PI-Po)/Po=~AK/ (~A + 0A + ~B + 0B)' (continued on next page) Economic Surplus Methods 217 Box 4.2: (continued) Multiplying through the numerator and denominator by PO/QA.O and converting to elasticities yields Z= - (PI-PO )IPo= (I3AK ) (Pc/QA.o)/[(I3A + OA + I3B +OB) (PC/QAO)] = f AKI[I3A (PI/QA.O) + OA (PI/CAO) (CA.1/QA.O) + (I3B + 0B) . [PI/(CB.O- QB.O)]· [(QA.O- CA.1/QA.O] = fAKI[ fA + SA llA + (I - SA )ll~ ] where we have defined the elasticity of the ROW's excess-demand curve as ll~ = (I3B + 0B) P1/(CB.O- QB,O) and used the fact that the traded quantity, QTo = CB,o - QB,O = QA,O - CAO' Welfare effects: In figure 4,2, the domestic consumer surplus change is given by ,1,.CSA = PoaePI in panel a. By analogy with the closed-economy case in figure 4.1, the domestic consumer surplus change is ,1,.CS A= PoCA .l7J..1 + O.5ZllA), Similarly, also by analogy with the closed-economy case, the domestic producer surplus change is ,1,.PSA = Plbcd = PoQA,o(K - Z)(l + O.S2£A)' The ROW surplus change is equal to area PI~mPI in panel b, and this area is equal to PoQ1(,z(1 + O.SZllB) by analogy with the closed-economy case. The same answer for welfare effects could have been obtained by substituting the relative price change, Z, into equations 4.4a and 4.5b after substituting for E(Pd,i) = E(P,) = Z for i = A or B and defining E(Qd) = lli Z and E(Q.,.A) = fA (K - Z) and E(Q.,.B) = fB Z. The disaggregated ROW welfare effects would be ,1,.CSB = POCB.O Z (I + O.SZllB) = area PofgP I in panel c of figure 4,2 and MSB = POQB.O Z (1 + 0,52£B) = - area PohiP,. The sum of these two effects (Le" the difference between the two areas) is ,1,.TSB = ,1,.CSB + ,1,.PSB = P1lCB.O- QB,O) z11 + 0.5 Z (llB CB.C/[CB,o - QB,O] - fBQB,o/[CB.O- QB.oD) = Po QTo Z (1 + 0.5 Z ll~ ) and therefore this area is equal to area PokmP, in panel b of figure 4.2. (domestic consumers gain Poi}P, in panel a and ROW consumers gain PoklP, in panel c), producers in country A gain (area P,bcd in panel a) and ROW producers lose (area PJhP I in panel c). Once again, the innovating country unambiguously gains. In contrast with the previous case, the ROW loses because the loss to ROW producers (area PufhP, in panel c) exceeds the benefit to ROW consumers (area PoklP, in panel c). The net ROW loss is shown as the area PoegP, in panel b (which equals area PcjhP, minus area PoklP, = area kjhl in panel c) of figure 4.3. Figure 4.3: Size and distribution of research benefits for a traded good (importer innovates, no technology spillovers, N large country) "- 00 (a) Country A production, (b) Excess supply, demand (c) ROW (region B) production, consumption & trade & trade consumption & trade Price Price Price DB,O SB,O ESB ,O d i 7" _E D!... :.....---7QT.O~ : " , ' ' :• '-QT1 4: I ,I ,I ;+-QTl-: o QA,O QA,1 CA,O CA,1 o QTl QTo o CB,OCB,I QB,I QB,O Country A Quantity Traded Quantity ROW Quantity Economic Surplus Methods 219 Technology Spillovers Technology spillovers arise when some parts of the ROW are able to adopt the results from country A's research. Thus the research-induced supply shift in country A is accompanied by a supply shift in the ROW. The ROW supply shift is likely to be smaller than country A's because a given country's research results are likely to be less applicable elsewhere, especially results from relatively applied research. Figure 4.4 duplicates figure 4.2 (the case where country A is a "large country" exporter) with adjustments to reflect spillovers of technology from country A to the ROW. In figure 4.4, all of the curves from figure 4.2 are as previously defined, but there are two extra curves (SB.I and EDB•1). In panel c, SB,I represents ROW (i.e., region B) supply after a research-spillover-induced supply shift. Cor­ respondingly, in panel b, there is a reduction in excess demand facing country A from EDB,o to EDB.I' Thus, the international spillover of technol­ ogy augments the initial effect, further depressing the world price from PI to P2, and in consequence, consumer benefits in both countries are greater. In addition, producer benefits in country A are smaller and ROW producer losses are reduced and perhaps even turned into a gain. The ROW unambiguously benefits from both the initial supply shift in country A and the spillover supply shift in the ROW. The benefits to country A are reduced by the spillover. One way to see this is to view the initial supply shift in country A and the ROW supply shift as independent events. The initial (own) supply shift in the exporting country yields a net domestic benefit (see figure 4.2). The subsequent supply shift in the (foreign) import­ ing country imposes a cost in the exporting country because it leads to a reduction in demand for exports from which the loss to producers is greater than the benefit to consumers in the exporting country. Nonetheless, country A still benefits overall from the research-induced supply shift. Indeed, country A producers unambiguously benefit so long as the overall price reduction (from Po to P2) is smaller than the initial vertical supply shift in country A, and that will be so even in the extreme case when the ROW supply function shifts by the same amount (i.e., when the technology is fully transferable). As the diagram in figure 4.4 is drawn, ROW producers are net losers, even with some adoption of country A's research results (i.e., area P2ij is less than area Pr/lk in panel c), but in general they could gain or lose. The benefits to producers in country A are shown as the area P2bcd in panel a; the benefits to consumers in country A and the ROW are shown as area PoaePz in panel a and area PJgPz in panel c, respectively. The case of research spillovers when country A is an importer can be illustrated by including a research-induced supply shift in the ROW in figure Figure 4.4: Size and distribution of research benefits for a traded good (exporter innovates, with technology spillovers, N large country) aN (a) Country A production, (b) Excess supply, demand, (c) ROW (region B) production, consumption & trade & trade consumption & trade Price Price Price SB,O SA,O SB, I Po P2 d j 1 1 1 1 1/1 1 ~QTo~ 1 r 1 1 ':..!..-- QT 1 1 ~QTo~ 1 I"') - 1---;--+1 1 L.QT 1 )1 1 r- I· o CA,O CA,l QA,O QA,I 0 QTOQTI o QB, I QB,O CB,OCB,I Country A Quantity Traded Quantity ROW Quantity Economic Surplus Methods 221 4.3. As an importer, country A benefits from transfers of its research results to the ROW, although research benefits to domestic producers are reduced as a result of the price-depressing effect of overseas adoption of those results. A general method of deriving the formulas for calculating economic surplus changes in the case of technology spillovers is presented in the next section. The Case oin Countries or Regions The above analysis has dealt with two regions with research-induced technical change, allowing the possibility of interregional technology spill­ overs. More generally, an arbitrary number of groups of suppliers and consumers can be considered with market-clearing conditions imposed on quantities and prices. This corresponds to the approach of Davis, Oram and Ryan (1987). In the context of multiple regions of a nation, technology may be better regarded as being specific to agroecological zones rather than to geopolitical regions. With this in mind, the formulation of technology spill­ overs can be recast to reflect multiple zones adopting new technologies to a greater or lesser extent, depending on their natural endowments of resources and climate, where the potential impact (i.e., the local supply shift or K effect) of the new technologies also varies spatially. Thus, supply-shift variables among regions may be related, either because technology spills over from one region to another or because technology becomes available to all regions but is not equally applicable, and also not as readily adopted, in all regions. The following model assumes that total quantity demanded and total quantity supplied are equal and prices are set competitively with zero transport costs among markets. 8 Supply: Q. .. ,; = f; (Ps,i, B;) (4.2a) Demand: (4.2b) n n Market clearing: L Q.,,; = L Qdj (4.2c) ;= I j= I Ps ,; = Pd,; = P"j = Pdj = P for all i and j (4.2d) where Qs,; is the quantity supplied, Pr,; is the supply price, and B; is a supply-shift variable for the ith group of suppliers or producers; QdJ is the quantity demanded, PdJ is the demand price, andAj is a demand-shift variable for the jth group of demanders or consumers. Equation 4.2d is the market- 8. Market-distorting policies are easily incorporated in this model as ad valorem taxes or subsidies. For example, see Alston (1991) and chapter 5. 222 Economic Surplus Methods clearing price condition that could be adjusted to reflect price wedges due to policies or transportation costs. This model can be solved either by using specific functional forms for the supply-and-demand equations or by taking a logarithmic differential approx­ imation.9 When a change in technology causes a small shift from an initial equilibrium, changes in prices and quantities may be approximated linearly by totally differentiating equations 4.2a through 4.2d and converting them to elasticity form. Differentiating throughout and adding exogenous shocks yields the following system of equations - expressed in terms of relative changes and elasticities: 10 Supply: E (Qs,i) = ti[E (P"',i)+ 13;] (4.2a') Demand: (4.2b') n n Market change: L ss;E (Q.,,;) = L dsjE (Qdj} (4 .2c') i= I j= I E (PdJ) = E (Pd.i) = E (Ps,i) = E (P.,) = E (P) (4.2d') for all i and j where E denotes relative changes (i.e., E(Z) = dZ'Z = dlnZ), 'Ilj is the absolute value of the elasticity of demand, and cxj is a vertical shift upwards in the jth demand function (an increase in demand relative to the initial equilibrium price); ti is the elasticity of supply and l3i is a vertical shift down in the ith supply function (reflecting an increase in supply, measured relative to the initial equilibrium supply price).11 In equation 4.2c' the share-weighted sum of relative changes in quantities supplied equals the share-weighted sum of relative changes in quantities demanded, where the ith supply share is sS; = Q,./ ( Li Q"i ) and the jth demand share is dSj = Qd/ ( Lj QdJ ). The exogenous shift parameters (cxj , 13;) express equilibrium displace­ ments relative to an initial equilibrium. Thus, for instance, setting cxj =0 .1 would imply a 10 percent increase in the jth group of consumers' willingness to pay for the initial quantity of the product. While the shift of demand is expressed as a fraction of the initial price, it cannot be presumed that there 9. Edwards and Freebairn (1981, 1982, 1984) and Davis. Oram and Ryan (1987) assumed linear supply and demand. for example. Alston and Wohlgenant (1990) have shown that the logarithmic differential (linear elasticity) approximation is good for small changes with constant elasticity supply and demand and is exactly correct with linear supply and demand. 10. Perrin and Scobie (1981) use this type of horizontal multimarket model to analyze Colombian food policy. They present general algebraic solutions as well as numerical results. II. Thus~; corresponds to K;. the proportional research-induced shift in the price direction of the ith supply function. Economic Surplus Methods 223 has been a proportional shift of demand. Rather, uj measures the vertical shift in the jth demand, gj, at a point, locally, for any type of demand shift (e.g., proportional, parallel, or pivotal). Similarly, ~i measures the shift down of the ith supply, (, with the magnitude of the reduction in marginal cost (at the point of approximation, the initial equilibrium) being expressed relative to the initial price of the good. In this specification nothing is presumed about the magnitude of the supply (or demand) shift at other points along the supply (or demand) curve. The nature of the shift (e.g., proportional, parallel, or pivotal) is treated as a separate question from the amount of the shift relative to the initial equilib­ rium. For the most part, parallel shifts of supply and demand are assumed but those shifts are expressed relative to initial prices and quantities. Further, for the most part it is assumed that all supply-and-demand curves are linear - at least in the relevant range of the equilibrium displacement. But these assumptions are not necessarily implied by the specification of the equations of the equilibrium-displacement model. At the same time, they are wholly consistent with the equations of the model when we make it clear that we are using the model to approximate the consequences of parallel displacements of linear supply-and-demand equations. The assumptions of (approximately) linear supply-and-demand functions with parallel shifts are required for the economic surplus measures that are used below. In cases where different assumptions are made about the func­ tional forms or nature of the supply-and-demand shifts, it may still be convenient to use the linear-elasticity equilibrium-displacement model to estimate changes in prices and quantities - and for small shifts, it is likely to be a good approximation. However, the surplus formulas below are correct only for parallel shifts of linear supply and demand; nonparallel shifts require different equations (albeit only slightly different in many cases) to compute changes in economic surpluses. The system of equations 4.2a' through 4.2d' can be solved for the endog­ enous relative changes in prices and quantities as functions of the elasticities of supply and demand, shares, and exogenous shift variables (see chapter 5). The solution for the relative change in price is l2 E(P) = L.i (dsi Ui 11i - SSi ~i Ci) (4.3) L.i (dsi 11i + SSi Ci ) This price equation shows how a research-induced increase in supply in one region (~i > 0 for region i) depresses the price in all regions. The extent of 12. In some situations it might be helpful to note that Lr~S/Ei =E wand L/dsiTJi =T Jw. where Ewand TJw are the aggregate (world) elasticities of supply and demand. 224 Economic Surplus Methods the effect depends on the size of the shift, the relative importance of that region, SSj, in world production of the commodity, and the elasticities of supply and demand. Alternatively, a research-induced increase in demand (perhaps as a reflection of improvements in processing technology) in region i (u;> 0) will increase price in all regions. Again, the price increase depends on the size of the shift and the share of that region in total consumption (i.e., dsJ Thus we have a relatively general representation of price spillovers in a model that allows for simultaneous, independent supply-and-demand shifts in any of the n regions. This is one type of spillover: the pecuniary effect when innovation by one group of producers affects prices received by another group of producers. The other type of spillover occurs when technology developed by one country (or group of producers) is adopted elsewhere. In the context of the model given above this "leakage" of research results can be analyzed by treating the supply shifters, the ~jS, as mutually dependent so that there is a supply shift in one region as a consequence of a supply shift in another region. Specifically, when an innovation in region i partially leaks out to other regions, we might define the shift parameters in the other regions as ~j =9 jiPiwhere9jiis the supply shift inregionj as a fraction of the shift in region i. 13 Similarly, a technical change that causes a supply shift and also involves a concomitant demand shift (say, due to a quality change) can be modeled by relating a demand shift to the research-induced supply shift. Once the situation of interest has been parameterized (by defining the nature of supply and/or demand shifts and the elasticities and market shares), equation 4.3 can be used to calculate the relative change in the product price. Effects on quantities may then be obtained by substituting this result into the country-specific supply-and-demand equations (in 4.2a' and 4.2b'). While all regions in this model experience identical changes in equilibrium prices, changes in quantities supplied may vary regionally because of differences in local supply-and-demand characteristics (as reflected in local elasticities of 13. For instance, suppose a technology developed by country I that reduces costs by one percent is adopted at the same time by country 2 but is not adopted by country 3. This could be analyzed by setting the shift parameters as 131 = I3:z = 0.01 and 133 = o. The qualitative results are that "leakage" of results will increase global total benefits, increase global producer benefits in total, increase global consumer benefits in total, and reduce benefits to producers in country I and country 3, relative to the case where only country I's costs are reduced. Country 2 clearly benefits from leakage. International leakage of research results from country I to country 2 may involve a net benefit or a net cost in countries I and 3. This is an empirical question, the answer to which will depend on the characteristics ofthe rn:uXets and trade in the affected commodity. More generally, for n countries, one can think of an n x n spillover matrix containing coefficients, 9ij' that are used as multipliers to indicate the extent of spillovers of supply shifts from country i to country j. The diagonal elements, 9ii, represent the own-country effects, and the off-diagonal elements, i1'j, represent international spillovers. This matrix need not be symmetric. Economic Surplus Methods 225 supply and demand) and the size of localized shifts in supply and/or demand. These changes in quantities and price, along with the original information on the supply-and-demand shifts, are sufficient to calculate the full welfare consequences of the equilibrium displacements. For large numbers of dis­ tinct groups of suppliers and demanders, particularly with price wedges in the model, it might not be sensible to try to obtain analytic solutions to the system of equations 4.2', but numerical solutions are possible.14 Gross annual research benefits accruing to the various groups of producers and consumers may be computed using !!.CSi = -Pd.i Qd,i [E (Pd) - ui ][1 + O.5E (Qd) ] (4.4a) !!.PSj = p'J Q'J [ E (P,,) + ~j ][ 1 + O.SE (Q,) ] (4.4b) (4.4c) where subscript d denotes demand prices and quantities and s denotes supply prices and quantities, subscript i denotes the different groups of consumers, and j the different groups of producers. Equation 4.4c measures global benefits. 15 Benefits to any subaggregate of consumers and producers can be computed using the relevant components from equations 4.4a and 4.4b. For instance, benefits to "country" i could be computed as !!.TS; = !!.PSi + !!.CS;. Suppose the subscripts "i" denote different countries. In the above model, a supply (or demand) shift in anyone country will affect the price, quantity, and economic surpluses in every other country. Thus research-induced technical changes in one country have effects that are not confined to the country where the innovation takes place. In this model it is easy to show the intuitively reasonable result that when a subset of producers adopt an im­ proved technology, all consumers benefit, while those producers who do not adopt the innovation lose (e.g., see Edwards and Freebairn 1982, 1984). The multiple market (or trade) model described here is essentially the same as those of Edwards and Freebairn (1981, 1982, 1984) and Davis, Oram and Ryan (1987) - i.e., it deals with a single commodity, research shifts a linear supply function down in parallel, and there may be spillover effects in other supply functions. There is one important difference, how­ ever: it is possible to combine this multi market model with the multifactor model described in the next section so that we can consider multiple markets (i.e., international trade) and multiple factors (i.e., multistage production) 14. The algebra and its solutions can become quite messy and complicated and the results may not be clear, in which case the value in obtaining analytic solutions is not clear. 15. If there were any tax wedges we would need to augment the measures with changes in government revenues as weU. 226 Economic Surplus Methods jointly and in a theoretically consistent manner. This was done, for example, by Mullen, Alston and Wohlgenant (1989) in their study of the international incidence of the benefits and costs of Australian investments in research applicable to various stages of the world wool industry, allowing for both technology and price spillovers. Special Cases The above model provides general results for the case of interregional trade among any number of regions under conditions of competitive market clear­ ing and zero transport costs (and no other price wedges). Using that model, we can measure the size and distribution of economic welfare effects (along with price, quantity, and trade effects), allowing for spillovers of prices and technology. It would be relatively simple to introduce price wedges into the model. Most studies do not use a model as general as the one shown above for n regions (even with the restriction of costless arbitrage so that prices are equal among regions). Typical alternatives can be represented as special cases of this model, however, and the welfare effects of research-induced supply shifts may be measured using the surplus formulas 4.4a, 4.4b, and 4.4c. In a common example, the innovating country, A, is a small country in trade. This can be defined by taking the limit of equation 4.3: as the supply share, SSi' of the innovating country i approaches zero so does E(P). The "small country" model takes as an extreme approximation that E(P) = O. Then, setting all of the other demand-and-supply shift variables at zero (i.e., assuming an absence of locally generated or spill-in technical change else­ where and no change in demand structures), the welfare consequences of innovation in country A are given by substituting ~A into equations 4.4a through 4.4c. In this case, all of the research benefits accrue to country A's producers regardless of whether the country is an exporter or an importer. The small-country assumption is often appropriate - most agricultural products are tradable and most regions or countries do not influence interna­ tional prices significantly. The impact of research on a small-country impor­ ter or exporter of a commodity is illustrated in figure 4.5 (panel a represents the case of a small exporter and panel b represents a small importer). The initial equilibrium is defined by consumption, Co, and production, Qo, at the world market price, Pw, with a traded quantity, QTo (representing exports or imports), equal to the magnitude of the difference between consumption and production. Research causes supply to shift from So to S] and production to increase to Q]. As a result, exports increase (or imports decrease) to QT]. Because the country does not affect the world price, the economic surplus change (equal to area !r,abl]) is all producer surplus. One advantage of the Economic Surplus Methods 227 Figure 4.5: Research benefits in a small open economy (a) An exponer (b) An imponer Price Price So Sl Pw 10 I I o I :'---;""'-QTo-': I II I I I I I I I :.- QTIr-: :---QTI -----+: I I :+-- QT I -:--------+: a CO ~I QI Quantity 0 Q I Quantity small-country assumption is that, even when the government intervenes in the commodity market (see section 4.4 below), all the research benefits continue to accrue domestically and there is no need to consider the ROW (unless there is leakage of research results and consequent price spillovers feeding back from the ROW). The world price, Pw, is a constant in the analysis and defines the opportunity cost of resources used in production and consumption. In this case of a small open economy, the formula for research benefits from a K percent (where K = ~A) parallel shift down of supply is simply (4.Sa) This equation can be obtained by taking the limit of equation 4.1b or 4.4b (where equation 4.2a' has been substituted for E[Q'JD as the demand elastici­ ty approaches infinity (and E[P'J] goes to zero). It applies for both an exporter and an importer. When we allow for trade in the product, we may also have to consider the possibility of technology spillovers, as discussed above in a more general context. Technology spillovers are relatively simple to incorporate as (4.Sb) where ZB is the proportional change in the world price due to the technology spillover, which may be calculated as ZB = KB £B /(£B + l1B)' The subscript B denotes ROW (i.e. , region B) coefficients corresponding to the domestic 228 Economic Surplus Methods coefficients defined in equations 4.1 a through 4.1 c, and KB denotes the ROW supply shift due to the leakage of the technology that caused the K propor­ tional home-country supply shift. We can also see the price change in a closed economy (the basic model) as a special case of equation 4.3. In this case there is only one region, and we can drop the subscripts and use supply-and-demand shares (SSj and ds) equal to one. Making these simplifications in equation 4.3, a research-induced supply shift of ~A causes a price change of E(P) = ~AEI(E+ll). This price change is exactly that defined above in relation to the basic model (using K instead of ~A)' Furthermore, substitution of this measure of the price change into equations 4.4a, 4.4b, and 4.4c will yield precisely 4.1a, 4.1b, and 4.1c. Finally, the two-sector case (country A and the ROW) analysed in detail above is easily dealt with as a special case of the n-sector framework. Thus, the equations of the model can be used to measure the areas of surplus changes shown on figures 4.2, 4.3, and 4.4. 4.2.2 Disaggregating Consumer and Producer Surplus So far we have shown how to disaggregate welfare among a range of producer and consumer groups. In the discussion above, in the context of traded goods, the interpretation was that the groups differed according to where they lived. More generally, the multi group model- as defined above in equations 4.2a' to 4.2d' - could be interpreted with groups defined according to any criterion. For example, the groups of consumers could be various categories of domestic consumers (e.g., according to income class) in a closed-economy model. Elasticities of demand that vary across income classes could be incorporated, and thus the differential impacts of technical change on the various groups could be identified. Similarly, producer cate­ gories of various types could be included (e.g., large versus small farms, adopters and nonadopters). The main limitation on such analyses is likely to be the availability of meaningful estimates of parameters and data to describe the groups of interest. Here, we introduce some special examples of disag­ gregating consumer and producer surplus according to characteristics other than regional location. Household Consumption In many countries, a significant proportion of food production is con­ sumed in the farm household where it is produced, with the proportion varying by commodity, location, and farm size. When producers of a good consume significant quantities of the good they produce, producer surplus Economic Surplus Methods 229 alone will be an incomplete measure of the welfare consequences of price changes for producers. A more complete measure would augment their producer surplus with their consumer surplus from consuming their own product. In the context of the model above, this could be done by having one group of producers and two groups of consumers (i.e., producers and non­ producers) with producer welfare being measured by their consumer and producer surplus combined. The remaining consumer surplus would accrue to consumers who are not also producers. In a developing-country context, household consumption of home-pro­ duced goods tends not to be very responsive to price and, therefore, has been represented in several studies with a perfectly inelastic demand (e.g., Haya­ mi and Herdt 1977; Nagy 1984; Norton, Ganoza and Pomareda 1987). Thus the consumer surplus benefit to producers is equal simply to the research-in­ duced price change multiplied by the quantity consumed by producers. The change in total economic surplus is the same as before but part of the consumer benefits remains with producers. This highly simplified approxi­ mation of the benefits to home consumption could be refined by assessing more accurately the price elasticity of demand for home consumption. Benefits Among Producer Groups The impact of technological change on the distribution of income among producer groups can be assessed in many dimensions (corresponding to several of the distributional objectives described earlier). Producers at dif­ ferent income levels, with different farm sizes, in different locations, and with diverse tenure situations can gain or lose depending on the suitability of the new technology to their particular situations. The supply curve can be disaggregated to allow measurement of these distributional consequences within the economic surplus framework (e.g., Hayami and Herdt 1977; Binswanger 1980). The n-sector model above could be reinterpreted to represent groups of producers according to some characteristic other than geographic location (i.e., adopters or nonadopters, small or large farms) and the surplus formulas could be applied without any modification. A related type of analysis can be used to show the distributional effects of technological change on the relative incomes of landlords and tenants. If the tenant pays a fixed amount to the landlord, the tenant's producer surplus is always reduced by the amount of rent paid to the landlord (which may represent all the producer surplus on marketed production). After the techni­ cal change, the tenant can earn rents (producer surplus) on the land being farmed until the landlord raises the rent charged for the use of the land. If the tenant pays the landlord a fixed share of the output, the division of any 230 Economic Surplus Methods producer surplus depends on the relative sharing of production costs as well. If all output and costs were shared equally, they would each earn the same producer surplus but the amount would be one-half that earned by an owner-operator. Of course these are ad hoc approaches and it would be better to analyze the functional distribution of income between landlords and tenants using the methods described in section 4.3 for analyzing surplus distribution among factors of production in vertically related markets. Benefits among Consumer Groups The relative benefits by consumer income groups can be calculated by disaggregating the market demand curve into demand curves by income groups, as illustrated by Pinstrup-Andersen, Ruiz de Londono and Hoover (1976), Pinstrup-Andersen (1977), and Perrin and Scobie (1981).16 Separate price elasticities of demand could be obtained for each consumer group and then applied to calculate benefits by group. A rougher but quicker method for apportioning total consumer benefits is to distribute them across con­ sumer groups in proportion to the quantity consumed initially by each group.17 Similarly, consumer benefits can be dis aggregated by region and combined with the regional benefits to producers to arrive at an estimate of total research benefits by region. As an approximation, benefits may be distributed across consumer groups in proportion to the quantity consumed initially by each group. This is equivalent to assuming equal demand elastici­ ties among the groups. This approach was used by Hayami and Herdt (1977) and Scobie and Posada (1977, 1978) and was criticized by Pinstrup-Ander­ sen (1977, p. 27). 4.2.3 Multiple Products: Some General Issues Types of Multiproduct Situations A few studies have dealt formally with the incidence of research benefits and costs among industries (e.g., Coble et al. 1992). Changes in technology in one industry (affecting one commodity) will affect industries producing different commodities that are related either in consumption or production to the one where the innovation occurs. For example, an improvement in technology in the chicken meat industry will result in a shift in the supply of 16. Scobie (1980) pointed out and corrected an error by Pinstrup-Andersen (1977). 17. Scobie and Posada (1978, pp. 88-89) followed a similar procedure in apportioning consumer benefits from rice research in proportion to the consumption by various income groups. Pinstrup-Andersen, Ruiz de Londono and Hoover (1976) conduct a more detailed analysis in which they estimate separate price characteristics for five consumer income strata. Economic Surplus Methods 231 chicken and a reduction in its price. A second-round effect of this change will be a reduction in demand for beef (a substitute in consumption) and, at the same time, a reduction in the supply of beef due to increased feed-grain costs (assuming the change in the technology was not feed saving) because beef and chicken compete for feed grains. Alternatively, as a second example, improvements in the technology of feed-grain production will lower costs for both chicken and beef, but the effects may differ in size and the net effects on either may be unfavorable because of substitution in consumption. A third alternative multiproduct situation is where the two goods are directly related in production (rather than through a shared factor), such as in the Australian sheep wool and meat industries or in the dairy-processing industry that uses milk to produce a range of products. In these two examples the various products will be related in three ways: (a) substitution in con­ sumption (say, between butter and cheese), (b) complementarity in produc­ tion (say, between butter and skim-milk powder), and (c) competition (or substitution) between products in the use of specialized factors (say, between using milk to produce butter and skim-milk powder and using milk to produce cheese). A change in technology may be of the kind that affects joint product relationships or the kind that affects factor use (neutral or factor-bi­ ased technical changes), as considered below. Particular situations might involve one or more of these types of interac­ tions among products. In the humid tropics, for example, multiple cropping is common. Technical change that affects one crop directly may affect another through its effects on resource requirements and growing season, through effects on soil structure and fertility and pests and diseases, and through substitution in consumption. General-Equilibrium Feedback and Double Counting Welfare measures taken in the context of a single-market model may reflect the welfare changes in related markets. Consider the first example, an improvement in the technology of chicken production. A conventional par­ tial-equilibrium analysis of welfare changes in the chicken market will reflect induced shifts in demand for beef but will still be correct so long as the beef price is exogenous. It will be incorrect, however, if the beef price is affected, and it will be complicated further when there is feedback from the beef market into supply or demand in the chicken market. A natural impulse would be to add up the areas of welfare changes in the beef and chicken markets, but adding up effects across markets could involve double counting. Alternatively, a general-equilibrium definition of the supply-and-demand equations for chicken could be used. The normal (ceteris paribus) Marshall- 232 Economic Surplus Methods ian demand curve shows how consumption of a good responds to changes in its own price, holding all other prices and money income constant. A general-equilibrium (mutatis mutandis) demand function shows how con­ sumption of a good responds to changes in its own price, allowing prices of related goods to adjust in response to the own-price changes and allowing the ceteris paribus demand for the good to shift in response to these induced changes in prices of related goods. The elasticity of this type of general-equi­ librium demand curve corresponds to the "total elasticity" concept intro­ duced by Buse (1958).18 The supply-side counterpart allows for induced changes in the prices of related products to feed back into the supply curve of interest. The welfare measures taken off those supply-and-demand equations will reflect welfare changes in the beef market (and any other related markets) as well as the welfare of the participants in the chicken market; the single-market model can measure the full effects. Just, Rueth and Schmitz (1982, p. 192) put it succinctly: "net social welfare effects over the economy as a whole of intervention in any single market can be measured completely in that market using equilibrium supply and demand curves of sufficient generality." There are two correct ways to measure welfare effects when there are multiple price changes and cross-price effects induced by a supply shift in one market. The first is to add up effects across markets using the welfare areas measured off ceteris paribus supply-and-demand curves in all of the affected markets, all of which may shift as a consequence of an exogenous supply (or demand) shift; it will be correct (and path-independent) only if integrability conditions are met. The second is to use the mutatis mutandis supply-and-demand curves for the commodity of interest, in which case there are no endogenous shifts of any of the curves, and there is no need to add up across related markets. When a "general equilibrium" model is used to measure the quantity and price changes caused by a supply shift in one market, those measured responses will reflect feedback from related mar­ kets. If those responses were used to compute welfare changes, it would be double counting to add them up across markets. The main point is that it is important to be aware of the dangers of double counting (and partial count­ ing) and to ensure that the welfare measures are calculated in a fashion that is consistent with the structure of the model. 18. The "total" elasticity is equal to the Marshallian elasticity obtained holding the price of other goods constant, dlnQ; I dinP;, plus a tenn reflecting the effect of induced cross-price changes and their effects on the demand for the good in question: dlnQ; _ dlnQ; + dlnPj dlnQ; dlnP; - dlnP; dlnP; dInPj Economic Surplus Methods 233 Correct Measures o/Welfare Change Two difficult questions remain. First, how can the effects among the related commodity markets be disentangled and, from there, apportioned among factor suppliers and consumers? Second, what can be done when two or more exogenous displacements occur simultaneously? Some recent work provides a partial answer to the first question: Thurman (1991a, 1991b) explores the welfare significance (and nonsignificance) of general-equilibrium supply-and-demand curves. He considers two goods that may be related through substitution in consumption, substitution in production, or both. He verifies the results of Just, Hueth and Schmitz (1982, p. 192) that, if care is taken, the total welfare effects in both markets can be measured in the context of a single market. In addition, he points out that the areas behind the general-equilibrium supply-and-demand curves for a com­ modity may have no welfare significance taken separately (in that they may not measure the welfare of an identifiable group) although they do have welfare significance when taken together (as a measure of the total change in welfare). This is because the conventional welfare areas reflect welfare changes in related markets (for example, the area that conventionally repre­ sents changes in consumer surplus for beef may now contain some compo­ nents due to changes in producer surplus for chicken). Thurman's most useful result - for the present purpose - is to show that when there is only one source of feedback (i.e., when the goods are related through either consumption or production but not both), the conventional measures of welfare change taken off the general-equilibrium supply-and-demand curves do have welfare significance (i.e., they do measure changes in the welfare of identifiable groups). These results mean that, with only one source of general-equilibrium feedback, it is possible to measure the total welfare change - and its incidence - due to a displacement in one market, taking account of general­ equilibrium adjustments. In applications to research-induced market dis­ placements, we have a remaining problem of defining the impact of technical change as a displacement of a general-equilibrium supply or demand func­ tion. What is the percentage shift of a general-equilibrium supply curve for chicken in response to a K percent reduction in the cost of chicken produc­ tion?19 The implications of Thurman's (1991 b) results are clearest for either (a) a supply shift when there is feedback to demand through substitution in consumption or (b) a demand shift when there is feedback to supply through substitution in production. In these cases we can use the conventional (partial-equilibrium) measures of research-induced displacements. 19. See chapter 5 for more discussion on estimating K. 234 Economic Surplus Methods With regard to the second question, welfare measurement with only one displacement is difficult enough in a general-equilibrium setting with only one source of feedback. To allow two or more changes to occur simulta­ neously would be very difficult, even with limited options for feedback. It is not difficult to model price, quantity, and revenue changes in multiple factor and product markets with general-equilibrium feedback and multiple dis­ placements occurring simultaneously (e.g., see Mullen, Alston and Wohl­ genant 1989). The difficulty is to measure the welfare consequences. Thus, where general-equilibrium issues are thought to be important, with multiple sources of general-equilibrium feedback, an entirely different approach is necessary. A full general-equilibrium treatment would allow measurement of the full welfare consequences (e.g., Ballard, Shoven and Whalley 1985). This requires a model of the entire economy, an exercise that is likely to be beyond the scope of most research evaluation and priority-setting studies. An Explicit General-Equilibrium Model Using the Balance of Trade Function Martin and Alston (1992, 1994) have described and illustrated an "exact" approach for measuring the benefits from new technology using a modifica­ tion of the widely utilized distorted trade expenditure function or balance of tradefunction (e.g., Lloyd and Schweinberger 1988; Vousden 1990; Ander­ son and Neary 1992).20 The balance of trade function may be defined for a single-household economy as H=e(p,w,ui)-g(p,w,v,'t)-(p-p"')' m(p,w,v)-f (4.6) and the money-metric measure of total welfare in the economy, Hi, given an intitial utility of ui, is based on four components: (a) the minimum expendi­ ture necessary to obtain a given level of utility from consumption, e(.), (b) income to owners of factors of production, g(.), (c) government revenues from trade taxes, (p - pw)'m(.), and (d) net transfers from abroad,! These four components are obtained as follows. The function e(.) is the net expenditure function of a composite household for a given vector of domestic product prices,p, a vector of domestic factor prices, w, and a level of utility that is exogenously specified at level ui since this measure is based upon the Hicksian money-metric measures of welfare change. The function g(.) defines the maximized profit generated from production in the economy 20. In this approach, advantage is taken of the modern, duality methods for modeling general equilibrium trade and welfare. While a complete model is required of the economy of interest, interest may be confined to a subset of the entire economy and, through judicious use of separability assumptions and aggregation, the model can be kept quite small. Economic Surplus Methods 235 for given domestic prices of outputs, prices of endogenously supplied fac­ tors, w, a vector of fixed factors, v, and a vector of technology variables, 'C, representing the state of the available technology. For a vector of world prices, pW, the second-last term in equation 4.6 is the government revenue generated by tariffs (or spent on export subsidies). It is calculated as the inner product of the vector of trade taxes on each commodity (p - pW) and the vector of actual import levels, m, which are determined by product and factor prices and the resource endowment. 21 The tariff revenues are assumed to be redistributed without cost to the composite consuming household. Finally,f, is the financial inflow from abroad in the form of net transfers, net factor income flows, or foreign borrowings utilized by the economy. The use of the expenditure function approach, as in equation 4.6, means that money measures of the compensation required to maintain a particular level of utility are deri ved in a consistent manner, avoiding the discrepancies that can arise when compensation is considered one market at a time (see Thurman 1991a; Hueth and Just 1991). The modified trade-expenditure function presented in equation 4.6 can be generalized in several ways. First, it can be extended from a single household to any number of households simply by identifying the expenditure and revenue functions associated with each household or household group. Similarly, vertical market linkages through intermediate inputs can be incorporated by identifying separate net revenue functions for the input-supplying and input-using sectors. Domestic taxation on production, consumption, or factor returns can be incorporated in the same manner by distinguishing between the prices paid by demanders and received by suppliers and by accounting for the resulting government revenues in the same way as for tariff revenues. An exact money-metric measure of the welfare change resulting from technical change can be obtained from equation 4.6 simply by comparing the net expenditures required to achieve a given level of utility, ui , under the initial technology, 'Co, and under the new technology, 'C 1• The compensating variation version of the measure, with the utility level in the expenditure function held constant at d\ is The equation for equivalent variation is the same except that utility is held at u l rather than uo. Since the components of changes in H shown in 4.6 are 21. Tariff revenues are calculated using the actual level of imports rather than the quantity of imports that would result if compensation had been made to hold utility at ri, following the approach of Martin and Alston (1992) (based on the suggestion of Mayshar 1990). If preferred, the actual import demands - m (p. w, v) - in equation 4.6 can be replaced with the compensated demands to provide a welfare measure based on actual, rather than hypothetical, compensation. 236 Economic Surplus Methods expressed purely in money terms, they can simply be added across different types of households or firms as long as different distributional weights are not being imposed. Different distributional weights can be incorporated if desired. The welfare evaluation in equation 4.7 is based on changes in product and factor prices from the intial period (subscript 0) to the final period (subscript I) as a result of the change in technology from to to tl, These price changes are established by the simulation of market equilibrium using Marshallian supply-and-demand curves, a simulation that is logically prior to and sepa­ rate from the welfare evaluation. For consistency, these Marshallian supply­ and-demand curves must be derived from, and share parameters with, the relevant components of the modified trade-expenditure function, equation 4.6. The basic surplus model in section 4.1 may be regarded as a special case ofthe more general model defined by equation 4.7, which applies when all product prices but one are exogenous and there are no distortions. In such a case, the only remaining differences are that (a) the basic model uses the Marshallian measure of consumer welfare change rather than the Hicksian measure and (b) the producer surplus measured in the basic model may differ from the change in producer profit in equation 4.7 The balance of trade function approach greatly expands the problems for which accurate welfare evaluation can be undertaken without requiring any data or parameters beyond those needed for the traditional approach. A behavioral model, constructed as described above, may be used to solve for a baseline and a perturbed solution for the price, quantity, expenditure, and revenue variables in the model. Because of the structure of the formulation, the behavioral model must be a general-equilibrium model. However, when the commodity (or commodities) of interest and all other goods can be assumed to be separable, the model need not be any more complicated (and requires little if any additional information to parameterize) than the more traditional partial-equilibrium approach. Once a general-equilibrium model has been used to compute the prices and quantities under the old and new technologies, the total welfare change and each of its components can be computed. If the supply-and-demand curves used to generate the prices and quantities are derived directly from the underlying preferences and technology, then the welfare measures will be exact. In sharp contrast with the traditional approach, multiple sources of general-equilibrium feedback present no problems for the calculation and interpretation of exact measures of the welfare effect and its distribution at any level of aggregation. The procedures identified above can be imple­ mented without modification in the presence of multiple sources of price change and endogenously determined prices. Economic Surplus Methods 237 The use of the balance of trade function approach in conjunction with a general-equilibrium model seems to hold the potential to resolve in a satis­ factory way many of the problems that have arisen when attempts have been made to expand the range of settings in which research benefits are evalu­ ated. The conventional graphical approach becomes increasingly difficult as the problem becomes more complicated. In principle one should get the same answer from the system of supply-and-demand equations, by integrating back, as one gets by starting with the expenditure function and modeling equilibrium in terms of the supply-and-demand equations derived from it (i.e., the duality should work in both directions). In practice, however, supply-and-demand equations may not be integrable and the results obtained by starting from supply and demand may not be the same as those obtained by specifying the full integrable model at the outset. At least in situations where the market situation to be studied is compli­ cated by distortions and multiple sources of general-equilibrium feedback, it will usually be better to use Martin and Alston's (1992, 1994) approach rather than to attempt to build welfare measures from a single-market analysis. It is also likely to be appropriate to apply that type of approach more generally. Whether that is so in a particular study will depend on the additional costs of using the theoretically more defensible approach relative to the benefits in terms of greater precision and consistency of results. Further work is necessary with these relatively new methods to establish the dimensions of that trade-off. 22 When markets are less complicated, the graphical single-market approach is adequate, and it is preferable in terms of its transparency and minimal requirements for data and modeling expertise. Such situations are likely to be the norm in typical research evaluation and priority-setting studies. Thus, the remainder of this chapter emphasizes relatively simple partial-equilibrium models that, for the most part, do not involve important problems of a general-equilibrium nature. 4.2.4 Multiple Products Related in Consumption The simplest case of multiple products has two (or more) products that are substitutes in consumption but entirely unrelated in production (i.e., the 22. As illustrated by Martin and Alston (1994). in the context of an evaluation analysis. the additional work may not be too onerous. They showed that a three-good model (i.e .• two goods of interest. and a numeraire representing all other goods) could easily be set up and run on a spreadsheet to evaluate benefits from research-induced changes in technology for one or more goods with market distortions. Martin and Alston (1993) report results from a full global general-equilibrium model. the OECD-World Bank RUNS model. What remains to be established by further empirical work is the magnitude of tbe gains in precision to be obtained by doing the extra work. to the extent that extra work is involved. in going from ad hoc partial-equilibrium approaches to a consistent general-equilibrium approach. 238 Economic Surplus Methods technologies of production are independent and there are no specialized factors in common - any factors that are used in both products are perfectly elastically supplied to both industries). A supply-and-demand model for this case with n products could be written as Supply: Qi = fi(Pi, Bi) (4.8a) Demand: Qi = gi(PI, . .. ,Pn, Ai) (4.8b) for i = 1, ... , n. The supply of each product i depends only upon its own price and exogenous supply shifters, B;, but demand for each product de­ pends on the prices of all the products and exogenous demand shifters, A;. In logarithmic differential form, for two products, the system of supply-and-de­ mand equations may be written as Supply: E(Qi) =£ i [E(P i) + ~i] (4.8a') Demand: E(Qi) =l 1ilE(P1) + l1izE(P2) -l1ijUi (4.8b') for i = 1 or 2. In these equations the parameter definitions are slightly different from those used in the single-product model. The elasticity of supply of product i (£;) is as before, but the own-price elasticities and cross-price elasticities of demand, l1ij' are the natural values rather than absolute values so that own-price elasticities are negative (11;;< 0 for i = 1, 2). The solutions for relative changes in prices are E(P1) = -[Cl1Il UI + £1~I)C£2 -1122) + l1dUz - ~2)]/d (4.9a) E(P2) = -[(1122U2 + £2~2)(£1 -1111) + 1121CU1 - ~I)]/d (4.9b) where The equations for gross annual economic welfare changes have the same form as equations 4.4a and 4.4b, but the interpretation is that the subscripts denote different commodities rather than the same commodity in different countries or regions: I1CSi = -P;Qi[E(Pi) - u;l[l + O.5E(Q;)] (4. lOa) I1PSi = PiQ;CECP;) - ~;l[l + O.5E(Q;)] (4. lOb) In addition, and in contrast to the single-product case, it is not appropriate to simply add these measures up across is (now commodities rather than countries) to get a measure of total welfare change. As suggested by the quote above from Just, Hueth and Schmitz (1982), the total welfare changes Economic Surplus Methods 239 due to a supply (or demand) shift in the ith market are reflected in the general-equilibrium measures of consumer and producer surplus changes in that market alone. The welfare measures in equations 4.1Oa and 4.1Ob are based on general-equilibrium changes in quantities and prices; so they are general-equilibrium welfare measures. Adding up these measures of the welfare effects of a particular supply (or demand) shift across markets would lead to double counting. In general, to measure the incidence of a change, we have to look across markets in a disaggregated fashion. For example, consider an increase in supply of good 1 (~I > 0) with no other exogenous shifts (a l = ~ = ~2 = 0). The correct measures of welfare change taken in the market for good 1 (assuming a parallel shift) are I1CS' = -P]Q]E(P])[1 + O.5E(Q])] (4.1Oc) I1PS1 = P]Q][E(P]) - ~1][1 + O.5E(Q1)] (4.1Od) (4. toe) where I1CS* is the change in consumer surplus measured off the general­ equilibrium demand curve for good 1, and it comprises consumer surplus from both goods plus producer surplus on good 2 (I1CS* = I1CS I + I1CS2 + I1PS2). To disaggregate these measures further, (4.1Of) (4. tog) To clarify these points, consider figure 4.6: panel a represents the market for one good (say, chicken meat) and panel b represents the market for a substitute (say, beef). The initial demand curves (Dc,o and DB•o) are defined in the usual way as conditioned on the price of the other good being constant at its initial value (PC ,o or P B.O)' When the supply curve for chicken meat shifts (from Sc.o to SC,I)' a series of general-equilibrium-type adjustments take place in both markets: a fall in the price of chicken causes a fall in demand for beef (because they are substitutes); the subsequent fall in beef price causes a faIl in demand for chicken and so on. Ultimately, a new equilibrium is achieved at prices PB .I and P C.I with corresponding demand curves DC.I and DB•I· The curve D~ is the "general" equilibrium demand curve for chicken that traces out the demand response to exogenous price changes in the chicken market holding constant the supply curve for beef The usual treatment - holding constant the price of beef - is a special case that applies when the supply curve for beef is perfectly elastic. The solutions from the equilibrium­ displacement model for relative changes in prices and quantities reflect these 240 Economic Surplus Methods Figure 4 .6: Welfare effects with feedback in consumption (a) Chicken (b) Beef pc.n pc. ) d Dc,n(P/J.n) In Dc}PH.) o QChickcn o general-equilibrium-type responses, a fact that must be borne in mind when the solutions are being used to compute welfare changes. In figure 4.6, the full welfare consequences of the shift in the supply of chicken can be measured as the area beneath the demand curve D ~ between the two supply curves (Sc.n and SC.l)' This area (Juab!)) comprises the "con­ sumer surplus" of area Pc .oabPc .) and (with parallel supply shifts) "producer surplus" equal to area P c. lbcd. In this case, the change in "consumer surplus" comprises changes in consumer surplus from consumption of both beef and chicken and changes in beef producer surplus. These components could be disentangled with a little effort; note that the fall in beef producer surplus is given by area P8 .nefP8 •1' 4.2.5 Multiple Products Related in Production So far we have considered interactions among products only through substitution in consumption. Now we consider cases where multiple prod­ ucts are related either through their production technology or through factor use. Clearly in some cases production processes are interdependent among products. An alternative way for product markets to interact is when they share the use of a specialized factor of production?3 23. Clear examples are (a) when livestock industries (e.g., hogs and chicken) affect feed-grain prices, the supply functions of livestock products wi ll be related, and (b) the use of milk in production of various dairy products (e.g., see Perrin 1980). In these examples the products are also related in consumption and, perhaps, through production technology. Economic Surplus Methods 241 Mullen, Wohlgenant and Farris (1988, pp. 247-49) presented a two-prod­ uct, two-input model that they applied to the U.S. beef processing sector. In this model two products are produced using two specialized factors. The products are related in production and through factor markets, but not in consumption. The key assumptions are that the production function is (a) characterized by constant returns to scale and (b) separable between inputs and outputs. Their modc;!1 is outlined below. The productionfunction has the form (4. 11 a) Because q is linearly homogeneous in XI and X2, the costfunction is separable in prices and quantity: (4.llb) Corresponding output-constrained input-demand functions are (by Shep­ hard's lemma) XI = hI (W1,W2)Q (4.llc) X2 = h2(WI , W2)Q (4.lld) The second part of the problem is to maximize revenue subject to a constrained level of inputs, F. Homogeneity conditions result in a separable revenue function: (4.11e) Corresponding input-constrained output-supply functions are (by Hotel­ ling's lemma) QI = rI (PI ,P2)F (4.1 If) Q2 = rz{P I ,P2)F (4.11g) The system of logarithmic differential equations describing equilibrium becomes Final demand: E(QI) = -llI[E(PI ) - ad (4.1 1a ') E(Q2) = -1l2[E(P2) - a2] (4.11b') Constrained output supply (transformation) and factor demand (substitution): E(QJ) - E(Q2) = 't[E(Pd - E(P2)] (4.11c') 242 Economic Surplus Methods (4.11d') Market equilibrium: m1E(P1) + mzE(Pl) = sIE(W1) + s2E(W2) (4.1 Ie') m1E(Yl) + m1E(Y2) = sIE(Xl) + SzE(X2) (4.11f) Factor supply: E(X1) =c l[E(W1) + ~d (4.11g') E(X2) = c2[E(W2) + ~2) (4.11h') In these equations (with some slight changes from the notation used by Mullen, Wohlgenant and Farris 1988), the parameters and variables are defined as follows: the quantity of product i is Qi and its price is Pi' the price of factor i is Wi and its quantity is Xi' the fraction of revenue accounted for by product i is m i, the fraction of cost accounted for by factor i is s;, the absolute value of the demand elasticity for product i is 11i, the supply elasticity for factor i iSci, the elasticity of product transformation is 't, and the elasticity of factor substitution is 0'. This model includes only two types of equilibrium displacements - those due to shifts of final demand and those due to shifts of factor supply (the shift of demand for product i is ai' and the shift of supply for factor i is ~i) - and it does not allow for the products to interact in consumption. Mullen, Wohlgenant and Farris (1988) show how to obtain numerical solutions to this model using matrix algebra. The solu­ tion is a vector of values for the relative changes in prices and quantities of the factors and products. Measures of welfare changes can then be computed by substituting the relative price and quantity changes into the following formulas: !:!CS; = -P;Q;[E(PJ - a;1[1 + O.5E(QJ] (4.12a) !:!PSi = WiX;[E(W;) + ~;1[l + O.5E(XJ] (4.l2b) (4.12c) !:!PS = Li!:!PSi (4.12d) !:!TS = !:!PS + !:!CS (4.12e) where equation 4.12a measures the change in consumer surplus in consump­ tion of good i, equation 4.12b measures the change in producer surplus in supplying factor i, equation 4.12c measures the change in consumer surplus across both products, equation 4.12d measures the change in producer surplus on all factors, and equation 4.12e measures the total welfare change. The aggregated results for producer surplus in equation 4.12d are the same Economic Surplus Methods 243 whether they are summed using components from 4.1 Ob to represent the sum of producer surplus changes across commodity markets, or whether they are summed using components from 4.12b to represent the sum of changes in surpluses accruing to factor suppliers, under competitive equilibrium as­ sumptions. The consumer surplus formulas, 4.1 Oa and 4.12a, are identical. 4.2.6 Demand Shifts Quality Change A recurring problem in analyzing the effects of new technology is the question of whether changes in technology involve changes in product quality characteristics as well as changes in factor use for a product. In rice, for example, broken grains, shape, chalkiness, amylase content, glutination temperature, gel consistency, and fragrance are varietal quality characteris­ tics that may be subject to research (Unnevehr 1986, 1990). A further example is the mechanical tomato harvester that required tomatoes to be sufficiently robust to withstand the process. Higher-yielding wheat varieties may have lower protein content, while barley varieties differ by malting characteristics (Brennan 1984; Ulrich, Furtan and Schmitz 1986, 1987; Macagno 1990; Voon and Edwards 1992). Lemieux and Wohlgenant (1989) considered the effects of consumer preferences for lean meat when analyzing the impact of porcine somatotropin, the primary impact of which is yield improvement or cost saving (see also Voon and Edwards 1991a). Differentiated products, which vary according to some quality character­ istics, face differential demands so that higher-quality goods command a premium. Farm products may be perceived as being of higher quality either because they have attributes that lead to higher quality from the retail viewpoint or because they have attributes that are advantageous from the viewpoint of intermediaries. For instance, Macagno (1990) used a multistage model to represent the malting quality of barley as an embodied technology, the initial benefits of which accrue to malsters and brewers. In this case there is no tangible change in the quality of the final product. Similar approaches may be appropriate for a wide range of other products (such as cotton, where milling costs are affected by uniformity offiber quality, and higher-protein grains that yield flour with better baking quality but not necessarily an appreciable change in the characteristics of the final product). In most cases it seems likely that technological changes will involve some changes in product characteristics, and sometimes these changes will be very important. For the most part agricultural economists have sidestepped the question of jointly modeling technical changes and associated changes in 244 Economic Surplus Methods product quality.24 One approach is to use a multiproduct model of the type described in the previous section(s) and either to treat product characteristics as products (so that "quality" is continuously variable) or to treat different qualities of products as different products (discrete variation in "quality"). The latter approach may be more restrictive but it is probably more practica­ ble. The most common approach is to introduce an ad hoc shift in demand for the product induced by changes in quality. Technical change that leads to a change in product quality is a change in supply conditions not demand conditions, and it would be better to model it as such.25 The implication is that different qualities should be modeled using a multiproduct modeling approach. The difficulty with this approach is that the substitution effects between the different qualities of a particular product that are the most important (that determine the own- and cross-price elasticities of demand) are very difficult to measure - especially for ex ante studies where the different qualities might not exist when the analysis is being undertaken. In addition, substitution effects in production (say, between two varieties of wheat) are likely to be too important to dismiss when various qualities of a particular product are dealt with and when these too are difficult to quantify. Thus, to model quality changes formally may require using a model with multiple sources of general-equilibrium-type feedback. We have seen above that measuring welfare changes may be difficult in such a setting. However, in some cases, where product quality change is important, a formal attempt to analyze its effects in a logically consistent fashion may be worthwhile. The previous approaches in the literature (treating quality change as a demand shift) have avoided the difficulties by treating different qualities as perfect substitutes in consumption (up to a constant premium for quality) and treating the supply choice as exogenous, implying no substitution in produc­ tion in response to price changes (either a total switch from one quality to another or a partial switch determined exogenously, independent of prices). 26 24. Some exceptions are Unnevehr (1986, 1990), Lemieux and Wohlgenant (1989), Voon (1991), and Voon and Edwards (I991a, 1992). The studies by Ulrich, Furtan and Schmitz (1987) and Macagno (1990) are pertinent as well. 25. For instance, the development of technology for filtered cigarettes may have had gross effects similar to those from an increase in demand for the aggregate good, "cigarettes" (i.e., greater sales at a higher price), but it might at the same time have led to a reduction in demand for tobacco per cigarette with an ambiguous net effect on demand for tobacco. Modeling this change simply as an increase in demand for cigarettes would lead to an erroneous conclusion that demand for all inputs used in cigarettes had increased. 26. Mullen and Alston (1994) treated different qualities oflamb as perfect substitutes in consumption (i.e., linear indifference curves) but with different production and marlceting costs and consumer willing­ ness to pay. Then they modeled quality change in the context of a model of consumption and production aggregated over different qualities. A change in the product mix was modeled as leading to a shift in Economic Surplus Methods 245 Under these restrictive assumptions, it is possible to analyze quality change as a demand shift (the addition of a premium for improved quality) and to obtain meaningful measures of the size and distribution of benefits. How­ ever, when less restrictive assumptions are applied, it may not be safe to treat quality change as an equivalent shift of demand. Income, Population, and Other Demand Shifters The production effects of agricultural research are generally realized over several years. As a result, demand can change a lot over time, particularly because of changes in population and per capita income.27 The effect of adding an exogenous demand shift is illustrated in figure 4.7. The original price and quantity are Po and Qo. If research were to shift the supply curve down with no exogenous shift in demand, the new price and quantity would be PI and QI' However, if demand were to shift out exogenously, the new (post-research) price and quantity would be P/ and Q/ and the research-in­ duced changes in total economic surplus, consumer surplus, and producer surplus would be IoabI l , Po'abP/, and Ir,abII - Po'abP/ = PI'bcd, respec­ tively. In addition, this diagram can be used to illustrate the case where the demand shift is endogenous. For instance, putting aside our reservations about this approach, when new technology involves both an improvement in quality and cost savings, we could model it that way (e.g., Lemieux and Wohlgenant 1989). Alternatively, when improved agricultural technology leads to capital accumulation and growth outside agriculture, with concom­ itant effects on per capita incomes, and those effects are not already repre­ sented in the demand curves, an adjustment to demand due to technical change will be appropriate. Equations 4.1 a, 4.1 b, and 4.1 c can be used to calculate changes in total, consumer, and producer surplus but with Po' substituted for Po, Qo' substi­ tuted for Qo, and with the elasticities of supply and demand and the percent­ age shift in supply defined at the initial equilibrium being adjusted to reflect the new pre-research equilibrium (i.e., the move from point d to point a in figure 4.7). Those adjusted numbers are not directly observable, but esti­ mates of rates of population and income growth can be used to inflate the initial quantities and prices from their base values (as described in chapter 5) before the research-induced supply shift is introduced. fann-Ievel supply for total Iamb. a change in the overall marlceting margin. and an increase in aggregate demand for lamb of all qualities. 27. These changes occur both domestically and internationally. One of the few studies to explicitly incorporate domestic demand shifts over time is Norton. Ganoza and Pornareda (1987). 246 Economic Surplus Methods Figure 4.7: Effects of exogenous demand shifts on the size and distribu­ tion of research benefits So ltice sJ Pc; Po p ' I PI d '0 D ~ Do 'I 0 QO QO QI Q; Quantity 4.3 Vertical Market Relationships To study vertical market relationships in multistage production systems, we abstract from the temporal ordering of the stages of production and treat the different stages as if they occur at one time. The participants in different stages of the production system are represented as input suppliers and their welfare is reflected in the distribution of economic surplus among inputs. The multistage nature of input decisions can be reflected in constraints on substitutability among inputs through separability assumptions. The case of a single product (in partial equilibrium) produced with two factors in fixed proportions is discussed first and then we proceed to variable proportions, and then multiple factors. 4.3. 1 Two Factors with Fixed Factor Proportions The simplest case we can consider is when two factors of production are used in fixed proportions to produce a homogeneous product. The case of derived factor demand and output supply with two factors and fixed factor proportions is illustrated by Friedman (1976) with the example of knives, blades and handles. That model can be used to show market equilibrium and surplus di stribution between, for example, two farming inputs (say land and Economic Surplus Methods 247 other inputs) used to produce a farm product. Alternatively, we may use the same approach to analyze a multistage production system - say when a farm product and marketing inputs (such as transportation, processing, and distribution inputs) are used to produce a retail product. Equilibrium in Factor and Product Markets Figure 4.8 represents the markets for a farm product and a composite marketing input that are used in fixed proportions to produce a retail food product. The market situation is defined by (a) the technology of production (i.e., the fixed amounts of the two factors used to produce a unit of the retail product), (b) the supply conditions for the factors of production (the farm product supply is SFo and the supply of marketing inputs is SMo with the units of factor quantities defined per unit oft he retail product), and (c) the demand function for the retail product, DRo. Because the factors are used in fixed proportions, it is straightforward to derive the retail supply and factor demand equations. The retail supply function, SRo is given as the vertical sum of the underlying factor supply functions (SFo and SMo) so that the marginal cost of a quantity of the retail product is equal to the sum of the marginal costs of the corresponding factor quantities. The derived demand function for the farm product, DFo, is given by the vertical difference between the retail demand and the supply of marketing inputs. Similarly, the derived demand for marketing inputs, DMo, is given by subtracting the supply function for the farm product (vertically) from the retail demand function. The initial equilibrium in the product market is defined by the intersection of retail supply and demand at price PRo and quantity QRo. Equivalently, equilibrium may be defined in terms of one of the factor markets: equilib­ rium of supply and demand of the farm product is at price PFo and quantity QFo; supply and demand for marketing inputs are in equilibrium at price PMo and quantity QMo. Increase in Supply of Marketing Inputs Now, suppose the supply function for marketing inputs shifts down (say, in response to technical change) in parallel from SMo to SMJ • This shift affects the equilibrium in all three markets. The supply of the retail product shifts down (by the same absolute amount per unit) from SRo to SRJ. The demand for the farm product shifts up in parallel (also by the same absolute amount per unit) from DFo to DF • All quantities increase in proportion (to J QR J, QMJ, and QFJ). The prices of the marketing input and the retail product fall (to PMJ and PRJ), and the price of the farm product rises (to PFJ). 248 Economic Surplus Methods Figure 4.8: Research benefits with two factors infixed proportions Retail price SRo SR J PRo PRJ DR J d DRO 10 IJ o Retai l quantity Marketing margin o Marketing inputs quantity Fann price o Fann quantity Economic Surplus Methods 249 As a consequence of these changes, there is a total welfare gain of loabl1, comprising a change in consumer surplus, !::.CS = PRr,abPR1, and a change in producer surplus, !::.PS = PR 1bcd. The change in producer surplus com­ prises a change in surplus to suppliers of marketing inputs (!::.MS = PM /gh) and a change in surplus to suppliers of the farm product (!::.FS =P F1ijPFo). We can express these effects algebraically in the same form as we did for the basic model as follows: !::.CS = PRoQRr7(1 + O.5ZTJ) (4. 13 a) !::.PS = PRoQRo(K - Z)( 1 + O.5ZTJ) (4.13b) !::.TS = !::.CS + !::.PS = PRoQRoK(l + O.5ZTJ) (4.13c) where K is now the vertical shift of the supply function for marketing inputs expressed as a percentage of initial retail price, PRo, TJ is the absolute value of the elasticity of demand at retail, C is the elasticity of supply to retail and Z =K eI(£+TJ) is the percentage reduction in retail price due to the supply shift. The components of the change in producer surplus are !::.FS = PFoQFo(K - Z)( c/C[)(1 + O.5ZTJ) (4.13d) !::.MS = PMoQMo(K - Z)( c/cm)(l + O.5ZTJ) and (4.13e) !::.PS =!::.MS + !::.FS (4.13f) where cr is the elasticity of supply of the farm product and Cm is the elasticity of supply of marketing inputs. Equivalently, we could measure the total benefits in the market for marketing inputs as the area beneath the demand curve, DM(), between the two supply curves (SMo and SM1). This area comprises "producer surplus" (i.e., !::.MS = PM/gh) and "consumer surplus" in the market for marketing inputs (PMoefPM1 - which includes !::.CS to final consumers and !::.FS to suppliers of the farm product). Alternatively, we could measure the total benefits and their distribution in the market for the farm product; the total benefits in this case are equal to the area between the two demand curves and above the supply curve. The increase in "producer surplus" in the market for the farm product reflects benefits to producers of the farm product, !::.FS, and the increase in "consumer surplus" reflects benefits to final consumers and suppliers of marketing inputs (i.e., !::.CS + !::.MS). This set of results may be extended to any arbitrary number of factors of production. Considering individual factors, in any factor market the "pro­ ducer surplus" refers to surplus of suppliers of that factor while the "con­ sumer surplus" refers to surplus of both final consumers and suppliers of all 250 Economic Surplus Methods other factors. Alternatively, we can consider surplus in markets for interme­ diate products. At any market level, the "producer surplus" is the sum of quasi-rents accruing to all factors used in the production of the intermediate good (i.e., factors used up to that market level). The "consumer surplus" is the sum of final consumer surplus and the quasi-rents accruing to all factors used in conjunction with the intermediate good (i.e., beyond that market level). Another feature of the results warrants emphasis: the distribution of benefits is entirely independent of which of the curves shifts. That is, the total benefit and distribution of benefits would be the same from a shift down of the farm product supply function by the same amount per unit - i.e., to SF, (or, for that matter, from a shift up of the final demand function by the same amount per unit - i.e., to DR,), so long as the shifts are parallel. Thus, in this setting, farmers could afford to be indifferent both about where new technology applies in the production and marketing system and about where a levy to fund research is collected; maximizing total benefits will maximize farmer benefits. Change in Processing Technology So far we have treated technical change in terms of either a shift of the supply of the marketing inputs or a shift of the supply of the farm product. An alternative type of technical change would be a change in the production function that combines the raw materials (the farm product and marketing inputs). The change could be neutral (reducing the amount of both inputs required to produce a unit of the product but maintaining factor proportions), biased (reducing the amount of only one of the inputs required per unit ofthe product), or some combination of biased and neutral changes (changing the proportions and amounts of both inputs required per unit of the product). Figure 4.9 shows the effects of a biased technical change (saving marketing inputs) in the context of the market model described in figure 4.8. The technical change reduces - in proportion - the amount of marketing inputs used per unit of the farm product and per unit of the retail product. This amounts to a proportional shift down of the supply of marketing inputs (where the input quantities are expressed per unit of the final product) from SMn to SM, from the point of view of the producers of the retail product (equivalently, a percentage reduction in the cost of supplying "efficiency units" of the marketing input). The welfare effects are slightly more complicated in this case. For the retail product, consumer surplus is increased by tlCS = PRnabPR, and total surplus is greater by tlTS = lnabl,. For suppliers of the farm product, Economic Surplus Methods 251 producer surplus increases by !1FS = PFlcdPFo' For the marketing input, the number of "efficiency units" (QM~) is greater, but the actual use of market­ ing inputs (QM]) is smaller. The effect of the technical change on surplus accruing to marketing inputs is equal to the difference !1MS = PM Jh - PMoeg. This difference may be positive or negative, depending, primarily, upon the elasticity of final demand; a sufficient condition is that it will be negative when final demand is inelastic. Thus, farmers and consumers necessarily benefit from a biased (marketing-input-saving) technical change; marketing input suppliers may gain or lose. Technical change biased against the farm product could be modeled in the same way by switching the roles of the farm product and marketing inputs in figure 4.9. By analogy, then, farmers may gain or lose from a farm-prod­ uct-saving technical change in the food industry. Notice that biased technical change has effects that are similar to those of a proportional downward shift of the factor supply function.28 With a neutral technical change, it is relatively easy to show that when both factor supply functions slope up, both inputs will benefit when demand for the product is elastic (in which case total expenditure on both inputs rises with an increase in output), and conversely, both will lose when demand for the product is inelastic. These issues are more easily addressed as a special case in the context of the model of technical change with variable factor proportions that will be developed next. 4.3.2 Two Factors with Variable Factor Proportions The assumption of fixed factor proportions is an extreme one. Clearly, the extent of input substitution possibilities is an empirical matter, and there is some empirical support for using a less restrictive assumption that allows the possibility of substititution between farm inputs or substitution between farm products and marketing inputs (e.g., Wohlgenant 1989). We saw above that the analysis of research benefits and their distribution for any number of inputs (or stages of production) is quite straightforward under the assump­ tion of fixed factor proportions. With variable proportions it is difficult to get useful algebraic results for more than three factors of production?9 2K As shown by Mullen, Wohlgenant and Funis (1988), a biased technical change that is XI-saving may be modeled as an "equivalent" shift in the supply of X I' It will not be equivalent in all senses, however, and care must be exercised in assuming equivalence. 29. For two factors the results are fairly transparent (e.g., see Alston and Scobie 1983) but for three factors the analytics are quite cumbersome and the transparency is reduced (see Holloway 1989). Numerical rather than algebraic solutions are likely to be necessary for studies involving three or more factors. Wohlgenant (1982) provided a general solution for the case of one output and n factors. 252 Economic Surplus Methods Figure 4.9: Biased technical change with fixed proportions Retail SRo price '0 " DR 0 QRo QR, Retail quantity Marketing margin SMa PMo SM, PM, g h 0 QM, QMo QMj Marketing inputs quantity Farm price o Farm quantity Economic Surplus Methods 253 Freebaim, Davis and Edwards (1982) analyzed the distribution of research benefits in a three-stage model with fixed proportions between purchased farming inputs, a farm product, and marketing inputs. They illustrated their results with an application to the U.S. hog industry. Their key points were that with parallel shifts of linear supply functions, (a) innovation at any stage of a multistage production process confers positive benefits on consumers and producers in all stages (i.e., factors) of production and (b) the distribution of benefits is independent of where the innovation applies in the system. In a comment on Freebaim, Davis and Edwards (1982), Alston and Scobie (1983) demonstrated that the distribution of research benefits among factors (or stages) of production depends crucially upon the elasticity of substitution. Their approach to modeling benefits from technical change has subsequently been adopted, adapted, and extended in several studies?O The approach owes its origin to Muth (1964), who presented an elegant, simple model of equilib­ rium displacement in a two-factor model of supply and factor demand in a competitive industry.31 First, a slightly modified version of Muth's (1964) model of market equilibrium displacements is presented below. Then the welfare economic effects of research-induced technical changes are con­ sidered. The Muth Model Following Muth (1964), we can model the market equilibrium of a competiti ve industry producing a homogeneous product using two factors of production in terms of the following six general equations: Consumer demand: Q =f (P) (4.14a) Production: Q =q (XJ, X2) (4. 14b) Factor demand: WI=Pql (4. 14c) W2 =P q2 (4. 14d) Factor supply: XI =g (WI) (4. 14e) X2 =h (W2) (4.14f) The endogenous variables in the model are industry output, Q, the amounts 30. Examples include studies by Mullen, Wohlgenant and Fanis (1988), Mullen, Alston and Wohlgenant (1989), Lemieux and Wohlgenant (1989), Holloway (1989), Mullen and Alston (1990), and Wohlgenant (1993). 31. Gardner (1975) used a very similar model to analyse marketing margins. Miedema (1976) clarified the connection between the Muth ( 1964) and Gardner (1975) models. More recently, Gardner (1987) applied the same type of modeling approach to a range of agricultural policy issues. 254 Economic Surplus Methods of the two factors used by the industry (XI and X2), the price per unit of the final product, P, and the factor prices (WI and W2). Table 4.1 summarizes the notation used in the Muth model. Equation 4.14a is the demand for the industry'S output, equation 4.l4b is the production function, equations 4.14c and 4.14d are factor-demand equations with each factor being paid the value of its marginal product (q; = dq[.]/dX; for factor i), and equations 4.14e and 4.14f are the factor-supply equations. Constant returns to scale is assumed at the industry level.32 Totally differentiating equations 4.14a through 4.14f, converting them to elasticity form, and adding exogenous shocks yields the following system of logarithmic differential equations - expressed in terms of relative changes and elasticities:33 E(Q) = -11 [E(P) - a] (4. 14a') E(Q) = S1E(X1) + S2E(X2) + 0 (4. 14b') E(W1) = E(P) - (s2/cr)E(X1) + (s/cr)E(Xz) + 0 + y (4.14c') E(Wz) = E(P) + (s1/cr)E(X1) - (s1/cr)E(Xz) + 0 - (S1/sZ)Y (4.14d') E(X1) = E1 [E(W1) + ~1] (4. 14e') E(Xz) = Ez[E(Wz) + ~z] (4.14r) where E denotes relative changes (i.e., E(Z) =d Z/Z = dlnZ), 11 is the absolute value of the elasticity of demand, a is a vertical shift in the demand function reflecting an increase in demand, S; is the cost share of factor i (s; = W; X; / PQ) and, under an assumption of constant returns to scale, Sl + S2 = 1,0 is a (neutral) upward shift in the production function, Y is a biased (X2-saving) technical change, cr is the elasticity of substitution between XI and X2, E; is the elasticity of supply of factor i, and ~; is a vertical shift down in the supply of factor i reflecting an increase in its supply.34 The exogenous shift parameters (a, ~I' ~2' 0, and y) express equilibrium displacements relative to an initial equilibrium. For instance, setting a =0 .1 32. See Diewert (1981) for a discussion of this assumption. 33. Muth (1964) shows how to do this for the case being considered here. Mullen, Alston and Wohlgenant (1989, pp. 44-5) show the steps involved in this transition for the three-factor case, approach­ ing the problem from the dual side (i.e., using a cost function rather than a production function). 34. Freebairn, Davis and Edwards (1983) objected to the Muth (1%4) specification of biased technical change and suggested an alternative treatment in which only one factor-demand equation is affected. This objection is primarily terminological. Muth claimed correctly that any technical change could be modeled a~ a combination of his biased component (twisting the isoquant - y) and a neutral component (relabelling isoquants or relabelling the axes -Ii). Neither treatment allows the possibility of a technical change that would alter the elasticity of substitution (i.e., the curvature of the isoquants). Economic Surplus Methods 255 Table 4.1: Notation Used in the Muth Model Variable or parameter Definition Endogenous variables Q Quantity of product p Price of product Xi Quantity of factor i (for i = 1,2) Wi Price of factor i (for i = 1,2) Qi Marginal product of factor i (for i = 1,2) Market parameters 11 Absolute value of the elasticity of final demand fi Elasticity of supply of factor i (for i = 1,2) Si Cost share of factor i (for i = 1,2) cr Elasticity of factor substitution Exogenous shift variables a Relative increase in demand (vertical shift up in the price di!:ection) l3i Relative increase in supply of factor i (vertical shift down in the price direction) y Relative increase in marginal product of factor XI due to an X2-saving biased technical change, holding output constant Relative increase in output and marginal products of both factors due to a neutral technical change would imply a 10% increase in consumers' willingness to pay for the initial quantity of the product. As in the case of the multiproduct model, while the demand shift is expressed as a percentage of the initial price, a proportional shift of demand cannot be presumed. Rather, ex measures the vertical shift in demand at a point, locally, for any type of demand shift (e.g., proportional, parallel, or pivotal). Similarly, ~i measures the shift down of the supply of factor Xi with the magnitude of the reduction in marginal cost (at the point of approximation, the initial equilibrium) being expressed relative to the initial price of the factor. These shifts are shown in figure 4.10 which is a diagram­ matic representation of the model in equations 4.14a' through 4.14f'. Solutions to the Muth Model As we can see in the equations of the model (i.e., equations 4. 14a' through 4.14f') or in figure 4.10, mutually consistent changes in prices and quantities of factors and products may arise from shifts of the final demand, a, either factor supply function, bl or b2, a neutral technical change, d, or a biased technical change, g. Algebraic solutions may be obtained by a sequence of 256 Economic Surplus Methods substitutions (as by Muth 1964, see also box 4.3a) or by matrix algebra methods (as shown in box 4.3b). The parameters and variables in the Muth model are defined in table 4.1, and the reduced-form solutions are shown in table 4.2 as equations 4.15a through 4.15f. These equations are derived from Muth (1964, p. 233) but with slightly different notation (the parameters are all defined as positive and the shift variables are defined so that when they have positive values, the relevant quantity increases), and we have corrected the error in Muth's equation 24 noted by Freebairn, Davis and Edwards (1983). 4.3.3 Research Benefits with Input Substitution To measure the surplus changes associated with the equilibrium displace­ ments described by the two-factor model above, it is necessary to define the functional forms of the factor-supply and-demand functions and the nature of the shifts induced by the various changes. As Lindner and Jarrett (1978) and others have shown in the context of the "basic model," the functional form and nature of the supply shift have important implications for measures of benefits - the nature of the shift is especially important. Surplus Measures The model in equations 4.14a' through 4.14f' does not involve any explicit or implicit assumptions about the functional forms of supply and demand. It is a local approximation to unknown functions; the approximation is linear in logarithmic differentials (i.e., relative changes) and elasticities; it is not assumed that the elasticities are constant.35 In the work that follows, it is assumed that supply-and-demand functions are approximately linear in the region of interest and that the curves shift in patallel as a result of exogenous factors (a, ~i, y, and 0). Under these assumptions, the benefits accruing to consumers (LlCS) and factors of production (MS; for i = 1,2) may be measured in terms of the changes in factor and product prices and quantities from equations 4.15a through 4.15f, using LlCS = - PoQo[E(P) - a][l + O.5E(Q)] (4.16a) (4.16b) 35. For instance, it is perfectly valid to use this type of model to analyze the effects of parallel shifts in the case of linear supply and demand functions. Alston and Wohigenant (1990) have shown that this type of linear elasticity model is exactly correct for linear supply and demand and only approximately correct for constant elasticity functions when using E(X) = dX/X. The opposite is true when using E(X) = dlnX. Economic Surplus Methods 257 Figure 4.10: The Muth equilibrium-displacement model (a) Consumer demand shift Q (b) Neutml technical change (c) Biased technical change _xl> oxp (d) Shift in supply of Xl (e) Shift in supply of X2 xO2 X2 258 Economic Surplus Methods BOX 4.3a: Solving the Muth Model by Substitution The equations of the model may be written as E(Q) = -TlE(P) + Tla (4.3.1) E(Q) =s ,E(X,) + S2E(X2) + 0 (4.3.2) E(W,) = E(P) - (s/a)[ E(X,) - E(X2) ] + 0 + 'Y (4.3.3) E(W2) = E(P) + (s/a)[ E(X,) - E(X2) ] + 0 - (S/s2)'Y (4.3.4) E(W,) = (1/f)E(X,) -~, (4.3.5) E(W2) = (l/E2)E(X2) - ~2 (4.3.6) Substituting 4.3.5 into 4.3.3 and 4.3.6 into 4.3.4 yields (l/E,)E(X,) - ~, = E(P) - (s/o)[ E(X,) - E(X2) ] + 0 + 'Y (4.3.7) (1/E2)E(X2) - ~2 = E(P) + (s /0')[ E(X ,) - E(X2) ] + 0 - (s /S2)'Y (4.3.8) Setting 4.3.1 equal to 4.3.2 and solving for E(P) yields E(P) = a - (s/TI)E(X,) - (s/TI)E(X2) - OITI (4.3.9) Substituting 4.3.9 into 4.3.7 and 4.3.8 eliminates E(P) and yields (lIEj)E(Xj) - ~} = a - (s}/TI)E(Xj) - (st'TI)E(X2) - OITI - (sia) [E(X}) - E(X2)] + 0 + 'Y (4.3.10) (lIE2)E(X2) - ~2 = a - (slrl)E(X}) - (siTl)E(X2) - OIT] + (s}/a) [E(X}) - E(X2)] + 0 - (s)ls2)'Y (4.3.11) Collecting terms involving X, and X2 yields [(liE)) + (s)/T]) + (s2/a)]E(X}) + [(s2/T1) - (sia)]E(X2) = ~) + a + O(l-lIT1) +'Y (4.3.12) [(lIE2) + (siTl) + (s}/a)]E(X2) + [(s/TI) - (s}/a)]E(X}) = ~2 + a + O(l - "T1) - (s/s2)'Y (4.3.13) or, more succinctly, A,E(X,) + A2E(X2) =A 3 (4.3.12') B,E(X,) + B2E(X2) = B3 (4.3.13') where A, = (liE,) + (siTl) + (s!a) = (O"T\ + aE,s, + TlE,s2)la TIE, (4.3.14a) A2 = (s/TI) - (s/a) = s2(a - TI)la TI (4.3.14b) A3 = ~, + a + O(l - "T1) + 'Y (4.3.14c) B, = (s/T]) - (s/a) = sICa - TI)la TI (4.3.14d) (continued on next page) Economic Surplus Methods 259 Box 4.3a: (continued) B2 = (1/E2) + (s/l1) + (s/<"5) = (<"5T\ + <"5E2S2 + TjE2S])/<"5 TjE2 (4.3.14e) B3 = 132 + a + 0(1 - IITj) - (S/s2)Y (4.3.14f) Solving 4.3.12' and 4.3.13' for E(X j ) and E(X2) gives E(X,) = ( A2B3 - A3B2 ) 1 D (4.3.15a) E(X2) = (A)B, -AlB) 1 D (4.3.15b) where D =A2B, -A,B2 , Substituting the terms above into the expression for D and simplifying it yields D = - [<"5 Tj + <"5 (S,E,+S2E2) + Tj(S,E2+S2E,) + E,E2] 1 <"5 TjE,E2 Multiplying the numerator of equation 4.3.15a by - <"5 TjE,E2 yields E(X,) = [(A3B2<"5 TjE,E2) - (A 2Bp TjE,~)] 1 D' (4.3.16) where D' = <"5 Tj + <"5 (s ,E,+ S2E2) + Tj(s ,E2+ S 2E,) + E,E2 Substituting from equations 4.3.14 and simplifying we get A)B2<"5 TjE,E2 =A 3 (<"5 Tj + S2<"5E2 + s,Tj~) E, and B02<"5 llE,E2 =B 3s2(<"5 - Tj)E,E2 After substituting for A3 and B3 we have the solution for E(X,). Sillce the solution is linear in the shift terms, its elements can be derived for each shift parameter in turn. Thus, the term involving the demand shift, a, is equal to E(X\i a) = { [(<"5 Tj + s2<"5E2 + s\TjE2)Ed - [s2(<"5 - 11)E1E2] ) aID' = TjE\(<"5+E2) al D' The terms involving the factor-supply shifts are given by E(X,i 13,) = (<"5 Tj + S2<"5E2 + S,TjE2)E, 13,1 D' E(X,i 132) =- S2( <"5 - Tj)E,E2 1321 D' The terms involving the technical changes are given by E(X\i 0) = { [(<"5Tj +s2<"5E2 +sITjE2)Ed-[s2(<"5-11)E1E2]) (1-1111)01 D' =- (I-Tj)E\(<"5 +~) 0 / D' E(Xli Y) = { [(<"5T\ + s2<"5E2 + slllE2)Ed + (s lls2)[s2(<"5-Tj)E1E2] ) yl D' = <"5E1( Tj + E2) y 1 D' The equation for E(X,) is given by the sum of these five terms. The equation for E(X2) can be derived using the same approach but eliminating E(X,), instead of E(X2), from equations 4.3.15. Equations for the other four endogenous variables (the output quantity and the three prices) are readily derived using the solutions for the input quantities and the equations of the model (4.3.1,4.3.2,4.3.5, and 4.3.6). 260 Economic Surplus Methods BOX 4.3b: Solving the Muth Model Using Matrix Algebra The matrix algebra approach is more tractable than the substitution approach, espe­ cially for problems involving larger numbers of equations (when there are more inputs or outputs, as shown later in this chapter). The first step is to transform the model so that the exogenous shocks are on the right-hand side. In the two-factor, one-product case, the model (i.e., equations 4.14) thus may be represented as follows: E(Q) + 1'] E(P) = 1'] - s)o(€) -~) y]/D (4.15b) E(X) = [T\€) (0 + 102)0: + I T\o + (s20 + s)T\)~) €)~) - s2(0- T\)€2€)~2 - (0 + 102) (1 - T\)€)O + €)o(~ + T\) y]ID (4.l5c) E(X2) = [T\~(o +£)0: - s)(o- T\)E)~~) + I T\O + (s)o + S2T\)€) ) ~~2 - (0 +£)(1 - T\)E20 - (S)IS2)E20(E) + T\) y]ID (4.15d) E(W) = [T\(o + ~)o: - (s)o + s2T\ + E2)E)~) - s2(0- T\)~~2 - (0 + E2)(1 - T\)O + 0(E2 + T\) y]/D (4.15e) E(W2) = [T\(o + E)O: - s) (0- T\)E)~) - (s20 + s) T\ + E) )E2~2 - (0 + E)(l - T\)O - (s)1 S2)0(E) + T\) y]/D (4.15f) Note: D =c r(11+s)£) +s2£2) + 11(s2£) +S) Ez) + £) Ez ' and D > °f or 11 > 0, cr> 0, and £) and £2> 0. 262 Economic Surplus Methods n !J.TS = !J.CS + 'L!J.PSi (4. 16c) i=l Mullen, Alston and Wohlgenant (1989) present equivalent formulas for the case of three factors. Equations 4.16a through 4.16c may be used for an arbitrary number of factors (or stages of production) to estimate total benefits and the distribution of those benefits from equilibrium displacements under the assumptions being used here. They can also be used to examine the effects of a combination of displacements (which add linearly) or individual changes. Freebairn, Davis and Edwards (1983) present formulas for surplus changes that correspond to these equations after substituting terms from equations 4.15a to 4.15f. Qualitative Results The qualitative results considering individual exogenous shifts in isola­ tion are shown in table 4.3. With the exception of the biased technical change, consumers always gain from the displacements associated with positive values for any of the exogenous shift variables; they either shift demand up, ex, or shift final market supply down (Bi' 0). In the case of a biased (X2-saving) technical change, consumers will benefit only when the elasticity of supply of XI is greater than that of X2 (i.e., tl > t2). Freebaim, Davis and Edwards (1983) suggested an alternative specification of biased technical change that they found more plausible and which avoided this ambiguity. Factor suppliers gain from a parallel shift down of their own supply function (i.e., surplus to producers of XI increases with positive values of BI and surplus to producers of X2 increases with positive values of B2). However, the cross-effects of factor-supply shifts may be positive or negative, depending upon whether the two factors are gross substitutes or gross complements. When the elasticity of substitution is less than the absolute value of the demand elasticity (cr < 11), the two factors are gross complements (i.e., the cross-price elasticity of factor demand is negative so that a fall in price of either factor will increase the demand for the other factor). In this case, both factors benefit when either factor-supply function shifts down. In the ex­ treme case of fixed proportions (cr = 0), the distribution of benefits is independent of which factor-supply function shifts and the results are as derived in section 4.3.1 above. When the elasticity of substitution is greater than the absolute value of the demand elasticity (cr > 11), the two factors are gross substitutes (i.e., the cross-price elasticity of factor demand is positive so that a fall in price of either factor will reduce the demand for the other factor). In this case, suppliers of XI lose when the supply function for X2 shifts Economic Surplus Methods 263 Table 4.3: Incidence of Benefits from Technical Change in the Muth Model Interest groups Type of change in technology Suppliers of XI Suppliers of X2 Consumers Demand increase (n> 0) + + + Increase in supply of Xl (~l > 0) + cr 0) cr 0) + £1> £2 Neutral (0 > 0) TJ>l TJ>1 + Note: Entries denote conditions under which interest groups benefit. + indicates that benefits are positive under all conditions. - indicates that there are no conditions under which benefits are positive. All entries are subject to the assumptions that Tj, (J, £1' and ~ ~ a. Entries in the row for X2-saving technical change assume (J is strictly positive. When (J =0 , there are no effects from biased technical change as defined by Muth (1964); all the entries in that row become zeros. down (~2 > 0) and suppliers of X2 lose when the supply function for Xl shifts down (~l > 0). Both factors gain from a neutral technical change (8 > 0) when demand is elastic (11 > 1); both factors lose when demand is inelastic. Factor Xl benefits from a biased (X2-saving) technical change and factor X2 loses unless we have fixed proportions (cr =0 ), in which case there is no effect on quantity or price of output and no effect on quantity or price of either factor. Alston and Scobie (1983) and also Freebairn, Davis and Edwards (1983) considered the distribution of benefits of these various types of technical change in the two-factor case (between a farm product and marketing inputs). They concluded that in contrast to the case of fixed-factor propor­ tions, when there is input substitution, the distribution of benefits depends on the nature of the research-induced technical change. They also suggested that the model can be used to measure the incidence of costs of a levy to fund research. When there is input substitution, the incidence of a research levy on the farm product will be different from the incidence of benefits from research, other than research directed at shifting the farm-product supply function. These issues have been explored further in empirical models?6 36. Empirical studies that have considered the implications of input substitution for the distribution of benefits from different types of technical change include, for example, Mullen, Wohlgenant, and Farris 264 Economic Surplus Methods One issue that has not been resolved in this literature is how best to model biased technical change. Muth (1964) suggested one approach: shifting both factor-demand curves - in effect, twisting the isoquant to change the ratio of marginal products but holding output constant. Freebairn, Davis and Edwards (1983) criticized that approach and offered an alternative: incorpo­ rating a shift variable in only one factor-demand equation. Mullen, Wohlge­ nant and Farris (1988) suggested that a biased (X2-saving) technical change (of the type defined by Muth) could be modeled as an "equivalent" shift of factor-supply functions (i.e., there is some combination of values for ~l and 132 that has effects equivalent to those from a particular value of y). It is not completely clear in what sense(s) the shifts will be "equivalent," however. 4.3.4 Models with More Than Two Factors of Production Three-Factor Models Several studies have provided numerical estimates of the size and distri­ bution of research benefits across three (or more) factors of production (e.g., Mullen, Alston and Wohlgenant 1989). However, the only published alge­ braic solutions are those of Holloway (1989). Those results serve, among other things, to illustrate how quickly the analysis becomes intractable for analytical results (although numerical simulation is always possible) when the number of stages of production increases. Holloway (1989) extended the two-factor case studied by Alston and Scobie (1983) to a three-factor case (a farm product with two marketing stages, processing and distribution). Hol­ loway's (1989) key results are summarized in box 4.4. n-Factor Models Muth (1964), Gardner (1975), Perrin (1980, 1981), and Holloway (1989) all tackled the problem from the primal side (specifying production func­ tions). Wohlgenant (1982) suggested using a dual approach (specifying a cost function instead), and he used it to illustrate solutions for the case of n factors. Several studies have followed that suggestion (e.g., Mullen, Wohl­ genant and Farris 1988; Mullen and Alston 1990; Mullen, Alston and Wohlgenant 1989). In this approach, the equations of the model in the case when n factors are used to produce a single product are specified in logarith­ mic differential form as Final demand: E(Q) = -11 [E(P) - a] (4.17a) (1988), and MuUen, Alston and Wohlgenant (1989). Alston and Mullen (1992) looked at the differential incidence of different ways of funding R&D and different types of technical change in Australian wool. Economic Surplus Methods 265 BOX 4.4: Research Benefits in an Industry with Two Marketing Stages Holloway (1989, p. 341) showed that farmers always gain from increases in final demand or from biased technical change that is farm-product using (i.e., technical change that saves distribution or processing services). Conditions for farmers to gain from other types of research are 1. Increase in the supply of (a) "distribution services": 1) > ad (b) "processing services": (ad-ap) (Ed+1) >s;(ad-1) (Ed+ap) 2. Neutral technical change in (a) "distribution": 1) >1 (b) "processing": (ad-I) (Ed + 1) >S;(ad-1) (Ed + 1) 3. Primary-input-saving technical change in (a) "distribution": 1) >ad (b) "processing": (ad-ap) (Ed+1)>S; (ad-1)(Ed+ap) where ad = the elasticity of input substitution in the distribution industry, ap = the elasticity of input substitution in the processing industry, Ed =t he elasticity of supply of distribution services, 1) = the absolute value of the elasticity of final demand, and s; = the cost share of the intermediate input. n Market clearing: E(P) == L Sj E(~) (4.17b) j=1 Factor supply: E(X) == Cj [E(W;) + ~jJ (4.17c) n Factor demand: E(Xj) == L 11(; E(Wj ) + E(Q) + OJ (4. 17d) j=1 This system consists of 2n+2 simultaneous equations in which the vari­ ables are as previously defined (i.e., W j is the price of factor i, P is the final product price, X; is the quantity of factor i, and Q is the quantity of the product). The parameters are the absolute value of the elasticity of final demand, 11 > 0, elasticities of factor supply, c; (where i = I, ... , n), output-constant own- and cross-price elasticities of factor-demand, 11~, and factor cost shares, S;. The exogenous-shift variables are a final demand shift, a, shifts of factor supply functions, Pi' and shifts of factor demand functions, 0;. This specification has used the assumption of constant returns to scale of the industry production function. The elasticities of factor demand may be expressed in terms of cost shares and Allen partial elasticities of factor substitution (i.e., 11ij == spij for i:t:. j). Restrictions on the parameters can be derived from assumptions of symmetry of the cost function (crij = crj ;) and 266 Economic Surplus Methods homogeneity ofthe cost function in the factor prices Lfol T\~ = O. Using these restrictions, the full set of n2 (output -constant) factor-demand elasticities can be represented by n - I factor shares and n(n - I )/2 elasticities of substitution. Equations 4.17a through 4.17d can be used to solve numerically for the price and quantity effects of a range of types of technical changes in the case where a single product is produced using a variety of factors of production (as described in box 4.5, for example). Then the size and distribution of the total benefits from research may be computed by substituting the results into equations 4.l6a through 4.l6c. 4.4 Market-Distorting Policies and Research Benefits The benefits from agricultural research can be influenced by government policies that distort output and input prices. These distortions can reduce short-run allocative efficiency, can significantly alter the distribution of research benefits, and may influence the size and direction of research investments and technical change in the long run.37 Several recent studies have examined the benefits from agricultural research under a variety of output pricing and other government policies. 38 In one of the first of these studies, Alston, Edwards and Freebairn (1988) analyzed the qualitative implications of a range of commodity price policies for the size and distri­ bution of research benefits under a range of market conditions (e.g., closed economy, small or large country, importerorexporter).39 Their main findings (pp. 285-7) may be summarized as (a) all of the forms of intervention studied modify the pattern of research benefits relative to free trade, (b) world research benefits may be increased, reduced, or left unchanged, depending on the market circumstances and the form of intervention, (c) a government intervention reduces (increases) total welfare gains from a research-induced 37. For example, see Schultz (1977,1978) and Ruttan (1982, pp. 88-90). Mellor and Johnston (1984, p. 558) suggest that "the indirect long term effects of price distortions on the orientation of research and the bias of technical change may well be more important than their adverse effects on short-run, allocative efficiency." Alston, Edwards and Freebairn (1988) explored the effects of price policy on research investments informally. More recently. Gardner (1988) presented a formal political economy model in which research policy and price policies are jointly endogenous. See also de Gorter, Nielson and Rausser (1992), Roe and Pardey (1991), and Alston and Pardey (1991). 38. See Akino and Hayami (1975), Nguyen (1977), Edwards and Freebairn (1981), Alston, Edwards and Freebairn (1986, 1988), Norton, Ganoza and Pomareda (1987), Oehmke (1988a, b), Haque, Fox and Brinkman (1989), Zachariah, Fox and Brinkman (1989), de Gorter and Norton (1990), Anania and McCalla (1991), Martin and Alson (1992, 1993, 1994), Alston and Martin (1992, 1995), Murphy, Furtan and Schmitz (1993), and Chambers and Lopez (1993). 39. In their earlier paper, Alston, Edwards and Freebairn (1986) provided some quantitative illustra­ tions in an application of their analysis to the Australian wool industry. Economic Surplus Methods 267 BOX 4.5: A Nwnerical Solution/or the n-Factor, One-Product Problem Computer programs are available for solving linear systems of simultaneous equations. Alternatively, the model can be solved using matrix algebra methods in a generalization of the solution to the 2-factor, one-product problem (the Muth model) as shown in box 4.3. The first step is to transform the model so that the exogenous shocks are on the right-hand side. In the n-factor, one-product case, the model (Le., equations 4.17) may be represented as follows E(Q) + T]E(P) = T]u s,E(W,) + S2E(W2) + ... + snE(W)n - E(P) = 0 E(X,) - T]i,E(W,) - T]i2E(W2) - ... - T]in(E(Wn) - E(Q) = 0, E(X2) - T]2,E(W,) - T]22E(W2) - ... - T]inE(Wn) - E(Q) = Oz E(Xn) - T]~,E(W,) - T]~2E(W2) - ... - T]~nE(Wn) - E(Q) = on E(X,) - E,E(W,) = E,~, E(X2) - E2E(W2) = E2~2 In matrix form this can be written as MY=X where Y is a vector of the endogenous prices and quantities of interest, of length 2(n+ 1), X is a vector of exogenous shocks of length 2n+2, and M is a (2n+2) x (2n+2) matrix of parameters. That is 0-0 0 0 I T] E(X,) T]u 0-0 S, Sn 0 -1 I 0 1 - 0 -T]i, - -T]in -I 0 E(Xn) 0, I \ I I \ I I E(W,) I o - 1 -T]~, - -T]~n -I 0 I on 1-0 -10, 0 0 0 E(Wn) E,~, I \ I I I E(Q) I 0-1 0 -En 0 0 E(P) En~n The solution vector is, as in the two factor case Y=M·'X. A numerical solution can be obtained after substituting values for the elements into the parameter matrix M and choosing values for the exogenous shocks in the X vector and simulating solutions using any computer program with capability to invert matrices. 268 Economic Surplus Methods supply shift by an amount equal to the increase (reduction) in social costs of the market intervention resulting from that same supply shift.40 Unfortu­ nately, there are no more general rules about the implications of commodity market distortions for the size and distribution of research benefits. Thus, each type of intervention in each market situation must be considered in a case-by-case fashion. More recent studies have extended the range of policies analyzed to include input market distortions (e.g., de Gorter and Norton 1990), to measure the quantitative importance of the issue (e.g., Oehmke 1988b, 1991) and to adjust measures of research benefits to allow for the effects of price-distorting policies (e.g., Haque, Fox and Brinkman 1989). Some further studies in this area have taken a different tack. Gardner (1988), Roe and Pardey (1991), and de Gorter, Nielson and Rausser (1992), for example, have argued that price policies and public-sector research investments are jointly determined in a political-economy process.41 Accord­ ing to their arguments, the instruments of policy are chosen to maximize a weighted sum (rather than a simple sum) of benefits to producers and consumers (who are also taxpayers). From the standpoint of those papers it does not make sense to examine the implications of price policies for incentives to fund research because the price policies themselves are determined jointly with the research policies. This argument is plausible, and integrates research policy into the public­ choice models of agricultural policy that to date have focused on price policy. It does call into question Alston, Edwards and Freebairn's (1988) inferences that, since price policies affect the size and distribution of re­ search benefits and, therefore, the incentives of different groups to fund research, price policies might account for underinvestment in research. It suggests, alternatively, that another factor, differential political power of different groups, accounts for both price-distorting policies and research policies. This line of argument might not work so well in relation to some price distortions (e.g., exchange-rate distortions) that are unlikely to be endogenous to agriculture in the same way as commodity market distortions. In any event, Alston, Edwards and Freebairn' s (1988) results would lead to the same predictions if combined with welfare weights. In the present context, we are not concerned with deducing the optimal combination of research and commodity price policies to maximize a weighted welfare 40. Alston and Martin (1992) prove this proposition formally. 41. Alston and Pardey (1991) reviewed this literature and argued that, given the nature of the timing of the impacts of decisions on research policy and price policy, and given the typical separation of the decision-making bodies for the two types of policies, it is too great a simplification to treat the two policies as being simultaneously determined by a single decision maker. See also Alston, Chalfant and Pardey (1993). Economic Surplus Methods 269 function. Rather, we are concerned with the problem facing research admin­ istrators: evaluating research and setting research priorities treating any price policies as determined elsewhere. For this problem, the issues raised by Gardner (1988), Roe and Pardey (1991), and de Gorter, Nielson and Rausser (1992) are not relevant and the approach suggested by Alston, Edwards and Freebairn (1988) is applicable. Since there are no simple general rules, in this section we provide details on the size and distribution of research benefits in a range of commonly occurring market and policy situations. We use supply-and-demand dia­ grams to illustrate the economic surplus measures of research benefits accruing to domestic consumers, !1CS, domestic producers, !1PS, domestic taxpayers, !1GS, and the ROW (foreigners), !1FS. In addition, total domestic benefits, !1TS, and world research benefits,!1 WS, are measured as aggregates of the other measures. Throughout, we assume a parallel research-induced supply shift in the home country. We show these surplus measures for each market and policy situation, both as areas on diagrams and by formulas to compute those areas in terms of market, policy, and technological parame­ ters. The analysis includes a fairly comprehensive range of typical policies, including (a) price-fixing schemes, such as minimum (target) prices, maxi­ mum (ceiling) prices, or variable import levies, (b) subsidies or taxes on production, inputs, or trade, (c) quantitative restrictions (on inputs, outputs, or trade), and (d) exchange-rate distortions. We begin with a closed-econ­ omy case, then we extend the analysis to the case of a small, open economy. Most countries (or regions within a country) are in one of these two catego­ ries in relation to most agricultural commodities they produce. This is especially so in relation to the relatively long-run horizon in which research benefits accrue. We show a limited number of examples for the more unusual situation in which a country has market power in international agricultural commodity trade. A number of issues, which warrant some consideration in many situations, are put aside in this analysis. First, in the calculations of surplus areas, changes in government revenues are used to represent changes in taxpayer surplus. This ignores the deadweight cost of taxation to raise government revenues (as discussed, for example, by Fox 1985; Dalrymple 1990; Alston and Mullen 1992). Second, the analysis is conducted at a single market level - although this raises no special problems as long as care is taken in the interpretation of the calculated surpluses. Third, the analysis retains Har­ berger's postulates for the surplus measurement after adjustment for the particular distorting policies of interest. That is, we assume no other relevant distortions. 270 Economic Surplus Methods Once we allow for one distorting policy (e.g., a target-price policy) it is tempting to worry about another as well (e.g., the deadweight costs of government spending). Indeed, in some cases the policies must be analyzed as consisting of the joint action of several instruments, as pointed out by de Gorter and Norton (1990) in relation to U.S. farm commodity programs in which supply controls offset the output subsidies. It is tempting, in particular, to include production externalities along with market-distorting policies. But this can go too far. We end up quickly in a second-best world in which we can say little unequivocally about economic welfare effects. This situation arises, in particular, in the context of analyzing the implications of exchange­ rate distortions that are addressed later in this section. The questions of production externalities and sustainability, however, are deliberately kept separate from the question of commodity market policies and are dealt with separately in section 4.5, below. The appendix to chapter 5 includes formulas for computing the price, quantity, and welfare effects of research-induced supply-and-demand shifts in the presence of the types of policy distortions discussed below. 4.4.1 Closed-Economy Examples Price Supports (Minimum Target Prices with Deficiency Payments) The benefits from research in the presence of a simple price-support scheme are illustrated in figure 4.11.42 The output price to producers is supported at PM1N (by government deficiency payments) above the competi­ tive equilibrium price, Po. As a result, the quantity supplied increases from Qo to Qo', and Po' is the price at which the commodity is sold on the domestic market in order to clear that quantity. The government incurs a cost of area PM/~bPo', while the net social cost of this policy is the triangle abc. As a consequence of research, the supply curve shifts from So to S" producers gain area Ir,adI" consumers gain Po'bfPi> the government incurs additional costs due to its price-support policy of adjP,Po'h, and the social cost of the price policy is gi ven by triangle djg. Research benefits are estimated as the change in producer surplus plus the change in consumer surplus minus the change in government cost. Although research-induced changes in total producer and consumer surpluses are larger, the net social benefits to research under this regime, compared with a situation without a price support, are lower by 42. This case has been discussed by Alston. Edwards and Freebairn (1988), Oehmke (1988b), and de Gorter and Norton (1990). Economic Surplus Methods 271 an amount equal to the difference between dfg and abc. Suppose, instead, a government wants to support producer prices but only up to a certain amount of production (which is assumed to be less than or equal to the free trade quantity, Qo).43 The quantity produced in excess of that amount receives only the market price. Such a policy is also illustrated in figure 4.11, where QR represents the quantity on which producer price supports are paid. In this situation, the price policy acts like a decoupled income transfer to producers. The income from price supports is independent of production beyond the supported quantity. Marginal decisions relate to market prices, so producers continue to produce the competitive quantities and research benefits are unaffected by the subsidy transfer. Price Ceilings (Maximum Prices) The price-support policy illustrated above includes subsidies to producers while consumers gain because of lower prices. An alternative "cheap-food" policy, at least in the short run, is to set a price ceiling, say at PMAX, which is below the competitive equilibrium price, Po, as illustrated in figure 4.12. The Figure 4. 11 : Research benefits in a closed economy with a target price and deficiency payments Price PM1N 1----+---:::.....-------- Po' o Quantity 43. There are two alternative situations where QR is greater than the competitive quantity. When it is greater than the quantity supplied at the minimum price, Qo', the quantity limitation is irrelevant. When it is between Qo and Qo', it is binding on producers and consumers. 272 Economic Surplus Methods result is a reduction in production to Qo' and an increase in quantity de­ manded to Co'. The implications of this response for consumption and welfare depends on how the government acts to clear the market (e.g., see Alston and Smith 1983). Unless the excess demand (Co' - Qo') is satisfied somehow (e.g., by imports that are purchased by the government), shortages will occur in the market and some other mechanism (e.g., queueing or black markets) will be necessary to clear it. A variety of policies have been used by governments when (maximum) price ceilings have been imposed to ration demand. For purposes of illustration, we assume the government buys imports (at a price PM above the regulated maximum price) and makes them available at PMAX' Thus, in the initial (distorted) equilibrium situation, the government incurs a net consumption subsidy cost equal to the price differ­ ence (PM - P MAX) per unit times the quantity of imports (Mo =C o' - Qu') . When research causes supply to shift from So to SI> the quantity supplied increases from Qo' to QI. consumption is not affected, and imports fall by this amount to M,. The benefits from research are equal to a gain in producer surplus of area locd/, and a gain to the government of (PM - PM AX)(Q, - Qo') of reduced subsidy expenditure.44 In the absence of the price ceiling, total research benefits would be equal to area I('pbl, and would be shared between producers and consumers depending on the elasticities of supply and de- Figure 4.12: Research benefits in a closed economy with a maximum price ceiling Price PMM ~-------------- :":(--~_MO .:, If.:c- ------,- M I -----+! o Quantity 44. If the supply curve shifts out beyond point e in figure 4.12, the policy becomes irrelevant and the calculation of research benefits is slightly more complicated. Economic Surplus Methods 273 mand. Clearly, consumer research benefits are reduced (i.e., eliminated) by this price policy. On the other hand, the government becomes a beneficiary of research under a price ceiling. The effects of the price policy on research benefits to producers and the nation are ambiguous, depending upon the size of the price distortions between the regulated price, PM AX' the unregulated price, Po, the cost of imports, PM, and the relative sizes of the elasticities of supply and demand. Producer research benefits could be greater under this price policy than under a competitive closed-economy arrangement. Subsidies on Inputs or Output The effects of an output subsidy on the size and distribution of research benefits in a closed economy are illustrated in figure 4.13. This case is described in detail by Alston, Edwards and Freebairn (1988). Here we consider a research-induced shift down of supply by k per unit, with and without an output subsidy. An output subsidy or negative tax (- 'rQ per unit) shifts the commodity supply curve down from So to So'. Research shifts the supply curve from So to SI without the subsidy and from So' to S/ with the subsidy. In summary, the effects of the subsidy policy on research benefits are to change the distribution but not the size of total benefits. Producer and consumer benefits from research are greater but this is exactly offset by the increase in government costs as a result of research. By the same token, research does not change the social costs of the subsidy policy - initially the social cost is triangle ace; after the research-induced supply shift, it is triangle bdf(which is equal to ace). The effects of output taxes are symmetric but opposite. The producer research benefits under this policy are shown as area P/d hg and consumer research benefits are equal to area Po' cdPI '. Total research benefits are equal to the sum of these areas less the increase in government subsidy costs due to research, the subsidy per unit multiplied by the research-induced change in output I 'rQI (Q/ - Qa'), where 'rQ < O. In figure 4.13 the total research benefit - with and without the subsidy - is also given by area loabll • Developing countries often subsidize inputs such as fertilizer and pesti­ cides. In some instances an input subsidy is exactly equivalent to an output subsidy. Assuming that all producers use the same (fixed) amount of the input, X, per unit of production, Q, before and after the technical change (i.e., it is a neutral change with fixed factor proportions), an input subsidy or negative input tax of - ~ per unit of input (where ~ < 0) is identically equi valent to an output subsidy of - 'rQ = - ~ (XI Q) per unit of outpUt.45 The 45. Of course this is a very restrictive special case. When input subsidies change relative factor prices, 274 Economic Surplus Methods Figure 4.13: Research benefits in a closed economy with a per unit subsidy p) PI) 10 P' ) g I) i · t Q II') 1 I ') 0 Quantity shift down in the product supply curve would be equal to the shift down in the factor supply curve multiplied by the number of units of the factor used perunit of the product (i.e., XlQ) . Equivalently, the percentage (parallel) shift of output supply could be determined by mUltiplying the percentage cost reduction in the input supply curve by the share which that input represents in the cost of producing the commodity. Under different assumptions (i.e., biased technical change or variable factor proportions), the effects of input price policies may be quite different from output price policies, requiring additional work to evaluate the effects. Output Controls The policies discussed so far all involve the use of government revenues. Supply controls are sometimes used as a way of supporting producer in­ comes at the expense of consumers, avoiding any budget costs, The idea is to restrict supply to the market and thereby raise prices. In many cases the and when there are opportunities for factor substitution, the input mix wiU change in response to an input subsidy and there wiU be a greater output supply shift than in the case of fixed factor proportions. On the other hand, the benefits from greater precision in aUowing for this substitution effect in the context of measuring research benefits are unknown and in most cases it would not be feasible or worthwhile to go beyond treating input subsidies as equivalent output subsidies. Economic Surplus Methods 275 controls are applied to inputs (e.g., land used to grow a crop or the number of livestock on hand) rather than outputs as such; the usual explanation is ease of enforcement. Input controls are identical to output controls when there is no opportunity to substitute inputs in order to increase yields and reduce the constraint of the policy. In most cases there are some opportuni­ ties for substitution, and "slippage" becomes a problem. As with input subsidies, substitution among inputs complicates the economic welfare im­ plications of input controls. Where it is thought to be important to do so, the effects of input subsidies or input controls with input substitution could be explored formally using models of the types developed in section 4.3. Here we consider only explicit output controls (or equivalent input controls with fixed proportions). The case of an output quota in a closed economy is illustrated in figure 4.14. When a quota restricts output from Qu to Qu', the price paid by consumers rises from Po to Po' and quota owners receive a quota rent equal to area Po' abc. Often, but not always, quota owners are producers. When research causes supply to shift from So to St, there is no effect on quantity supplied or on price. All the research benefits in this case accrue to quota owners in the form of increased quota rents (which increase to area Po' ab' c'); there are no research benefits either to consumers or to producers per se. The total research benefit (the increase in quota rents) is equal to the research-induced cost saving per unit multiplied by the quota quantity. In the absence of the quota, research benefits would be equal to area lodelt• Total research benefits are lower under the quota, compared with the unregulated situation, by an amount equal to area bdeb'. 4.4.2 The Small-Country Trader Case The models of effects of price-distorting policies discussed thus far have assumed a closed economy. The more typical example is one where trade occurs or would occur in the absence of trade-distorting policies. In addition, in most cases, the small-country assumption is appropriate. The impact of research on a small-country importer or exporter of a commodity in the absence of market-distorting policies was described earlier (see figure 4.5). In the closed-economy case, we considered policies of (a) price fixing, (b) input and output subsidies (and taxes), and (c) output controls. In the context of traded goods, we can consider all of these policies, and in addition, there are border policies including (a) trade taxes (or subsidies) and (b) quantita­ tive restrictions on trade. Often a country will use a mix of policies for a traded good - in particular, to be effective, domestic policies often require the assistance of an embargo (or some other restriction) on imports. 276 Economic Surplus Methods Figure 4.14: Research benefits in a closed economy with an output quota Price c D c I, o~ --------------~----~~---------------- Q~ Quantity Price Supports (Minimum Prices) or Output Subsidies The benefits from research in the presence of output price supports (i.e. , target prices with deficiency payments or the equivalent per unit output subsidy) in a small country (exporter or importer) are illustrated in figure 4.15.46 Before the research, with prices supported at PM IN, producers supply QUi. Domestic consumers continue to face the world price, Pw, and consume Co, as if there were no policy . The government incurs a subsidy cost of deficiency payments equal to (PMIN - Pw)Qu/. After the research, production increases from Qo' to Q I and exports increase (or imports decrease) by the same amount. As a result of research, producer surplus rises by area lr,cdl l • Government subsidy costs rise by an amount equal to the per unit subsidy times the increase in quantity supplied (PMIN - PW)(QI - Qo/), which in this instance, is equal to area acdb. National research benefits are unaffected but the distribution is changed by the policy. Specifically, the increase in pro­ ducer surplus is greater by the amount of the increase in government subsidy costs, which is used as a measure of the reduction in taxpayer surplus, !lGS. (4.18a) 46. The size and distribution of benefits in the case of free trade are iUuslJated in figure 4.5 and described in the accompanying text. Figure 4.15: Research benefits in a small open economy with price supports (or output subsidies) N '-l '-l (a) An exporter (b) An importer Price Price So SI P M1N PM1N PW I It Pw 10 10 D 11 II D' o Co Qo QO QI Quantity o Qo QO QI Co Quantity 278 Economic Surplus Methods !J.PS = PMJN Qo' K' (1 + O.SK'£') (4.18b) !J.TS = !J.PS + !J.GS (4.18c) where £' is the elasticity of supply and K' is the proportionate supply shift effect both defined at the (generally observed) preresearch, distorted equilib­ rium price and quantity (i.e., PMJN, Qo'). The results are identical when an output subsidy of (PMJN - Pw) per unit is used instead of the price support and deficiency payments. As in the case of the closed economy, under some circumstances input subsidies are equiva­ lent to output subsidies, and this model could also be used to represent the research benefits in the presence of input subsidies. Output Price Ceilings (Import Subsidies or Export Taxes) When a small country imposes a price ceiling below the world price to protect domestic consumers, it usually must also introduce a trade barrier or tax in order to prevent the policy from being undermined by trade. Assuming that producers and consumers face the same (maximum) price, there will be increased consumption, lower production, and lower exports (increased imports) due to the ceiling. The example of research combined with a price ceiling is illustrated in figure 4.16. In this analysis it is assumed that the government effectively taxes exports (subsidizes imports) to ensure domes­ tic market clearing at the regulated maximum price, PM AX' Although this policy does not change the total net domestic research benefits, it does affect the distribution. In the context of given world prices, consumers do not benefit from research - whether there is a maximum price or not. Producer benefits are lower but government revenue is greater because a research-induced supply shift leads to reduced subsidy costs (in the case of imports) or increased tax revenue (in the case of exports). Of course, if there were different approaches taken by the government to clear the market, entirely different results could obtain. Before the research, with prices fixed at PMAX, producers supply Qo' and consumers consume Co'. In the case of an exportable good (panel a), the government imposes an export tax and generates tax revenue of (Pw - P MAX)(QO' - Co'); in the case of an importable good (panel b), the government incurs a subsidy cost of (P w - P MAX) ( Co' - Qo'). After the research, production increases from Qo' to Q, and exports increase (or imports decrease) by the same amount. As a result of research, producer surplus rises by area IoabIJ• Government revenues rise (or subsidy costs fall) by an amount equal to the per unit tax (or subsidy) times the increase in quantity supplied domestically (Pw - PMAX)(Q, - Qo'), which is equal to area acdb. Figure 4.16: Research benefits in a small open economy with maximum domestic prices N ~ (a) An exporter (b) An importer Price Price So Sl PW 1 \ 7(' ,< Pw PMAX J - - - - - -. P MAX I I . o II ·D 0 11. I I I 0 Co CO QO QI QO Quantity 0 QO QO QI CO Co Quantity 280 Economic Surplus Methods National research benefits are unaffected but the distribution is changed by the policy. Specifically, the increase in producer surplus is smaller by the amount of the increase in government revenue in the case of an exporter (or by the amount of the decrease in subsidy costs in the case of an importer), the taxpayer benefit, flGS. The formulas are (4.19a) flPS = PMAX Qo'K' (1 + O.SK'E') (4.19b) flTS = flPS + flGS (4.19c) where E' is the elasticity of supply and K' is the proportionate supply shift effect defined at PM AX' Qo'. These results are identical to those that would hold if an export tax (for an exportable) or an import subsidy (for an importable) of (Pw - PMAX) per unit were used as such without setting a ceiling price. As in the case of the closed economy, when different policies are used in conjunction with a ceiling price to clear the market, they could have entirely different effects from what has been shown here. Import Tariffs and Import Quotas Import tariffs are widely applied in agricultural markets, in developed and developing countries alike. The range of tariff policies employed includes ad valorem (percentage) and specific (per unit) tariffs and they may be either fixed or variable (i.e., most tariff rates are fixed but some countries use variable tariff rates, such as the European Community variable levies). For present purposes, the differences among types of tariffs are relatively unim­ portant, especially in the small-country case.47 We wiJI consider an ad valorem tariff, only. In some contexts an import quota has effects identical to those of an import tariff but, as we shall see below, not in the context of measuring the size and distribution of research benefits. Figure 4.17 shows the effects of a tariff (1 OaT percent) on imports by a small country. The tariff raises the domestic price from P w to (1 +n p wand generates tariff revenue for the government, equal to TPv/.Co' - Qu') in the absence of research. Producer benefits from research, with the tariff, are 47. Differences between per unit and ad valorem tariffs arise when the world price changes (i.e., the two are equivalent only for a particular value of the world price). Similarly, there is a particular import quota that corresponds to a particular tariff (of either sort), given particular positions of the functions and competition. However, this equivalence breaks down when there is imperfect competition or when the curves shift (due to growth, for instance). In a comparative-static, competitive model the three policies are equivalent in the small-country case because the world price is constant. Economic Surplus Methods 281 Figure 4.17: Research benefits in a small importing country with a tariff u Qo c­o Quantity equal to area fncdf,. The research-induced supply shift reduces imports from (Cn' - Q()' ) to (Cn' - QI) and reduces tariff revenue by area cdgf= area acdb = TPJQ, - Q()') .48 The total benefit from research (area Irpbl ,) is unaffected by the tariff but the producer benefit is greater, offsetting reduced govern­ ment revenues from tariffs. The formulas for research benefits in this case are identical to those for a support-price or output-subsidy scheme - equations 4.18a through 4.18c - after we replace the support price, PM IN, with the tariff-ridden price, ( 1+1)Pw: !1GS = -TPw Qo'K'£' (4.20a) !1PS=(1 +7)Pw Qu'K'(1 +O.5K'£') (4.20b) tJ.TS = tJ.PS + tJ.GS (4.20c) This equivalence of formulas ari ses fro m the equivalence of the policies in their effects on the size and distribution of research benefits. In the former case, research increased output-subsidy costs; in the latter, it reduced gov­ ernment revenues from tariffs. In both cases, the increased private benefits for producers exactly matched the reduction in government revenues. Now we consider the effects of an import quota that is sometimes used as a substitute for a tariff. In most analyses, the main difference between a tariff 4~ . Note that the change in quantity due to the tariff is Qn' - ~) =T E~) where e is the supply elasticity de fined at the preresearch, undisto rted equilibrium. Alternatively the change in quantity Q()' - Q() = TE' Q()' / ( 1+1) where e' is the elasticity of supply defined at the preresearch, distorted equilibrium price and quantity (i.e., ( 1+1) P w' Q()/). See appendixA 5.2 for more details. 282 Economic Surplus Methods and an import quota is that the tariff generates tariff revenue for the govern­ ment whereas the import quota generates quota rents, which are often private benefits. Thus, a tariff and a quota are in some senses equivalent, and often are regarded as such. An import quota is illustrated in figure 4.18. Initially, the quota is set at Mo (corresponding to [Co' - Qo'] in figure 4. 17 if the quota were set to be equivalent to the tariff rate, n. What happens when research causes supply to shift from So to SI? In the case of the tariff, increased domestic production was accommodated with reduced imports. In the case of the import quota, imports will continue at the quota level, Mo, as long as there are quota rents to be earned, and any research-induced increase in production must be sold on the domestic market. Thus, unlike the case of the tariff, in this case research causes a decline in domestic price (from Po' to PI)' As a result, there are now some consumer benefits, area Po' abPI> but pro­ ducer benefits (shown by area Plcde) are lower. Quota rents to owners of the import quota are reduced by the amount of the decline in the domestic price multiplied by the import quantity, (Po' - Pw)Mo. The research has led to a reduction in the price distortion (the distortion in both production and consumption) due to the import quota. Initially, the net social cost was triangle fgh plus triangle amk; after the research-induced supply shift, it is triangle cji plus triangle bml. Compared with free trade, research benefits are greater by the amount of the research-induced reduction in the triangles of deadweight loss associated with the import quota policy .49 In the absence of Figure 4.18: Research benefits in a small country with an import quota D () Quantity 49. This result was asserted by Alston, Edwards and Freebaim (1988), who illustrated a number of cases, and proved by Alston and Martin (1992) as a general second-best proposition. Economic Surplus Methods 283 the quota the total research benefits (also equal to the national research benefits in this case) would be area IohiIl; with the policy, total research benefits are equal to area Ir/liIl plus lfgh - cji) plus (amk - bml). Output Controls In the case of a small-country trader, output controls would seem to serve little purpose - they involve a net social cost and there are no clear beneficiaries. Output controls reduce the domestic quantity produced and the balance of trade, and research benefits continue to be concentrated in the hands of producers (quota owners). How do we explain the pervasive use of measures to control supply in countries that seem to have no market power in trade? The most likely explanation is that the output control is only one component of a package of policies (e.g., as a complement to a subsidy scheme to limit government obligations, as discussed above). There are relatively few examples of a pure supply control program in a small-country setting. Hence, it is important to model the actual policies accurately. When a quota scheme is applied in the context of a small trading country, it will usually be in conjunction with a prohibitive trade barrier (e.g., an embargo against imports, as is common for fresh milk or eggs) and the correct form of analysis will be as described above for the case of a closed economy with additional steps to account for research-induced changes in the social costs of the embargo. Revenue-Pooling Price-Discrimination Schemes As a final example, many countries have schemes that price-discriminate against the domestic market and pay producers a weighted average of the (high) home price and the (low) export price.50 In figure 4.19, the domestic price to consumers is fixed at Pd, above the world price Pw' Producers are paid a weighted average of these two prices so we can represent average revenue - the effective demand facing producers - with the pooled price line, aPp' which asymptotically approaches the export price line from above. Equilibrium is now defined by the intersection of supply with the pooled price line at Po' and Qo' before a research-induced supply shift from So to SI. and PI and QI afterwards. Whether the policy is applied or not, all of the research benefits accrue to producers because research does not affect the price paid by consumers in either case. Thus, in either case, all of the research-induced increment to production is sold on the export market. 51l. Alston and Freebaim (1988) provide a comprehensive discussion of this type of scheme. Alston, Edwards and Freebaim (1988) show the results described here. See also Freebaim's (1992) study of the Australian dairy industry. 284 Economic Surplus Methods Figure 4.19: Research benefits in a small country with a revenue-pooling price-discrimination scheme Price "o ~------~--~------------~--~-------- c~ Q~ Quantity In the absence of the policy, the total producer (and national) research benefit is equal to area lobel,. Under the policy there is a deadweight cost of resource distortions in production, shown as triangle bcd before research and the smaller triangle efg afterwards. In this case, as with an import quota, an added benefit is that research reduces the social cost of the policy. Using Alston and Martin's (1992) results, total producer (and national) benefit from research is equal to the benefit in the absence of distortions, area lobel" plus the change in cost of distortions induced by the supply shift, bcd - efg. 4.4.3 Large-Country Trader Models We have examined the effects of a range of policies in the context of a closed economy or a small open economy. These cover the vast majority of situations likely to be encountered in the context of work on research priority setting and evaluation. In a few cases, an individual country will produce enough of a commodity that it can affect international prices, but such cases are rare, especially over the relatively long run, which is the appropriate context for research evaluation and priority setting. Often, when a country is treated as having market power in a commodity market, the context is the short to medium term. Even when countries have some market power in the Economic Surplus Methods 285 short to medium term, they usually do not have significant market power in the relevant long run. In most cases, therefore, even when it seems that a country might have some market power in international trade, a small-coun­ try model is better for evaluating research and setting priorities.51 In this section we illustrate the effects of selected policies in the case when domestic policies and output can affect international prices. A key point to be made here is that, when a country can affect its international terms of trade in a commodity market, free trade in that commodity may not maximize domestic welfare - even when we maintain all of the standard assumptions. When a country has some monopoly or monopsony power in international markets, it can gain by exercising that power. When the private sector is competitive, and therefore cannot take advantage of that market power, the government can intervene to achieve an equivalent result (from the point of view of a nation as a whole) through the application of optimal export taxes or optimal tariffs. The story becomes more complicated when we consider the potential for strategic interaction among governments and retaliatory policies, especially when the industry is not purely competitive. 52 The distinction between market power of firms and market power of nations becomes especially important in this context. We retain the assumptions of competitive behavior of firms and assume no retaliation. Then, a nation with market power in trade may benefit by applying an import tariff, an export tax, or an equivalent policy. Alston, Edwards and Freebairn (1988) showed that generally benefits from a research-induced supply shift are greater or smaller under a price policy than under free trade, to the extent that research reduces or increases the magnitude of the social costs of divergence from the policy that maxi­ mizes domestic welfare. For a closed economy or a small-country trader, the optimal policy is free trade; for a large country, it is the optimal trade tax. We illustrate these ideas with four examples: (a) output price supports for an exporter, (b) export taxes, (c) import tariffs, and (d) supply controls. Output Price Supports The benefits from research in the presence of output price supports in a large exporting country are shown in figure 4.20. This is virtually identical to figure 4.10, which represents the closed-economy case. Here we identify the domestic component of total demand as Dd , and the curve Do represents 51. This line of argument implies that an explicitly dynamic treatment of the response of supply and export demand to price changes (i.e., dynamic price elasticities) is required for optimal exploitation of market power in trade in general as well as in relation to understanding the interaction of research-induced supply shifts with trade policy. 52. For example, see Vanzetti (1989) and McNally (1993). 286 Economic Surplus Methods total demand - the sum of domestic and export demand (in the case of a nontraded good in figure 4.10, the corresponding total demand was entirely domestic demand). The output price to producers is supported at PMIN (by government deficiency payments) above the competitive equilibrium price, Po. As a result, quantity supplied increases from Qo to Qo' and Po' is the price at which the commodity is sold on the domestic and export markets in order to clear that quantity. The government incurs a cost of area PM IN abPo'. As a consequence of research, the supply curve shifts from So to SI' producers gain area load/l' domestic consumers gain Po'cePJo and the gov­ ernment incurs additional costs because of its price-support policy of adfPIPo'b. Research benefits are estimated as the change in producer surplus plus the change in consumer surplus minus the change in government cost. Although research-induced changes in total producer and consumer sur­ pluses are larger, the net social benefits from research under this regime, compared with a situation without a price support, are lower. Here, research widens the disparity between the actual policy and the optimal policy in two ways. It widens the divergence between domestic costs and prices while, at the same time, it increases the effective subsidy on exports when the optimal policy is to tax exports. Thus the exporter's research benefits are lower than they are with free trade - indeed, research benefits could be negative (e.g., Oehmke 1988b, 1991).53 However, ROW research benefits are greater in the presence of an export subsidy in the innovating country. Export Taxes Less-developed countries often impose export taxes either to raise gov­ ernment revenue or as part of a "cheap food" policy. The implications of agricultural research in the context of an export tax in a large country are illustrated in figure 4.21. The initial quantities produced, consumed, and exported by country A are Qo, Co, and QTo. First we introduce an export tax of 1E per unit. Then we introduce a research-induced supply shift. The export tax is represented as a shift down in the ROW excess-demand curve from EDB.ot o EDB.1 such thatthe preresearch net oftax export (and domestic) price falls from Po to Po'. As a consquence of research, the supply curve shifts from So to SI' At the same time, the excess-supply curve shifts from ESA .o to ESA•I 53. The idea of negative returns to research is not new. The literature on irnrniserizing technological change includes articles by Bhagwati (1958, 1968), Gruen (l %1), and Johnson (1958, 1967)that illustrate the possibility of irnrniserizing effects due to tenns-of-trade effects or to government intervention in rnrukets. Some recent articles have shown examples from agriculture (e.g., Oehrnke 1988b, 1991; Chambers and Lopez 1993; Murphy, Furtan and Schmitz 1993) where a negative return may be attributable to distortions in commodity rnrukets. Alston and Martin (1992) synthesize the recent literature and the older, irnrniserizing growth literature. Economic Surplus Methods 287 Figure 4.20: Research benefits in a large country with a target price and deficiency payments Price PM1N r-~------------~~------~ Po P~ o c~ Q1 Quantity and the final price is PI' Relative to the tax-distorted equilibrium, research has caused an increase in domestic consumption (from Co' to CI), an increase in production (from Qo' to QI)' and an increase in exports (from QTo' to QTI)' Associated with these changes is a gain in producer surplus equal to area Plcde in panel a and a gain in consumer surplus equal to area Po' abP I in panel a. If we were to ignore the tax, the sum of these two areas would be a measure of total domestic research benefits. However, the research-induced supply shift also leads to an increase in tax revenue equal to area fghi in panel b (because exports increase), which is an additional component of domestic benefits. Exactly offsetting this, from the world standpoint, is a reduction in research benefits to the ROW. Import Tariffs The implications of agricultural research in the context of a tariff imposed by a large importing country are illustrated in figure 4.22. The initial quantities produced, consumed, and imported by country A are Q(l> CO, and Q~) . First, we introduce an import tariff of ~ per unit. Then we introduce a research-induced supply shift. The tariff is analyzed as a shift up in the ROW excess supply curve from ESB•n to ESB•I such that the preresearch, duty-paid Figure 4.21: Research benefits for a large-country exporter with an export tax gNg (a) Domestic quantity produced, consumed, (b) International trade and traded Domestic Domestic price price So ESA,O Po ESA,I Po' PI e o Co C~CJ Q~ Q J QO Quantity 0 QT~ QTJ QTO Traded quantity Economic Surplus Methods 289 import (and domestic) price rises from Po to Po'. 54 As a consequence of research, the supply curve shifts from So to SI' At the same time, the excess-demand curve shifts from EDA,o to EDA,I and the final price is PI' Relative to the tax-distorted equilibrium, research has caused an increase in domestic consumption (from Co' to C I ), an increase in production (from Qo' to QI) and a decrease in imports (from QT;/ to QTI)' Associated with these changes is a gain in producer surplus equal to area Plcde in panel a, and a gain in consumer surplus equal to area Po' abP I in panel a. If we were to ignore the tariff, the sum of these two areas would be a measure of the total domestic research benefits. However, the research-induced supply shift also leads to a reduction in tariff revenue (because imports decrease) equal to areafghi in panel b, which offsets some private domestic benefits. Supply Controls For a large-country exporter, supply controls have some of the virtues of export taxes (for example, see Sumner and Alston 1986; Johnson 1965): they do allow the monopolistic exploitation of foreign markets but they involve some distortions in domestic consumption that the export tax avoids. Output quotas greatly simplify the problem of measuring research benefits. The same result holds regardless of whether the country is an importer or exporter, large or small. So long as the quota is binding on production - with and without the research - the total research benefit is equal to the cost saving per unit multiplied by the total quota quantity. This total benefit accrues entirely to quota owners who may also be producers. Thus, the model described in detail for the closed-economy case is equally valid for the small- or large-country trader, although as discussed above, pure supply controls are unlikely to be used in isolation from other policies by a small­ country trader. A More General Approach The four cases discussed in this section illustrate a general point. When measuring research benefits (with or without policy-induced price distor­ tions), there is one set of formulas to measure the welfare consequences that can be applied in every case. The implications of price distortions are that (a) different market participants may face different prices, (b) the policies may affect the research-induced price changes, and (c) government revenues may 54. This policy could be analyzed, alternatively, by treating the tariff as an equivalent shift of the importer's excess demand for imports. As always for competitive industry, the ultimate incidence of a tax does not depend on its initial incidence. The choice is arbitrary but the shifting ofe xporter's excess supply yields a clearer picture. Figure 4.22: Research benefits for a large-country importer with an import tariff to-.) ~ (a) Domestic quantities produced, consumed, (b) Excess supply, demand, and trade and traded Domestic Domestic price price ESB,J P0' PI I ~--------T ~ ~ V ESB,O Po e A I I :\ . hX: '- EDA,O I I I I I 10 r / I I I Dd iJ1" ~ EDA,J I I I I I I I I I I I I I I II r I I I , 0 Qo Q~ QI C~CICO Quantity 0 QT QTO QTO Traded J quantity Economic Surplus Methods 291 change as a result of the policies and effects on government revenues must be accommodated. Some relati vely general formulas for components of the domestic welfare change due to research in the presence of policies are dPS = PpQp[E(Qp)/E][1 + 0.5E(Qp)] (4.21a) dCS = P cQdE(QcVrlHl + O.5E(QC>] (4.21b) dGS = 'tc Qc E(Qc) + 'tp Qp E(Qp) + 'tr Qr E(Qr) (4.21c) dTS =d PS + dCS + dGS (4.21d) where Pp and Qp denote the producer price and quantity before research, Pc and Qc denote the consumer price and quantity before research, Qr is the quantity traded, E and 11 are the absolute values of elasticities of supply and demand, and 'tc, 'tQ, and 'tr are per unit taxes on consumption, production, and trade (for subsidies they can be negative numbers). What is needed to complete the analysis is a model of the market that can be used to calculate the relative changes in quantities produced, consumed, and traded, given by E(Qp), E(Qc), and E(Qr), respectively. Specific reduced-form formulas for research benefits under particular policies and market structures can be derived (as done above, for example, in equations 4.18, 4.19, and 4.20) by substituting analytic solutions for the relative changes in quantities into equations 4.21 a through 4.21 d or their equivalent. However, it may be more useful and appropriate not to do that. Instead, in many cases, it will be better to keep the model of research-induced price and quantity effects separate from the calculation of the welfare effects associated with those market adjustments, as is shown in the appendix to chapter 5. 4.4.4 Overvalued or Undervalued Exchange Rates Exchange rate distortions are pervasive, especially in agricultural markets in less-developed countries. The most common scenario is one where the exchange rate is overvalued (i.e., fixed above the rate that would be reached if the currency were freely floating). In such cases it is effectively a tax on all exports (and a subsidy on all imports) although it does not generate any tax revenue as such. An undervalued exchange rate has the opposite effect. Sometimes a country will use different official exchange rates for different commodities or sectors and these exchange-rate distortions - in relation to the market exchange rate - amount to discriminatory taxes. To fully analyze exchange rate distortions requires a general-equilibrium model with the macro­ economy explicitly treated. For the present, we will use a partial-equilibrium approach and, where appropriate, introduce general-equilibrium connotations. 292 Economic Surplus Methods First, consider the case where an official exchange rate (overvalued relative to the market rate that applies to goods in general) is applied only to one commodity that is exported by a small-country trader. What are the implications for research benefits? The answer depends on the mechanism used to enforce the policy. Suppose the government insists on changing all the foreign currency earned from exporting the good in question into local currency at the official rate. This is a typical policy when a specific official rate is applied to a particular commodity; it is tantamount to an ad valorem export tax on the commodity at a rate given by the ratio of the official and market exchange rates. This case can be seen by reinterpreting panel a of figure 4.16 so that P w denotes the world price in domestic currency at market exchange rates and P MAX denotes the corresponding price using an overval­ ued official exchange rate. Differences between a policy of an overvalued exchange and an export tax may arise in the greater potential for tax avoidance through falsifying invoices or through barter trade under the former. When an overvalued exchange rate is enforced costlessly in a way that enables the home-country government to keep the difference between export revenues and payments to producers, the appropriate measures of research benefits are those that would apply for an explicit export tax. Changes in government revenues (area acdb in figure 4.16a) must be incorporated explicitly in total benefits if private benefits are measured using the distorted prices (area fr/lbf t ); an equivalent measure of total benefits is obtained by using the undistorted prices (and the corresponding quantities) to measure what benefits would have been without the distortion (area f(.pdf t ). An extreme alternative situation is one where all of the "export tax" revenue is wasted from the home-country viewpoint, either because the policy is operated in such a way that it creates rents for foreign importers or because all of the rents are dissipated in "tax" avoidance, black markets, and so on. In such a scenario, the only research benefits that exist are the private benefits measured by the supply shift relative to the distorted "net of tax" price received by producers (i.e., area Irpbft in figure 4.16a), and the poten­ tial "tax revenue" benefit (area acdb) is wasted. When enforcement of exchange rate rules is costly, something that lies between these two alterna­ tive measures may be appropriate. Now consider the case where the exchange rate distortion is not discrimi­ natory but applies to all traded goods. This creates two additional problems for the analysis of research benefits for a particular good. First, there will be general-equilibrium feedback of the effects in all commodity and factor markets, displacing both supply and demand for the good in question in ways that may be difficult to quantify. Second, it becomes less clear what happens Economic Surplus Methods 293 to the potential rents or implicit tax revenues created by the policy. As before, in the partial equilibrium context, if the "rents" are not collected by the government, they may be lost from the country altogether, either through waste or to foreigners. Given these problems, and additional problems that are encountered when an attempt is made to define and deduce a measure of the "market exchange rate" that would apply if official rates were absent,55 it may be best simply to ignore the existence of general exchange rate distortions when research benefits are calculated using partial-equilibrium commodity models. Moreover, even when a country does not have an explicit policy of exchange rate management, other trade-distorting policies (such as general tariffs against imports) have similar effects through their effects on the balance of trade and exchange rates. For example, from the point of view of exporting industries, a general tariff on imports leads to an appreciation of the exporter's currency and is tantamount to an export tax applied to all exports. Sometimes an explicit exchange rate policy serves to counter other trade-distorting policies. It may be more misleading to account for only part of the policy set rather than ignore it altogether. Thus, accounting for economywide policies such as those affecting exchange rates and general trade policies may be too difficult to be worth attempting. This is an example of the more general second-best problem. 4.5 Sustainability Issues and Other Externalities A further potential source of distortions in incentives is externalities in production. While there has been some work on the general economic implications of environmental externalities in agriculture, we are unaware of any studies of the implications for returns to research.56 Externalities from agricultural production - and broader, related, environmental or "green" issues such as sustainability, global warming, preservation of wilderness areas, animal welfare, food safety, and species preservation - have received increasing attention in discussions of agricultural policy in recent years. That trend can be expected to accelerate. All of these issues, conceptually at least, can be considered in the framework of a conventional supply and demand model, allowing for a divergence between private and social costs or benefits from production. 57 This is similar to incorporating price-distorting policies 55. For example, see Krueger, Schiff and Valdes (1988, 1991). 56. See Pingali and Roger (1995) for an example of the environmental implications of agriculturaI research. 57. As Summers (1992, p.71) recently observed, "Certainly, the idea of sustainable development has 294 Economic Surplus Methods but different in that the distortions are not the creations of governments. We begin with an iIIustrative example of an environmental externality and then discuss some issues that relate to agricultural research in this context. 4.5.1 Research Benefits in the Presence of Environmental Externalities There are many types of external effects in agriculture, and we have already considered two types of spillovers - price spillovers and technology spiIIovers. An externality arises when there is a spiIIover effect of one person's actions on another person's economic opportunities and where that effect is not fully compensated through a market transaction. Price spillovers are not externalities, but technology spillovers usually are. Here we are concerned with externalities of a different type. One example would be where the use of agricultural chemicals on a crop suppresses the population of beneficial natural predators and increases the costs of pest control (in that industry or in some other industry), but individual producers disregard that external effect when making their pest-management decisions. Another example is where agricultural production causes pollution of groundwater with agricultural chemicals or salt or causes pollution of surface water with eroded soil, salt, or agricultural chemicals. A third and more abstract exam­ ple is when the clearing of rainforests results in species depletion, reduction of pristine wilderness, or other environmental damage that is regarded as a cost by some people but where that cost is not considered in decisions about clearing the forest. In all these cases, the social cost of agricultural produc­ tion is greater than the private cost perceived by farmers. In most cases, the externality is borne within the country (or region) within which it is created, although not necessarily by consumers or producers of the commodity in question, or even within the agricultural sector. In some cases (such as global warming), the effects are also borne by foreigners. A further example is where the use of an agricultural chemical leaves residues on food that are hazardous to consumers but difficult to detect. Here the cost of the residue may be borne by the consumers of the product in the first instance, but eventually, if the practice continues, all producers might experience a loss of markets due to consumer concerns about the unreliability of the product. Figure 4.23 illustrates supply and demand for a commodity produced by a small-country exporter in the case where there is a negative externality drawn attention to environmental problems that were ignored for too long. But there is no intellectually legitimate case for abandoning accepted techniques of cost-benefit analysis in evaluating environmental invesbnents ... The answer does not lie in blanket sustainability criteria, or in applying special discount rates, but in properly incorporating environmental costs into the appraisal of projects." Economic Surplus Methods 295 associated with production (e.g., pollution created by the use of agricultural chemicals). The preresearch supply curve and demand curves are So and D, and the market clears at the world price Pw with quantities produced, consumed, and traded equal to Qo, Cu, and QTo. However, the marginal social cost of production is given by So' and it does not coincide with supply. Instead, marginal social cost is greater by E per unit (which is a measure of the external costs experienced by all agents in the domestic economy). From a national viewpoint there is excess production equal to the difference between Quand Qu' and there is a net social cost due to the externality equal to the triangle a' ca (the total cost of the externality is E per unit times a quantity Q(l> equal to the parallelogram 10' calo • but some of that is offset by producer surplus equal to 10' a'a lo on the extra production, Qo - Qo'). A constant per unit externality, E, is assumed for convenience and means that Su and Su' are parallel. Suppose research reduces the social costs of production by k' per unit without affecting the externality. Then, when research shifts supply from So to S), it also shifts the marginal social cost from So' to S)'. The net social cost of the externality is unaffected (triangle b'c 'b is identical to triangle a'c a). However, the total social cost of the externality does increase by E(Q) - Qu), which is equal to area acc'b or area a' cc'b' . The producer benefit from Figure 4.23: Research benefits in a small country with a negative production externality Price So' }kf s)' I0' I') 10 0 I) :t 0 : QTo : 0 " 0 Cu Qu Q; Q() Q) Quantity 296 Economic Surplus Methods research is area loabl l (= 10' ee'1 /). Total benefits are given by deducting the amount of the increased external cost from producer benefits (there are no effects on consumer welfare). The result is that total research benefits are equal to area 10' a'b'I I' and are unaffected by the presence ofthe externality. Only the distribution of benefits is affected - producer benefits are greater, but offsetting this is increased external costs. This example is relatively stylized. A large number of alternative scenar­ ios might be considered. For instance, suppose the research-induced supply shift was not associated with an identical shift in marginal social costs - so all of the private cost savings were merely increases in the externality without any reduction in social costs. In this case, the producer benefits would be as described above but the research would lead to a net social cost equal to area aedb. Alternatively, suppose the effect of the research were to reduce the externality by k' without any effect on private costs. In this case, there would be no producer benefits but total research benefits would be equal to 10' eft/. Both of these are extreme examples, but it is perhaps equally unlikely that a research-induced technical change would be neutral with respect to an externality, especially if that externality is associated with the use of a particular input.58 As a further complication, an externality may not be constant per unit of output, and research-induced technical changes might cause nonparallel shifts of nonlinear functions that are difficult to judge. Finally, for both traded and nontraded goods, the international distribution of externalities might be an important issue from an international perspec­ tive, if not a national one. In particular, for example, this might apply in the context of global warming. What should we do about externalities in the evaluation of agricultural research and priority setting? The general answer is the same as the one given for agricultural policies. Externalities are like taxes or exchange rate distor­ tions. When measuring research benefits, we ought to take into account the total effects on the welfare of all affected groups when we can. To do this, we compute the effects of research-induced supply shifts on producer sur­ plus, consumer surplus, government revenues, and now, those who bear the costs of externalities. Then the total benefit is obtained as the sum of benefits and costs to all groups. A shortcut approach would be to evaluate the effects by adjusting first for the externality and working off the marginal-social-cost curves instead of the supply curves. This is not recommended for two reasons. First, it is difficult and, if the information is available to allow it to be done, it is not necessary. Second, information on the distribution of 58. Indeed, a great deal of current research in many countries is directed at reducing the demand for chemicals and enhancing water-use efficiency, all with a view to reducing adverse environmental effects. Economic Surplus Methods 297 benefits and costs will be lost. In the event that good information is unavail­ able on the size of externalities and the effects of agricultural research on them but these effects are thought to be important, there is little that can be done except to heavily qualify the results of an analysis that does not take them into account. 4.5.2 Resource Depletion, Intergenerational Equity, and Agricultural Research An increased rate of exploitation of the resource base is inherent in the "green revolution" technologies that involved a greater intensity of use of land and other resources. Problems that have arisen from intensified land use include soil salinization and acidification, erosion, and the like, and in some cases this has been reflected in declining yields. Greater reliance on chemical pesticides and cherbicides, especially in monoculture systems, leads to problems over time because of the development of resistance and the de­ struction of natural predators. Many people are concerned that the capacity of agricultural systems (globally or locally) is being depreciated too rapidly by excessive exploita­ tion of the natural resource base.59 Underlying this concern is an implicit belief that agricultural decision makers are discounting the future too heav­ ily, that they find it optimal to consume the natural resource base too quickly, compared with some standard. Two possible rationales are that (a) private discount rates are greater than social discount rates and (b) some individuals attach too little weight to the welfare of future generations. These rationales are both open to challenge, but it must be said that they are among the thornier issues in economics and there is no clear consensus. Why do we think landowners fail to value their assets properly? By whose values are future generations being underweighted? By the same token, what about the starving members of the current generation - are they being underweighted, too, according to the same argument? Graham-Tomasi (1991) and Crosson and Anderson (1993) discuss some of these questions in the context of sustainability and agricultural research. Notwithstanding the heat generated by this debate, it is not clear that there are any useful implications for the measurement of expected economic surplus arising from improved technology. Lacking clear evidence to the contrary, the best approach remains that prices are our measure of opportu­ nity costs and the unadjusted economic surplus measures are appropriate. 59. As Alston, Pardey and Carter (1994, p. 97) ObselVed, "Ajlaw of selVices from the stock of natural resources is always used in production but the concern here is with changes in the stock itself that wiU imply a reduction in future selVice flows." 298 Economic Surplus Methods Where there is evidence of a market failure such that social costs and benefits diverge from private costs and benefits, the analysis can be adjusted as shown above for the case of an externality. The political nature of these issues means that there may be implications for research priority setting and evaluation even when there are none for expected economic surplus. Projects aiming to develop more environmen­ tally sensitive technology, or likely to contribute to sustainability, seem likely to be given priority, everything else being equal. There are two roles for the economist here. The first is to clarify the issues surrounding sus­ tainability (and other similar objectives) such that decision makers can develop informed weightings for that objective compared with other objec­ tives. The second is to use the expected economic surplus analysis to demonstrate the opportunity cost of using research to pursue objectives such as sustainability. It is not clear that biasing the research portfolio is the best way to counter the effects of environmental problems. However, research has the potential to address some of the problems that have been identified as arising in part from the dynamics of modem production systems, and as those problems become increasingly important, the likelihood increases that such research will be a part of the optimal portfolio. 4.6 Conclusion A benefit-cost analysis based on the calculation of economic surplus provides a framework for agricultural research evaluation and priority set­ ting capable of incorporating most of the conceptual elements discussed in chapter 2. Although all procedures are subjective and only provide an approximation of the "true" model, the economic surplus model provides a logically consistent approximation and is a useful evaluation tool. A common misperception is that economic models, and their creators, relate only to pecuniary or monetary issues, so questions of income distribu­ tion, nutrition, or sustainability, for example, must be dealt with separately from the economic analysis. In this chapter we have shown that a com­ prehensive approach can explore these different types of economic questions at the same time. The analysis aims to quantify the physical effects of different technology scenarios, to translate them where possible into economic values, and to derive summary measures of those values that can be used to rank alternatives on a consistent basis. The great strength of this approach - over alternatives that consider multiple aspects of the problem in a piecemeal fashion - is that it is comprehensive and internally consistent. Economic Surplus Methods 299 We have illustrated the application of supply, demand, and economic surplus models to a wide range of situations. Often, a simple aggregate model of either a closed economy or a small open economy will suffice. For most cases this will provide the information necessary to evaluate alter­ natives against a criterion of economic efficiency. Sometimes, however, consequences other than economic efficiency will be of interest. In most instances, agricultural research is not the best way to achieve nonefficiency objectives. Usually there is a better policy available for achieving a specific nonefficiency objective (such as income distribution, nutrition), orto address sustainability or natural resource conservation concerns. Nevertheless, the economic surplus analysis can be extended to consider a range of the distributional effects of research along with the efficiency effects. A variety of horizontal disaggregations (according to agroecological zones, types of producers, income classes of consumers, countries or regions of a country, and so on) are readily incorporated in the analysis if data are avaliable. Similarly, vertical disaggregations across stages of a multistage system of production permit us to examine the distributional consequences among producers, consumers, and market intermediaries. Some problems might warrant relatively simple models. Some may call for sophistication to allow for research-induced quality changes, general­ equilibrium feedback effects, market-distorting policies, and other distor­ tions in incentives such as externalities. We have shown how to incorporate some of these aspects into an economic surplus model, too. While we have not considered every possibility, we have dealt with the most important ones and the reader should be able to draw on these examples to deal with most of the situations likely to be encountered. In particular, we have not consid­ ered the implications of distortions arising from the exercise of market power by individual firms. There is little written on the effects of the market power of firms on the size and distribution of research benefits in agriculture. This may be because of a widespread belief that the competitive model provides a good approximation for agriculture.6o The next step is to consider how to obtain and use the economic surplus measures in practice. The following chapter provides details on implement­ ing the procedures, including algebraic formulas for surplus measures for the main examples, and in an appendix examples of computer programs are provided that can be used in either a spreadsheet or in the interactive computer program, Dream©, that we have written for research evaluation. 60. One exception is Kim et al. 's (1987) reconsideration ofthe benefits from the tomato harvester in California, although their assumption of pure monopsony (i.e., a single processing firm) may be more misleading than an assumption of competition. Part III Evaluation and Priority Setting in Practice 5 Economic Surplus Measurement and Application This chapter describes how to implement an economic surplus analysis to evaluate agricultural research and how to use the results for setting priorities and allocating research resources. We break the implementation process down into five components and relate those components to the underlying conceptual constructs, such as the knowledge-production function and eco­ nomic surplus models presented in earlier chapters. Defining the problem: A research evaluation study is a research project itself and, like all research projects, should have clear, explicit objectives. At the outset, it is important to decide which questions the analysis hopes to answer, as well as how the results will be used and for what types of decisions. This defines the objectives for the analysis. I Only then is it possible to define the scope of the analysis in terms of the research programs for which benefits and costs are to be assessed and the kinds of decisions to be made about those research programs.2 A statement of the objectives oft he system, against which performance is to be measured, helps define the I. The client, who commissions a research evaluation study, might also be asked to identify the objectives lif the research system to which the study relates, as a basis for defining the relevant measures of benefits and costs. These two notions of objectives - the objectives of the study of the research system and the objectives of the system itself - can be related to one another (particularly in tenus of who defines them) but they are very different ideas. 2. Research programs are often defined by the institutional structure in which research is carned out. Sometimes they are defined according to the commodities they affect (in an institute structured along commodity lines), the disciplinary focus (where research is organized into fields such as entomology or soil science), and spatial aggregates (where research is organized among regional institutes or where spatial incidence is an issue). 303 304 Economic Surplus Measurement and Application relevant measures of costs and benefits (which will often be limited to measures of total economic surplus and its distribution, the subject of this chapter). Together, these two sets of objectives, along with the definitions of measures of costs and benefits, are the terms of reference for the study. Then, given a budget for the study and in consideration of other constraints such as data availability, a modeling strategy can be developed to define the degree of detail for the analysis in terms of the degree of disaggregation of both research program alternatives and measures of performance. Compiling the data: The next step is to define and compile the research­ and market-related data - some of which may be obtained from published sources and some elicited from scientists and others. The modeling strategy and scope of the analysis define the data requirements. Essentially, data on prices and quantities produced or consumed are required for each group of producers and consumers identified by the client as being of particular interest. Corresponding elasticities of supply or demand will be required for each identified group. In addition to these market-related data, research-re­ lated data are required to identify the nature, magnitude, and timing of the research-induced shifts of supply. Measu'ringK: As discussed in chapters 2 and 4, by far the most important parameter defining total benefits, and perhaps the hardest to measure, is the size of the research-induced supply shift, K. An ex ante assessment of program alternatives usually involves eliciting the values of the potential impact of successful research (often expressed as yield effects). These are combined with estimated probabilities of success and information on likely adoption paths to derive a time path for K for each program alternative. Conceptual and practical issues in estimating K are raised and resolved below. Analyzing the data: The fourth step is to calculate research benefits and validate the results. Spreadsheets or other computer aids are helpful for this task, and the Dream© interactive computer package, discussed and docu­ mented in appendix AS .1.2, could be used. Depending on the purpose of the analysis and the nature of the data, in addition to estimates of total benefits, estimates of the effects on the distribution of benefits among different groups or measures of the effects of research on income variability may also be needed. For ex post studies, the set of choices is limited, but it may still be relevant to compute both total benefits from the total investment and mar­ ginal benefits (using small changes from the actual past investments). For ex ante analyses, it may be necessary to compute benefits in relation to (a) the current allocation of resources to research, (b) allocations representing changes from a baseline of the current allocation, (c) allocations of a given amount to each of a range of programs, or (d) more than one of these options. Economic Surplus Measurement and Application 305 In every case, benefits and costs have to be converted to comparable units (such as net present value per unit of research resources or internal rates of return) by capital-budgeting methods, as described in section 5.4.2. Interpreting and using the results: The fifth component involves mak­ ing the results useful for the client, either as a set of summary statistics, a set of decisions about priorities, or an institutionalized process of thinking about decisions regarding the allocation of resources to research. Once gross and net benefits have been calculated for each of the alternatives being analyzed, they can be converted into summary statistics such as rates of return or net present values (perhaps per dollar of research spending) and used as aids in decision making. Usually, it is important to validate the results in consulta­ tion with the client (the decision maker for whom they are being developed), which might involve some adjustment of results while a consensus is being built.3 The process will usually not be linear, as laid out here, but will involve iteration in which goals for the analysis are reviewed and revised after preliminary results have been obtained and digested. 5.1 Defining the Problem The variety of problems and situations that call for research evaluation is too great to be covered fully here. It is not our purpose to provide a manual that deals with all possible situations. Rather, we provide a set of principles and general approaches that can be tailored to specific situations. In order to illustrate those principles, we use as our central example the common case of a public agricultural research system in a developing country deciding how to allocate a fixed amount of research resources among a given set of research program areas. 5.1.1 Clients for the Analysis - Decisions to Be Served The process by which resources are allocated to agricultural research is complex. It is shaped by the conjunction of a large number of scientific, economic, and political factors, and many decision makers have an interest in or are responsible for allocating resources to agricultural research (chapter 1). While many are interested in (and, indeed, have an economic stake in) agricul­ tural research decisions, few have any direct say in decision making. For any particular research budget, there is a person or group of people who have the 3. Both as part of the validation, and as a management strategy to effectively implement the results of the exercise, other scientists and lower-level managers typically are (and should be) involved at this stage. 306 Economic Surplus Measurement and Application authority and responsibility for allocating funds among the broad alternatives being considered in strategic analyses such as this. This person or group is the primary client for a research evaluation or priority-setting study. Several individuals may be potential clients. For instance, the minister of agriculture may be responsible for allocating a total budget for national agricultural research (and might delegate that authority to the permanent secretary). The secretary of agriculture may allocate a total research budget among research institutes (although it would be more typically delegated to a systemwide director of research, council, or board). In turn, a director of a research institute may be authorized to allocate the research institute's budget among programs and projects. Usually, however, the delegation is less than complete. The minister and secretary often take an interest in the distribution of the budget among institutes and the programs within them, and the national director or board often plays an active role in determining the research programs of individual institutes. A first step in any research evaluation project is to identify the client (or clients). Typically, the client is the senior manager (or a small group of senior managers) in the research system being evaluated. Often research evaluation is undertaken on an explicit consultancy basis, where the contractual ar­ rangement makes clear the objectives of the work and explicitly identifies the client. But in-house reviews are common, too. In those studies, as well, it is necessary to be clear about which decision maker within the organization will play the role of the client for the work, what decisions the work will serve, and what questions it will try to answer. 5.1.2 The Objectives of the Analysis - Terms of Reference Within a public agricultural research system, a variety of questions might be raised about the value of research, perhaps calling for a variety of evaluation approaches. There is often a demand for information on rates of return to justify support for research. Estimated rates of return are usually based on ex post analysis of past research (see chapter 3 for examples). Sometimes this ex post evaluation is treated as if it provided a basis for setting priorities for future research - an extrapolation that often is not justified. The other main type of evaluation work pertains to decisions about allocating resources for the future - either for ongoing research programs or for new initiatives, that is, for setting priorities. In a priority-setting context, the types of decisions being made can include • how much to allocate to research (and extension) in total • how to allocate that total among different programs of research • whether to accelerate, slow down, or even discontinue existing programs Economic Surplus Measurement and Application 307 • whether to introduce new programs In principle, we might envision a process of equating marginal social benefits and costs among all possible alternatives so as to maximize net social benefits. In a practical setting, it is necessary to define a specific finite set of alternative decisions about research programs so that those decisions can be evaluated and compared. For example, an estimate of the net benefits for the current program alone is of little use in decision making; it is much more useful to compare the net benefits of the current program to a relevant alternative (e.g., an across-the-board 10% cut in current programs).4 Here we focus on the example of research evaluation that relates to the question of how to allocate a fixed quantum of research resources among a given set of commodity research programs. This question calls for a forward­ looking analysis in which prospective programs of research (i.e., research not yet done) are evaluated ex ante. A great many troublesome issues arise even in the context of this somewhat circumscribed example. Along the way, we elaborate on some of these issues. And occasionally we refer to differences that arise in other cases that we do not cover in detail- such as ex post analysis of research that has already been done in either individual programs or total systems, or ex ante analysis when the research budget is not fixed. 5.1.3 The Scope of the Analysis - Research Programs and Program Alternatives Before collecting data, it is necessary to define the research programs of interest and the alternatives to be evaluated. The list of programs might be defined in terms of commodity and noncommodity programs and also might be disaggregated regionally. For an ex post research evaluation, the list may be relatively short. However, for a priority-setting analysis, the list can be long, and programs on commodities that are clearly of low priority, perhaps because of their very small current and potential value of production, are candidates for exclusion from formal analysis. If research is to be prioritized by noncommodity research programs or by components of commodity research programs (e.g., plant breeding, plant protection, or animal nutrition), a list of perhaps three to four broad programs 4. Even if the appraisal of current programs shows that the rates of return (or net present value per unit ofc onstrained resource) are different among programs, permitting a ranking of programs, more information is needed to support a decision to change the allocation of resources: Should programs with lower rates of return be cut and programs with higher rates of return expanded? Not always. If a program with a relatively low rate of return is to be reduced, how much should it be reduced? Answers to questions such as these depend on the curvature of the research-production relationship, the extent of sunk costs in a particular research program, and the degree of fixity in research resources, among other things. If a particular set of alternatives has been evaluated, decisions can be supported by the analysis. 308 Economic Surplus Measurement and Application (representing a logical aggregation of more specific research areas) that correspond to existing or potential programs can be developed. An example of typical research areas is included in table 5.1. However, because of the difficulty of quantifying the effects of research in certain noncommodity areas, a decision may be made to formally analyze only a subset of the areas explicitly using economic surplus models. For instance, economics and other social sciences, agroforestry, and agroclimatology might be excluded be­ cause of the conceptual and empirical difficulty of applying an economic surplus analysis to them. Alternative methods (perhaps informal economic surplus approaches) must be used to evaluate these programs. Table 5.1: An Illustrative List of Commodity and Noncommodity Research Programs Crop programs Cross-commodity programs Plant breeding Agroc1imatology Plant cultural practices Agroforestry Plant protection Economics and other social services Livestock programs Mechanization Animal health Postharvest technologies Animal nutrition Soils and fertilizers Animal reproduction Water management Having defined the relevant research programs, what remains to be de­ fined are the decisions about the programs to be evaluated - decisions such as whether to support particular programs at all or, alternatively, whether to redistribute funds in a particular way among existing programs. Even when the total budget for research is fixed, it is informative to consider varying the allocation of resources to particular programs in order to measure the bene­ fits and costs of potentially relevant alternatives. For instance, one approach is to evaluate each research program at the current level of funding and plus or minus 10% of that amount. The results from such evaluations could be used in a decision-making model to reallocate the fixed total resources for research among the existing programs in order to try to maximize overall net benefits. When shutting down a program or introducing an altogether new program is being considered, it is appropriate to compute the benefits and costs of those "all-or-nothing" decisions rather than the benefits and costs of marginal changes. Economic Surplus Measurement and Application 309 5. J.4 Objectives for the Research System - Measures of Benefits The scope of the analysis defines the alternatives to be assessed. The next step is to define the yardstick to be used in making that assessment: the measures of benefits corresponding to the objectives for the system being analyzed. The "clients" for the analysis - typically national and regional research directors or an agricultural research councilor board - must define the objectives for agricultural research. Usually it is productive for the analyst(s) conducting the study to meet with the clients to discuss the list of potential objectives and to ensure that they are in agreement. A generic list of objectives is presented in table 5.2. The list for a particular study is usually more specific. For example, if a regional distributional objective is specified, it must indicate which regions are to receive additional emphasis.s It is important during this step not to confuse objectives with the means and measures of achieving them. For example, improving nutrition, increas­ ing production and employment, and generating foreign exchange are means or measures of achieving increased economic and physical well-being. Each objective for the research system implies the economic impact of research to be measured, if possible. For example, an objective of increasing total net benefits implies a need to estimate research-induced changes in total eco­ nomic surplus. An objective related to the regional distribution of benefits implies a need to calculate changes in economic surplus by region. Also, it may be necessary to put some effort into resolving what the real objectives implied by the stated objectives are. For instance, a stated objec­ tive of "sustainability" is not clearly meaningful. Through discussion with the client, the analyst may be able to identify a more explicit concern about conservation of the natural resource base. In tum, that concern can be identified more explicitly as either a concern that the opportunity cost of natural resources be measured properly (an economic efficiency consider­ ation) or that resources be conserved so that they are available for the following generations (an intergenerational distributional objective). Failing to identify the real objectives in this fashion could lead to unnecessary complications in the analysis due to the inclusion of redundant objectives or, perhaps even worse, to double-counting of effects. As discussed in chapters 2 and 4, consideration of objectives other than economic efficiency makes the analysis much more expensive and problem- s. The decision to undertake economic sUiplus analysis at a spatial level other than national might also be due to a concern that the effects of research on reducing per unit costs or increasing yields cannot be modeled accurately at the aggregate level, or because decisions need to be taken within regions. Regions can be based on agroecological or political considerations. There may be some correlation between the two but seldom do politically based regions and agroecological zones coincide exactly. Suggestions on how to define the boundaries of agroecological zones and relate them to regions are presented later in this chapter. 310 Economic Surplus Measurement and Application Table 5.2: An lllustrative List of Agricultural Research Objectives Broad societal goals Objectives for the research system 1. Efficiency - raise the average level 1. Increase the total or average well-being of well-being in the economy of producers and consumers taken in aggregate 2. Equity - increase the well-being of 1. Improve the well-being of specified particular groups income group(s) (e.g." low-income groups) 2. Improve the well-being of people living in particular locations 3. Improve the well-being of owners of particular factors of production (e.g., providing employment for landless laborers) 4. Improve the well-being of people with specific farm sizes (e.g." small farms) 5. Improve the well-being of people in specified farm-tenure situations 6. Improve the well-being of consumers and producers differentially 3. Security - reduce the variability of 1. Reduce year-to-year income well-being over time or increase fluctuations food safety 2. Increase food self-sufficiency or self-reliance 3. Improve food safety atic. Limiting attention to economywide economic efficiency means that only the total domestic impact of economic surplus must be assessed. This is still a formidable task, given that the evaluation involves assessing the total costs and benefits for each research program (and each option being consid­ ered for each program) and given the likelihood that a regional disaggrega­ tion (either international or domestic) will be required to permit an accurate assessment of domestic benefits. Estimating the effects on different groups in the domestic economy might not involve too much extra work if the model has been parameterized in reasonable detail. Making use of the information about distributional impacts may be more problematic; as noted in the earlier chapters, research is a blunt and ineffective instrument for achieving dis­ tributional, or security, objectives. One valuable contribution of the economic surplus analysis is that it allows an assessment of the opportunity cost of using agricultural research Economic Surplus Measurement and Application 311 to pursue nonefficiency objectives, reflected as lower total benefits from the overall research program. Those opportunity costs can be estimated by considering research programs that are designed to pursue nonefficiency objectives such as income distribution or security. This requires measuring the contributions of different research program choices to the nonefficiency objectives of interest. With that in mind, although this chapter is primarily about measuring economic surplus and its distribution, some attention is given to measuring the impact on variability as an example of a noneffici­ encyobjective. 5.1.5 Strategy for the Analysis - Degree of Detail One primary decision is whether the analysis will lead to relatively precise, detailed, disaggregated estimates of benefits or, instead, will be relatively aggregative and approximate. Some studies require highly disag­ gregated and precise estimates of impact on a large number of groups of people (regional or otherwise); others are served adequately by rough esti­ mates of aggregated (say, national) welfare impacts. The Simplest Case - Aggregate Impacts, No Price Effects Suppose all weight is placed on the efficiency objecti ve so that only global total benefits are of interest (and no consideration is given to the distribution of benefits among various groups). In such a case, the gross annual research benefit from a K percent per unit cost saving (or increase in yield) can be closely approximated without regard to research-induced changes in prices and quantities (and therefore without requiring any information on elastici­ ties and market shares) as being equal to KVo, where Vc) is the initial value of production. Thus, the analysis simply involves estimating the time-specific values of the increased production or of the inputs saved due to research. This model, which was presented in figure 2.6 in chapter 2, is implicit in both simple benefit-cost analysis and the econometric approach to research eval­ uation discussed in chapter 3. Although changes in prices and quantities are not explicitly accounted for, this simple model can provide a first approxi­ mation of total research benefits because such market effects tend to influ­ ence the distribution of benefits more than the total benefits. This simple model requires roughly the same effort at identifying re­ search-related information (effects on yield or per unit costs, probabilities of research success, adoption rates, and base price and quantity data) as more sophisticated forms of economic surplus analysis (for a given set of alterna­ tives). But it is simplified in its requirements for market-related information: 312 Economic Surplus Measurement and Application it does not require information on price elasticities of demand or supply (assumed zero or infinite) or market shares. However, it implicitly imposes some extreme assumptions about price policies, technological spillovers (they are absent), and market structure (it is irrelevant), which can be relaxed in more detailed modeling approaches. It would be a minor variation on the theme in such an analysis to adjust for some of these aspects while preserving the essential parsimony of the approach. This type of simplified economic surplus model is used most often when a long list of alternatives must be compared but total resources for the analysis are limited, or where only rough approximations are judged to be adequate for the purpose at hand. This approach has been combined with other criteria, attempting to weight multiple objectives for research, in simple scoring models that are reviewed and critiqued in chapter 7.6 It has also been used for project-level evaluations of improved technologies (e.g., lAC 1976). Extensions - Some Price Effects. Some Disaggregation The approximations and the degree of aggregation involved in the simple model mean that it is unable to deal with the distribution of benefits. It may also result in significant measurement errors. Where commodities are traded and technologies are transferable outside the region of direct interest, in such a way that benefits accrue beyond the place where the research is carried out, the simple model might not be adequate for measuring aggregate domestic effects. In addition, a measure of distribution of domestic benefits and costs may be of direct interest (e.g., it may make a difference if benefits accrue to farmers rather than consumers or to the farmers in a particular region). A more detailed economic surplus analysis allows for public policies, interregional and interna­ tional trade, and regional price and technology spillovers, providing a measure of the distributional impact of research and thereby a more precise measure­ ment of the domestic impact of interest. This type of analysis distinguishes the impact of domestic welfare from international impact; it can also distinguish between impact on domestic producers, consumers, and taxpayers. The analy­ sis can be dis aggregated horizontally to account for the spatial distribution of research benefits and costs within a country and to incorporate research spillover effects both within and between countries. A country's research can affect world trading prices in one of two ways. First, if the country is large in trade in the commodity, its production response to research affects world trading prices (i.e., the price spillover effects of research). Second, if the country is large in research, the adoption of its research results by other countries leads to changes in world production that 6. See, for example, the study by Cessay et aI. (1989) of research priorities in the Gambia. Economic Surplus Measurement and Application 313 affect world prices (i.e., the price effects of technology spillovers). Thus, price spillover effects can arise without international technology spillovers only when a country is large in trade, but the price effects of technology spillovers can originate from a country small in trade if the country is large in research. Technology spillovers might be ignored if they had no price effects.7 However, when the domestic price effects of price and technology spillovers are significant, it is necessary to use a multimarket model to measure their domestic consequences. Of course, if the client is interested in international impact (e.g., the CGIAR or international donors to domestic systems), there is another set of reasons for paying attention to the global multicountry impact of domestic R&D.8 Even when the spatial distribution of benefits within the country is not of direct concern, there may be grounds for disaggregating some aspects of the analysis to improve the accuracy of the measures of total research benefits. In particular, agroecological diversity within a country often means that dis aggregated data on the local impact of research (in terms of cost effects and rates of adoption) are needed to obtain accurate measures of the aggre­ gate research-induced supply shift, K. Similar points apply to the estimation of other parameters (e.g., demand elasticities), but the potential for aggrega­ tion bias is typically more important in estimating K. The more detailed analysis requires the same research-related information as the simple model, but it is more detailed in its treatment of markets and, therefore, more demanding in its requirements for market-related data. In addition to the information required by the simple model, it requires esti­ mates of commodity-specific price elasticities of supply and demand, quan­ tities traded internationally, and agricultural policies. Also, depending on the concerns with aggregation and the potential for the price effects of technol­ ogy spillovers, it might also require a spatially disaggregated treatment of the supply response to research, both domestically and internationally. Some version of this more detailed approach is likely to be used in any serious ex ante priority-setting exercise conducted by a NARS (except per­ haps in the smallest systems) because it incorporates the virtual minimum of market-related and research-related data necessary to obtain meaningful 7. This typically would be the case, for instance, when the ROW countries that adopt the research results are collectively small in trade in the commodity, in that changes in their production have a negligible effect on global exports or imports. It is also true when the combination of relative smallness of spillover effects on production combined with relative smallness of countries in trade means that the spillover effects on prices are small, or when government interventions in markets prevent the effects fium being reflected in international prices. 8. For instance see Edwards and Freebairn (1981, 1982, 1984), Davis, Dram and Ryan (1987), McCalla and Ryan (1992), and Traxler and Byerlee (1994). 314 Economic Surplus Measurement and Application measures of national research benefits. Such an approach has been used extensively for research evaluation and for priority-setting analysis.9 While more time, information, and effort are required than for the simple model, the difference should not be too great if basic data are available on elastici­ ties, trade, and pricing policies. One purpose of this book is to facilitate the application of models with more detail (and hence more accuracy) in re­ search evaluation and priority setting. Further Extensions In chapter 4, we discussed extending the simple model to disaggregate impacts vertically among the stages of a multistage production system, which requires information on the elasticities of substitution among inputs, factor shares, and factor-supply elasticities for inputs used in the marketing chain. Such models have not been applied often in research priority-setting contexts/o but they are likely to become more widely used as data and techniques become better and cheaper. Also, some recent studies have developed and applied more comprehensive general-equilibrium models to evaluate specific questions in relation to research benefits (e.g., Coble et al. 1992; Martin and Alston 1993). Such studies have tended to relate more to broader questions of research policy than to applied evaluation and priority­ setting work. However, these relatively new approaches require further development and refinement before they can be applied to practical research evaluation and priority-setting problems. Hence, we see limited use of such models in the immediate future for most of the questions that concern us in this book, and we do not go into them further here. 5.2 Market-Related Data A major step in implementing an economic surplus analysis is defining the specific data and other information required, deciding who should pro­ vide particular types of information, designing questionnaires (if necessary) to obtain that information, and collecting the information. The specific types of information needed for our hypothetical study on behalf of a developing­ country NARS, and their likely sources, are described below. The economic surplus measures presented in chapter 4 require, at a mini­ mum, data on quantities produced and consumed, prices received and paid, and 9. See Norton, Ganoza and Pomareda (1987), Dey and Norton (1993), Palomino and Norton (1992a), and lima and Norton (1993). 10. Exceptions include Mullen and Alston (1990) and Scobie and Jacobsen (1992). Economic Surplus Measurement and Application 315 the corresponding price elasticities of supply and demand for each identified group of producers and consumers. An ex ante analysis requires a few years of price and quantity data (perhaps the most recent three or four years) for a benchmark. II Data on prices and quantities of exports and imports, exchange rates, rates of population and income growth, and information on government price policies and any other relevant market interventions may be required for some commodities. A discount rate is required for the capital budgeting framework. These items are discussed individually below. Decisions must be made about the "level" in the production-marketing­ consumption system at which to apply the analysis, and these decisions are usually governed by practical considerations of the availability of data and the degree to which a commodity is recognizable as such (e.g., wheat versus bread), as well as a desire to relate the analysis closely to the objectives identified by the client. This constellation of considerations usually implies that an analysis should apply at the farm level or at the wholesale level. The latter is especially apt to be used for traded goods to ensure that domestic and traded quantities and prices are treated consistently .12 It is unusual for a study of research benefits to use retail prices and retail goods. When farm-level prices are used, it may be necessary to consider spatial variation in prices for a gi ven quality and price variation due to variations in quality when forming a representative price index. For traded goods, it may be appropriate to use the border price - either c.i.f. or f.o.b., depending on whether the good is importable or exportable. As noted above, once the set of goods has been defined explicitly (includ­ ing the market level), at a minimum, quantities are required on production and consumption for every distinct group of producers and consumers identified in the objectives. Producer and consumer prices will be required for every such identified quantity. Of course, if there were no policy inter­ ventions or other reasons for price differences, all consumers and producers would effectively face the same price. II. An ex post analysis typically requires detailed data on prices and quantities for a single commodity aggregate of interest on an annual basis for all past years for which benefits are to be assessed. As discussed in chapter 3, econometric studies require even more data. 12. As discussed in detail in chapter 4, the choice of marl<:et level for tbe analysis involves an implict choice about the aggregation of surpluses accruing to farmers and others on the supply side (in "producer surplus") and consumers and others on the demand side (in "consumer surplus"). 316 Economic Surplus Measurement and Application 5.2.1 Price and Quantity Data Quantities Produced Infonnation is needed on the annual production of each commodity included on the list of research programs. Usually a three- or four-year average is taken to reduce the effects of abnonnal years. For ex ante evaluation, the most recent three to four years are typically used as a benchmark.13 If there are regional distributional objectives and a more completely disaggregated analysis is being undertaken, explicitly taking trade into account, the same infonnation must also be gathered on a regional basis. Data are also needed on the recent world production of tradable commodities for which the country influences world market priceS.14 Data on own-country production are usually available from national statistical sources. World production for particular commodities can be obtained from a source such as the most recent FAO Production Yearbook. 15 But users need to be aware that such data are not always completely reliable­ it is best to double-check. Quantities Consumed and Traded Unless the simplest model is being used, information is needed on exports and imports (for the same years as production) for those commodities for which the country can influence world prices, either through its own produc­ tion or through spillovers of its research results. The most recent FAO Trade Yearbook and the other references listed in footnote 15 are potential sources for these data. Data on interregional trade within the innovating country are needed as well when regional distributional objectives are specified. These data may be obtained from state, provincial, or national statistical agencies. Consumption data can be obtained from national statistical sources by adding imports to and subtracting exports from production (and by adjusting 13. For ex post analysis, there is a general problem of infening a stream of without-research prices and quantities using time-denominated with-research data that vary partly as a result of variables not included in the model (e.g., weather, policy changes), unlike the ex ante analysis where, by assumption, other things are held equal or are explicitly modeled. Problems of double counting or inappropriate attribution can arise when studies apply a measure of K to actual past quantities and prices (i.e., time-subscripted data) rather than projecting the entire series forward or backward, based on a benchmark as for ex ante analysis. It is important to be conscious of the potential problems. One "safe" course is to carry out an ex post analysis in the same way as an ex ante analysis, projecting the entire series from a benchmark as described in appendix AS.l.2, in which case the same information is required for benchmarlcing. 14. Most countries are unable to influence the world price of any commodity, over the relevant length of run, through international trade; at most there may be one of these commodities in a typical country study. 15. A number of government organizations and international agencies have made useful data available on diskette for a nominal charge. These include (a) FAO's AGROSTAT data, (b) USDA's World Agriculture and Trade Indicators files, and (c) The World Bank's World Tables data series. Economic Surplus Measurement and Application 317 for changes in stocks if data are available). Consumption data must be con­ verted to the same units as production and trade data. 16 For example, informa­ tion may be needed to convert slaughtered weight oflivestock to its equivalent live weight or to convert various processed forms of food crops (e.g., dried, shelled, milled) into a standardized equivalent form to maintain units consistent with production data. World consumption data are also needed for those commodities. World consumption of a commodity can be assumed to be roughly equal to world production. If there is an interest in the dis aggregated effects on consumers according to regions within the country, data on consumption by region may be required. These data are unavailable for many countries, but they do exist in several of the larger, more populous countries. In any event, even when not available directly, values for consumption can often be inferred using information on the regional population pattern and either (a) per capita incomes and results from studies on income-consumption relationships or (b) data on rural/urban per capita consumption patterns that can be used to apportion aggregate consump­ tion among regions. 17 The latter is likely to be more accurate. Prices Data to match the production and consumption data (i.e., the same commod­ ities at the same market levels in the same places and for the same years) are need for the prices on each commodity. It is usually necessary to determine whether these domestic prices are free-market prices or the result of a tax or subsidy policy and to measure the extent of any price interventions. When regional objectives on income distribution are included, and therefore regional income distribution effects are to be measured in the analysis, regional prices and information on regional pricing policies are needed as well. 18 If the analysis 16. Relevant conversion factors can usually be obtained from local statistical agencies. 17. The data on per capita consumption may be obtained from periodic consumption surveys (often carried out by local statistical agencies). 18. Although we have abstracted from transport costs, regional differences in prices attributable to transport costs might be significant - especially for perishables such as fresh fruit and milk - and it might not be appropriate to treat prices as being equal among regions for the welfare analysis, even when the supply-and-{!eroand model does not include explicit spatial equilibrium considerations. Options as approximations in models where prices differ regionally include (a) treating individual regions as closed economies in trade in certain collUIlOdities (not necessarily in technology), (b) treating regional prices as being in fIXed proportion to one another (equivalent to a constant percentage price difference) and solving for each regional price given a percentage change in the national price, and (c) treating the price differences as being constant and solving for each regional price given a value for the national price. Since it is the change in prices that matters for welfare analysis, it might not be worth the effort of trying to allow for regional price differences in such ways in the welfare analysis if the IIIIlIket equilibrium model does not explicitly account for the causes of those differences. In roany cases, it will be appropriate to use changes 318 Economic Surplus Measurement and Application is to be dis aggregated vertically in the marketing chain to deal with multiple market levels, as discussed in chapter 4, data will be needed on marketing margins. Data on marketing margins may be required for converting prices received by farmers to retail or wholesale, or vice versa, even when benefits are not to be disaggregated vertically. Typically, an analysis will be undertaken with all monetary variables ex­ pressed in real terms (usually current purchasing power). This can be achieved by deflating nominal prices by an appropriate price index, with the initial year for the economic surplus calculation taken as the base.19 When international data are included in the analysis, it is usually desirable to convert all money units to the same currency so that benefits can be compared and summed across countries. This can be accomplished by multiplying the local currency price by the exchange rate between that currency and, say, the U.S. dollar. In some instances it might be appropriate to adjust the published official exchange rate, based on an estimate of the percentage over- or undervaluation of the currency (e.g., Krueger, Schiff and Valdez 1988). Alternatively, purchasing-power-pari­ ty indexes, such as those developed for the agricultural sector by Rao (1993), could be used instead of market exchange rates. Government Policies The specific government price policies to be included in the analysis of individual commodities can be determined by examining historical policies and by consulting with policy makers about likely future output- and input­ pricing and trade policies. These policies can be incorporated into the economic surplus measures in ways described in chapter 4. It may be tempting to treat policies as ad valorem tax or subsidy equivalents and to use available data on producer subsidy equivalents. However, it is important to be aware that, as discussed in chapter 4, such approximations can invalidate the welfare analysis (the actual instruments of protection matter), especially in relation to the distributional impacts of research. Indeed, assuming that actual policies can be modeled as tax or subsidy equivalents may distort the findings more than assuming that price policies are absent (as pointed out by de Gorter and Norton 1990). More generally, if policies are to be included, it is important to take the problem of representing the actual policy and all of the relevant components seriously. Many analysts may be inclined to ignore domestic price policies. Although it is difficult to generalize, in most cases this will lead to an overstatement of in the national border price (or equivalent for nontraded goods) in conjunction with actual regional quantities that relate to region-specific prices. 19. Chapter 3 provides information on index-number theory and construction. Economic Surplus Measurement and Application 319 the social benefits from research on protected commodities and an understate­ ment of the social benefits from research on commodities that are taxed. The percentage over- or understatement will often be approximately equal to the percentage producer subsidy equivalent (which includes the effects of com­ modity policies, input subsidies, and other measures, as described below). Thus, a consideration of the differentials in rates of protection among commod­ ities will give some guidance as to the potential importance of explicitly accounting for policies in the analysis. Where distribution of research benefits is of concern, it is especially important to take explicit account of price policies and to consider the actual policies, because the main effect of price distortions is on the distribution of benefits between producers, consumers, and govern­ ment (Alston, Edwards and Freebairn 1988). Information on the policies, and measures of them that can be included in supply-and-demand models, can be obtained from the government of the country being studied or from published studies. Methods for representing agricultural policy in models have received a great deal of attention recently, especially in connection with the Uruguay round of GATT negotiations. Pro­ ducer subsidy equivalents (PSEs) and consumer subsidy equivalents (CSEs) have been proposed as summary measures of policy distortions and the FAD (1973), DECO (1987), and USDA (1988) have published PSE and CSE estimates for the major individual commodities produced by most countries. These measures are summary subsidy equivalents that might be equivalent to the actual policies for some purposes (e.g., measuring trade-distorting effects) but are surely not equivalent for deducing the size and distribution of research benefits. However, in estimating PSEs and CSEs, FAD, the USDA, and DECO have also quantitatively documented the specific instruments of protection as well.20 Thus, the PSE-CSE data could be useful for parameterizing the types of models laid out in chapter 4. Other studies have documented other aspects of policy distortions for modeling work (e.g., Tyers and Anderson 1992). 5.2.2 Elasticities When the simplest economic surplus model is being used, the domestic price elasticities of supply and demand are implicitly assumed to be zero or infinite. To go beyond the simplest models, domestic price elasticities of supply, as well as domestic price and income elasticities of demand, are required for all of the relevant commodities. For those commodities for which the country can influence world price, a "rest-of-world" excess-sup- 20. Josling and Tangerman (1988) compared the policy coverage of the PSE estimates available from these three sources. 320 Economic Surplus Measurement and Application ply curve (for an importer) or excess-demand curve (for an exporter) may be used to summarize the "rest-of-world" role in the market in conjunction with a model of domestic supply and demand. Then, a corresponding excess-sup­ ply or excess-demand elasticity is required along with the domestic supply­ and-demand elasticities. This excess-supply-excess-demand approach is appropriate for estimating domestic benefits, but more detailed elasticities are needed for all countries for which benefits are to be calculated when a client (e.g., a donor to a national system or the CGIAR) wants to know the detailed, international, cross-country distribution of benefits. Additionally, if interregional or other multi market research effects are calculated, domestic price and income elasticities may be needed by region, by consumer income group, and so on. Obtaining all these elasticities is a tall order. Fortunately, there are several ways to approximate both domestic and foreign elasticities. Demand Elasticities Domestic price elasticities of demand can be obtained from (a) published results of previous studies, (b) estimations of demand system equations, and (c) approximations using economic theory.21 Option b is usually too expensive for an agricultural research priority-setting study. Option a is the least-cost approach, but usually a combination of a and c is used. Economic theory can be used to deduce an estimate of an unknown price elasticity of demand. Theoretical restrictions (homogeneity, symmetry, and adding-up rules) can be used to get some economically meaningful consistency into the set of demand elasticities and fill in any missing values.22 For instance, the "homogeneity condition" means that, for commodity j, the own-price elasticity of demand, llii' the income elasticity of demand, llj/' and the relevant cross-price elasticities, llij' sum to zero. That is, n L llij + llj I = 0 (5.1) i=1 For normal goods, the income elasticity of demand is positive, and for highly aggregated commodities with limited substitution possibilities (e.g., 21. For example, see Nerlove (1956), Frisch (1959), Carter and Gardiner (1988), and Tsakok (1990, appendix D) and the references therein. 22. The restrictions on demand elasticities implied by theory are docwnented and discussed by, for example, Phlips (1974), Deaton and Muellbauer (I 980aand b), and Johnson, Hassan and Green (1984). In addition to homogeneity, the most useful restrictions are (a) that the Slutsky matrix of second derivatives of the conswner-expenditure function is negative semidefinite and symmetric (Slutsky symmetry), which implies that compensated elasticities of demand, llij, satisfy the restriction that llijlS j = llfthi where Sj is the budget share of goodj and (b) the Engel-aggregation condition in which the share-weighted average of income elasticities is one: r.j Sj llj/ = I. Economic Surplus Measurement and Application 321 meat or cereal grains), the sum of the cross-price elasticities is usually a small positive number, so the own-price elasticity of demand is usually a negative number that is slightly larger in absolute value than the income elasticity. Even when the income elasticity of demand is unknown, it can be approxi­ mated if one knows whether the good is a staple, a normal good, or a lUxury good. Staples in the diet have very small or even negative income elasticities of demand, normal goods (the majority of foods) have lljl between zero and one, and luxury foods (primarily meats and other higher-priced foods) have lljl greater than one.23 Anotherformula that can be used to estimate T\.u is llii =E j [Sj - (l-sJ~/co)], where Ej is the expenditure elasticity of demand for commodity j, which is roughly equal to llil , Sj is the proportion of the consumer budget spent on commodity j, and co is the "money flexibility," which equals (du/dy)(y/u), where u is the marginal utility of money income and y is money income.24 While co can vary from zero to a large negative number, in most developing countries co would lie between -1.0 and -5.0, decreasing in absolute value as incomes rise. A value of co =- 1.0 might be typical of a middle-income developing country while co =- 5.0 would be typical of a low-income country. Supply Elasticities Domestic supply elasticities can be obtained from previous studies in the country or region. Most published elasticities of supply for agricultural prod­ ucts fall between 0.1 and 1.0. These elasticities, primarily estimated with annual time-series data, are likely to have been biased downward, however, as a result of problems with the specification of the dynamics of supply response, problems with specification of price expectations, incomplete representation of alternatives, and the nature of data in which prices of alternative products tend to move together (e.g., see Cassells 1933; Colman 1983; Burt and Worthington 1988; Just 1993)?5 Part of the problem is that supply elasticities increase with 23. Note that the classification of goods as luxuries and normal goods varies with income as, along with the patterns of consumption, income elasticities depend on per capita income. In some relatively rich countries, for instance, most meats would be normal goods, not luxuries. See Schultz (1953b) for a good discussion of this issue regarding farm and food products. 24. In this approach it is assumed that the marginal utility derived from each good is independent of the quantity of any other good consumed. This is known as want-independence. Demands for goods are still related through the budget constraint, however, and therefore want-independence does not imply the much stronger assumption of demand-independence. Pinstrup-Andersen, Ruiz de Londoiio and Hoover (1976), Pomareda (1978), and Norton, Ganoza and Pomareda (1987) used this formula to calculate own-price elasticity. See, also, Scobie (1980). 25. For these kinds of reasons, part of what is often attributed to trend and technological change may be due to changes in relative prices that are not captured by conventionally used price indexes (Griliches 1960; Peterson 1979). See, also, Binswanger et aI. (1987). 322 Economic Surplus Measurement and Application increases in the length of run (i.e., as more time is allowed for adjustments in response to a price change and as more "fixed" factors become "variable"). As pointed out by Cassells (1933), econometric studies tend to estimate interme­ diate rather than long-run elasticities. The supply elasticity depends in known ways on a number of things in addition to the length of run. The supply elasticities of the factors used to produce a commodity make up an important determinant its own elasticity of supply. Arable land can be a limiting factor, but often this will be so only for very important commodities (e.g., corn in the United States, rice in Asia) or very aggregated commodities (e.g., grains). Thus, the supply of relatively important commodities (or relatively aggregated commodities) tends to be relatively inelastic because of an inelastic supply of factors that account for a large share of the total costs of production. Livestock industries and perennial crops tend to have smaller supply elasticities because a component of specific capital (existing breeding stock or trees) is fixed for a time. Other things to consider are the ease of factor substitutability (greater substitutability among factors in the production of a commodity leads to a greater supply elasticity) and the nature of economies of size or scale in the industry (diseconomies of size or scale lead to a less elastic supply). Thus, theory suggests that an industry such as the tobacco industry - that uses only a little arable land and relatively few other specialized factors and which can be regarded as having close to constant returns to scale at the industry level- is likely to have a highly elastic supply.26 Similar arguments indicate that in most countries, the intensive livestock industries (pork, poultry, and in some cases, dairy) are likely to have relatively elastic supplies. Clearly, long-run elasticities for most individual agricultural products are greater than one (and for many products they may be infinite), but even short­ or intermediate-run supply elasticities are probably close to one. For most priority-setting work, the relevant length of run will be intermediate, and in the absence of better information, an elasticity of 1.0 is an appropriate starting point - this is especially true in relation to translating is (measuring horizontal shifts) into Ks (vertical shifts). Excess Supply-and-Demand Elasticities In many studies, it is convenient to represent the "rest of the world" in summary form in the analysis, using excess-supply and excess-demand concepts. For a given country (or region or group of countries), the excess­ supply/excess-demand function is given by the algebraic difference between 26. This theoretical idea has been borne out in empirical work by Sumner and Alston (1986), Goodwin and Sumner (1990), and Fulginiti and Perrin (1993) who all found a highly elastic supply of U.S. tobacco. Economic Surplus Measurement and Application 323 its domestic supply function and domestic demand function. Global market clearing can be represented equivalently by either (a) the intersection ofthe algebraic sum of all demand functions and the algebraic sum of all supply functions or (b) the intersection of the "rest-of-world" excess supply (or demand) with the "own-country" excess demand (or supply) when the home country is an importer (exporter). In the most common case, the home country is a small country, unable to influence world prices. Then the rest-of-world excess-supply/excess-de­ mand function can be represented by a horizontal (i.e., perfectly elastic) price line. In a few cases, a country or region will be able to influence world prices for a product, and then an estimate of the price responsiveness of excess supply/excess demand will be needed. Own-country and the rest-of-world excess supply-and-demand elastici­ ties can be approximated as follows. First, for a good that is exported by the home country, tXA = Qs,A ) ( tA + (Qd'A ) llA Qs,A - Qd,A Qs,A - Qd,A (S.2a) where tXA is the elasticity of excess supply (i.e., supply of exports) for the commodity in the home country (country A), tA is the domestic supply elasticity, llA is the absolute value ofthe domestic price elasticity of demand for the commodity, and Qs ,A and Qc/,A are domestic production and consump­ tion of the commodity. Qx ,A is exports of the commodity. Second, for a good that is imported by the home country, llMA = Qs,A ) ( tA + ( Qd,A ) llA Qd,A - Qs,A Qd,A - Qs,A (S.2b) = Qs,A Qd,A --tA +--llA Qm,A Qm,A where llMA is the elasticity of excess demand (i.e., demand for imports) for the commodity in the home country and Qm,A is imports of the commodity into the country. Third, for a good being imported by the home country, 324 Economic Surplus Measurement and Application (S.2c) = Qs,B Qd,B --EB + -- 11B Qm,A Qm,A where EXB is the excess supply elasticity of the commodity in the rest of the world (i.e., region B), EB and 11B are the elasticities of supply and demand (absolute value) in the rest of the world, Q.,B and Qd,B are the rest-of-world production and consumption of the commodity, and Qm.A is imports of the commodity into country A (equal to exports from country B). Finally, for a good being exported by the home country, Qs,B ) ( Qd,B ) 11MB = ( £B + 11B Qd,B - Qs,B Qd,B - Qs,B (S.2d) = Qs,B Qd,B --EB + -- 11B Qx,A Qx,A where 11MB is the excess-demand elasticity for the commodity in the rest of the world (i.e., the rest of the world's demand for imports) and the other variables are as defined above.27 5.2.3 Discount Rate and "Exogenous" Growth Factors Discount Rate Economists do not agree as to whether the appropriate social discount rate should reflect the alternative value of public resources in consumption or in investment. There is little disagreement, however, that when the analysis is conducted using benefits and costs expressed in constant value (i.e., real) terms, the rate should be a real rate of interest (adjusted for inflation), and most would argue that it should reflect any restrictions placed on alternative uses of the funds. In many situations, the real discount rate will fall in the 3% to S% range. This rate corresponds to a long-term, risk-free rate of return (e.g., the real yield from long-term government bonds).28 Ray (1986, pp. 27. For discussions of these equations and their simplifying assumptions, particularly with respect to the elasticity of price transmission, see Horner (1952), Tweeten (1967, 1977). Johnson (1977), Bredahl, Myers and Collins (1979). and Carter and Gardiner (1988). 28. In some places the government bond rate might include a risk premium, but it could still be appropriate to use as a measure of the opportunity cost of government funds. Some have suggested that a Economic Surplus Measurement and Application 325 92-101) discusses theoretical and conceptual issues surrounding the choice of discount rate and reviews the relevant literature. In chapter 2 we argued that it is inappropriate to adjust the discount rate to reflect the riskiness of research or, for example, to reflect concerns with sustainability because adjusting the discount rate does not account properly for concerns about risk or intergenerational equity, and other approaches are better. Not everyone will take this advice, so it is important when comparing net present values among studies to check what discount rates were used and to make sure they are comparable. 29 Sensitivity analysis is useful for assess­ ing the effects of the discount rate on the net present value of net research benefits. Section 5.3 discusses using the results from economic surplus and provides details on the use of discount rates in capital budgeting. Exogenous Growth in Demand For priority-setting work, projections of population and income growth rates for the next 15 to 20 years are needed to project exogenous demand shifts. In a future year, Tyears from now, the projected population, Nt+T> will be equal to the current population, N scaled by the exponential population growth rate, gN, " t+k' in year t + k: T Nt+T= Nt II (1 + gN, t+k) k=l (5.3) = (l + gNlN t if gN,t+k is invariant over time. Then, for a given per capita consumption, total consumption would be projected to increase in proportion to population - i.e" scaled up by (1 + gNf. In the absence of information to the contrary, the same approach can be applied for income growth, but it is necessary to multiply the projected growth in per capita income by the relevant income elasticity to deduce the implied growth in consumption. Thus, ignoring the effects of population growth, total con­ sumption of commodity j in T years' time would be projected to be Cj,t+T = [(1 + g[)T - 1] 'IlJl Cj ,/, where g[ is the exponential growth rate in per capita income. More details on these procedures are provided in appendix A5.1.2, which documents the Dream© computer model. higher discount rate might be appropriate for developing countries to reflect their greater scarcity of capital. This ought to be reflected in the government bond rate, if it is relevant. 29. In particular, according to Birdsall and Steer (1993), the World Bank uses a discount rate of 8% to 10% for project evaluation in the context of a less-developed country - on the grounds that this represents the opportunity cost of capital in developing countries. They claim that the opportunity cost of capital is higher in developing countries than in industrial countries, belying the globalization of interna­ tional capital markets. 326 Economic Surplus Measurement and Application Population growth rates can be included at the current rate estimated by the World Bank or other sources and then allowed to follow standard projections for the country over time. Recent historical experience and other factors can be considered in projecting income growth rates. Exogenous Growth in Supply The "without-research" quantity may be projected to change either as a result of changes in area (for crops) or herd size (for livestock) or as a result of changes in yield. Such changes could result from (a) responses to market forces with given technology (e.g., changes in output prices leading to greater planned output or changes in input prices leading to intensification), (b) price-induced changes in technology, or (c) the effects of research spillover. In addition, when the relevant alternative is less research rather than none, the relevant projection of output for the benchmark case is output given the lower amount of research. What is important here is to be clear about the alternatives being compared and, therefore, about what conditions are applicable for projecting future outputs. In practice, current output may be projected forward using recent past changes in output to infer a growth rate and to obtain a first approximation of a benchmark stream of "without-research" quantities. Then appropriate revisions to that benchmark stream can be made, based on advice from scientists and others for the case of either (a) no research or (b) a baseline program of research spending. In either case, the benchmark should incor­ porate the effects of research spill-ins and autonomous growth in supply in response to exogenous changes in factor and product markets. Alternatively, in some cases it might be preferable to project yield and area (or herd size) in the baseline case and to combine the results to project output. In the Dream© model, as described in appendix AS. 1.2, exponential growth rates in area and yield that are not attributable to research are added together to derive an overall exogenous growth rate of output. 5.3 Measuring the Research-Induced Supply Shift To measure changes in economic surplus due to research-induced shifts in the supply curve, information on variables that quantify the knowledge production function described in chapter 2 is required. That function relates research costs (and how they are deployed) to actual or expected per unit cost reductions or yield increases; to lags in research, adoption, and depreciation; and to probabilities of research success. Economic Surplus Measurement and Application 327 The size of the research-induced supply shift - the K-factor - is a crucial determinant of the total benefits from research. The accuracy of the estimate of K and its path over time, reflecting adoption lags and so on, will determine the accuracy and validity of the estimates of research benefits and any research priorities that are derived, based on those estimates. In short, K (and its associated distribution) is critical. The most important questions concern the size of the research-induced shift in a commodity supply function for a given expenditure on research and how that shift varies over time. In order to answer these questions for a given research program, the analyst has to combine technical, scientific, and eco­ nomic information from a number of sources. For ex ante analysis, some of the information can be obtained primarily from researchers and extension workers - especially technical information about the likely impact of the research on, say, experimental yields or on commercial yields under various scenarios and on the likely time path of the adoption of various technologies. However, it is important to use appropriate sources for any such information. Scientists are often unable to give meaningful answers to questions about impact on indus­ try-level costs or supply functions, so economists can play an important role in translating scientific information into economic information. Also, scientists might be too optimistic about their chances of success, the likely size of the eventual impact of their work if successful, the time required to complete the research, and the speed and extent to which it would be adopted. If the economic surplus analysis is ex post, the results of previous experimental trials can be used to assess changes in cost and yield. Even with ex ante analysis, such past results are indispensable for assessing information from other sources and providing a benchmark for future projections?O In section 5.3.1, we layout some of the conceptual issues that arise in relating changes in production, productivity, or cost to a measure of K to be used in research evaluation. Then in section 5.3.2, we consider some specific measurement issues that arise in particular alternati ve approaches to estimat­ ing K, including ex post econometric studies and ex ante approaches based on elicited information. The ex post measures obtained from econometric models reflect the adop­ tion response. In section 5.3.2, we emphasize how to measure the maximum shift corresponding to full adoption when research is successful. We tum to the modifications that must be made in ex ante evaluation in section 5.3.3 - to account for the distribution of possible research outcomes and the time path of the adoption. Sample questionnaires and some discussion of the practical 3(), In appendix AS.3, we discuss the use of experiment and industry data to help assess potential supply shifts at the industry level. 328 Economic Surplus Measurement and Application aspects of eliciting the interview information needed for ex ante evaluation and for validating responses are provided in appendix AS.4. Although variables are discussed individually, they interact in shifting the supply curve over time, amplifying the importance of the manner in which questions are posed to scientists, extension workers, and those asked to validate their responses. 5.3.1 Conceptual Issues At the level of the individual farm firm, a research-induced supply shift may be decomposed into two components: (a) one part arising from changes in productivity that would occur if input use were held constant at the optimum that applied before the technological change and (b) one part associated with changes in the input mix to optimize input combinations under the new technology. The latter augments the former. 31 In order to measure the impact on profitability, the measured increase in productivity should reflect a change to a new optimal input mix, and corre­ spondingly, measures of research-induced cost savings ought to reflect research-induced changes in the use of inputs and their opportunity costs. At the level of the individual firm, this means that estimated per unit cost savings ought to reflect the complete difference in the commodity enterprise budgets between the new and old technologies. In principle, this difference could be measured by preparing a detailed enterprise budget for production (applicable at the national or, if relevant, regional preresearch quantity), using each of the alternative technologies. The resulting differences in unit costs could be used as a measure of the research-induced reduction in marginal and average costs at the preresearch equilibrium. But it is difficult to measure these differences in a reliable or meaningful way for technologies that are not yet developed, let alone adopted. At best, we can make an informed guess about the likely impacts on yield or on some aspects of cost, and this guess will be conditional to holding some things constant that will not be constant in practice.32 This highlights the point that it is the result of producers optimizing their responses to the availability of new technology, not the new technology per se, that is relevant for measuring K. 31. This is according to the Le Chfitelier Principle - when a constraint is removed, you can do at least as well as, and possibly better than, you could when the constraint was in force. 32. Evenson (1992) argued that unless the details of the research program are known, the effects of the research will be unpredictable. This is a reasonable view. But it is rarely true that we have no infonnation to support making an infonned estimate of the likely research outcome, and such an estimate may be more relevant than ex post infonnation on past research effects, or it can be combined with information on past effects in a Bayesian-type approach. We do not agree with Evenson's (1992, p. 68) conclusion that "one may as well acknowledge that ex post evidence is all that one can bring to bear on such questions." Economic Surplus Measurement and Application 329 Results from using shortcut methods to infer this difference - such as translating a yield increase into a cost saving - must be adjusted to reflect changes in input use. For instance, as a matter of sound scientific practice, experimental yields generally hold the input mix constant across experimen­ tal alternatives in order to isolate the effects of the specific component of the technology being studied.33 Thus, to the extent that optimal input mixes vary among alternatives, the experimental yield increase (or corresponding cost saving) will misstate the economically optimal yield increase (or cost sav­ ing). If the experimental input mix is optimal for the old technology but not for the new technology, then the cost saving from switching to the new technology will be understated (i.e., futher reductions in costs will be achieved by changing the input mix). Conversely, if the input mix is optimal for the new technology, the cost saving relative to the old technology will be overstated (i.e., the costs under the old technology will be overstated because they could be reduced by changing to an optimal input mix). Alternatively, if the input mix is varied across alternatives so that it is optimal for each of the alternatives, it will be necessary to account for changes in input use in deducing the research-induced cost saving (this also includes a charge for the opportunity cost of so-called "fixed" factors if their use changes). And in going from yield changes to cost changes, it will be necessary to account for that part of the increased yield that is attributable to changes in input use. The same kinds of issues arise when the analysis is at the industry level rather than at the level of the firm, but they are buttressed with some additional ones. First, some inputs that are "variable" for firms (having exogenous prices) are quasi-fixed from the point of view of the industry. The endogenous prices of variable factors can complicate the evaluation of the impact of an input-saving or input-using technical change on the per unit costs of outputs, but they are unlikely to be a serious problem in most instances. Second, the prices of outputs that are exogenous to individual firms can be endogenous at the industry level, giving rise to the possibility of general-equilibrium feedback from related product markets through shifts of the industry supply function. Such changes complicate both the problem of measuring the research-induced supply shift and the problem of interpret­ ing it in relation to a welfare analysis and evaluation (e.g., see chapter 4 for a discussion). 33. Thus, for example, in typical variety trials, the use of chemical fertilizers and the timing of operations will be held constant across different varieties even though different varieties might respond differently and call for different agronomic treatments. Similarly, for instance, in order to maximize profits when using bovine somatotropin to increase the milk yields of dairy cows, there must be an increase in rations and a change in the composition of the rations and the feeding schedules for the cows, compared with the optimum when Ire growth hormone is not being used. 330 Economic Surplus Measurement and Application These ideas are represented schematically in figure 5.1: four curves are shown representing different industry-level supply functions under different technological and market conditions. Initial output is Qo. So represents the initial supply curve, reflecting the use of the optimal input combinations under the original technology. SI represents the supply curve that would result if the new technology were adopted but the input mix were the same as under the original technology (SI is shown as lying below So, but with certain types of biased technological change, where achieving cost savings relies on changing the input mix, it might not lie below So and could even lie above it).34 S2 represents the supply curve that applies when the optimal input mixes are used for the new technology, but this assumes that variable input prices are fixed and the quantities of "fixed" factors (such as land) used in Figure 5.1: Components of research-induced supply shifts Price o Quantityl Year 34. Of course producers would be unlikely to adopt such technologies if they did imply higher costs. A case where a new technology implies higher costs when input combinations are not optimized - but lower costs when they are optimized - provides a graphic illustration of the importance of optimaIly varying input mixes when technologies are compared for research evaluation. Economic Surplus Measurement and Application 331 producing the commodity are indeed fixed. The difference between the curves Sl and S2 is the cost saving due to optimizing the input mix for the new technology. Finally. S3 represents the supply curve after all optimizing responses have been made. including the drawing of "fixed" factors into (or out from) the production of the commodity whose profitability has been increased by the introduction of the new technology. Which ofthese supply shifts should we attempt to approximate to include as K in our assessment of research benefits: So to Sl = k1• So to S2 =k 2• or So to S3 = k3? The research-induced cost saving is understated by kl because it does not allow for economizing on the input mix. The unthinking use of experiment data or the results from a production-function econometric study (where the quantities of variable inputs are held constant for evaluating the output-enhancing effects of research) could lead unconsciously to a measure that corresponds to kl unless explicit account is taken of the input-mix change. Adjusting for the optimal input mix would lead to a measure that corresponds to k2 (e.g .• Bernhart and Perrin 1989; Lemieux and Wohlgenant 1989; Perrin 1992). This could well correspond to the measures of the research-induced supply shift derived from a cost-function study. for in­ stance (where the prices of variable inputs and quantities of output and fixed inputs are held statistically constant). as well as the measures derived from a thoughtful ex ante study based on experiment data. What about k3? The estimate of the research-induced supply shift from an econometric estimation of either a directly estimated supply function or a single-commodity cost function might correspond to k3• which represents the entire research-induced supply shift. including the component of cost reduc­ tion (or output increases) that is attributable to drawing in quasi-fixed factors (e.g .• allocatable fixed factors such as land in a multi-output setting). The problem is that the measure of k here is a measure of single-commodity cost changes. some of which have been achieved at the expense of cost increases (decreases in producer surplus) in other commodities. from the production of which the quasi-fixed factors have been drawn. The difference between S2 and S3 is not a net benefit; it is a gross benefit for which there is a correspond­ ing cost (associated with a leftward shift of the supply of competing prod­ ucts) and the net social benefit is zero (see Martin and Alston 1994 for a discussion and heuristic proof).35 Unless a full general-equilibrium analysis is being undertaken. in which case it would be desirable to explicitly measure the impact of commodity-market­ factor-market interactions of the type involved in shifting from S 1 or S2 to S3. it 35. In conunenting on Lindner and Jarrett (1978), Rose (1980) was concerned with this issue of allocatable fixed factors (especially land) and deriving appropriate measures of quasi-rents. 332 Economic Surplus Measurement and Application would be best to attempt to estimate k2 rather than k3• This implies adjusting estimates obtained from either scientists, experiment data, or production-func­ tion studies for changes in the input mix, with appropriate cost adjustments. It also implies adjusting measures obtained from econometric models for the impact, if any, of changes in quasi-fixed factors. An advantage of the sector­ wide econometric models of total agriculture over their indi vidual-commodity counterparts is that at the sector level (say provincial, state, or national agricul­ ture as a whole), quasi-fixed factors may reasonably be regarded as fixed.36 In individual-commodity studies, a significant component of the supply response to research might be a reflection of changes in intensity of land use, which might lead to an overstatement of research benefits unless an appropriate adjustment is made for the opportunity cost of land. In a sectorwide model, it is necessary to account for the effects of research program alternatives that involve shifting several individual-commodity supply curves simultaneously. 5.3.2 Practical Measurement As noted above, the key piece of information for any research evaluation study is the per unit cost reduction that has resulted from research or that is anticipated if the research is successful and the resulting technologies are adopted. A number of options are available for estimating K, depending on the purpose of the analysis data available, and tbe overall methodological approach being applied in the study. In ex post studies, the cost and impact of the research can be known and measured, at least conceptually; in ex ante studies, they can neither be known nor measured. In either case, we must make an estimate that will be subject to error, but in the case of the ex ante studies, we know much less about the statistical distribution of the estimation errors. Unlike ex post studies, in ex ante studies we don't know whether the research will be successful; we must estimate the odds of success. We must draw on people's subjective estimates (either as individuals or in some type of consensus approach), both of the costs of a research program and of a number of the components of the corresponding K-factor: the probability that research will be successful, the likely impact on productivity if the research is successful and the results are adopted, and the likely time path of adoption (replicated by region when a geographically dis aggregated analysis is being undertaken). 36. Offsetting this "advantage" is the fact that the resulting estimate of the aggregate K is a weighted average of the individual Ks across individual comroodities. This will be inappropriate for any particular comroodity when the individual Ks vary much among commodities. Economic Surplus Measurement and Application 333 Research Costs Information is required on research costs, both for relating the calculated research benefits to their corresponding costs and for developing a bench­ mark of the size of the research program for scientists responding to ques­ tions about expected yield changes, probabilities of research success, and so on.37 Historical data on research costs are necessary for ex post analyses and are useful for establishing a benchmark for ex ante studies. For many developing and developed countries alike, cost data are seldom broken down by commodity and research program. Hence, it can be useful to use informa­ tion on numbers of scientists working in different programs (or, more appropriately, their full-time equivalents) to assist in apportioning the costs. Administrative costs must also be spread across programs (see chapter 3 and appendix A5.4 for details). The evaluation exercise requires a statement of total research costs for each program alternative to be evaluated. At the outset, it is usually appro­ priate to define a benchmark of the current research effort in terms of the total resources being invested, the composition of those resources - capital (buildings and equipment), personnel (disaggregated between nonscientists and different types of scientists), and operating expenditures, as well as their deployment among current programs. This information can be obtained either (a) from "central administration" files, reports, and other documents or (b) by asking scientists or administrators for information on the current scientific staffing and costs for each research program. Discrepancies from these alternative approaches to gathering data on the current research effort are likely to be evident in the results, and it may help to combine the two. The advantage of the second approach is that the costs reported by scientists are based on their recent experiences.38 The benchmark value for each research program can be used as a base for the analysis when information is elicited from scientists about expected research results. The base can be varied (perhaps 10% higher or lower) to 37. When an extension component is explicitly involved as part of the research program, its costs must also be measured. The inclusion of this aspect can have implications for the benchrnarlcing of research impacts, especiaUy as extension would be expected to affect the adoption process. In addition, it may be appropriate to scale the research program costs upwards to reflect a measure of the full social opportunity cost of government funds (e.g., as discussed by Fox 1985 and Dalrymple 1990 and done by Alston and Mullen 1992). Care should be exercised to ensure that results from different studies are consistent in their treatment of this issue. 38. A third approach can be used in which research costs are ignored initiaUy, a particular percentage (or set of percentages) reduction in production costs due to research is assumed, and then scientists are asked what it would cost to arrive at that percentage reduction (Davis, Gram and Ryan 1987). This has implications for the probabilities of success, research lags, and so on that are relevant. Various approaches to incorporating costs can be used, but the choice will influence subsequent steps in the analysis. 334 Economic Surplus Measurement and Application elicit information on expected results for different total research invest­ ments.39 Another plausible alternative to evaluate might concern the rede­ ployment of existing resources within a current program - i.e., changing the spatial orientation of the program, changing the problem orientation of the program (e.g., in relation to genetic improvement, agronomy, or pest and disease control), or changing the ratio of scientists to support staff. One way to help scientists provide meaningful information is to ask them what would be possible if resources were "entirely unlimited." Scobie and Jacobsen (1992) asked scientists two questions: First, how much total scien­ tific resources could be spent productively on a given research program? Second, given that amount, how would you spend it and what would be the expected outcome? As well as providing an observation of the maximum point of the research production function attainable in the short to medium run, this approach provides a fixed reference point against which to define what is possible with more realistic levels of resources - our primary concern. In this way, information is provided on the scientists' view of diminishing returns in the research production function. Econometric Measures of Research-Induced Supply Shifts Ex post evaluation studies using econometric approaches (as described in chapter 3) will, as a matter of course, yield parameter estimates that either directly represent the research-induced supply shift (when supply functions are estimated directly) or that can be translated into a measure of the research-induced supply shift (when a production function, cost function, or profit function is estimated with research as an argument or when index­ number methods are used to derive a measure of productivity growth). Directly estimated supply functions: A number of studies have esti­ mated commodity supply functions directly, with past expenditures on agricultural research and extension included as arguments.411 In these studies the lag relationships between research and adoption are estimated jointly as part of the supply response to research. Thus, it is possible to use the estimated supply function and deduce (or simulate) an entire time path of research-induced supply shifts associated either with the total research in­ vestment or with marginal changes in it. For example, suppose the supply function for commodity j was estimated as follows: 39. Scientists might not appreciate the possible impact of a 10% reduction in total budget. Often a higher percentage change in operating costs (say 30% to 50%, perhaps implying a 5% to 100/0 change in total costs) is used in this elicitation instead, in order to get scientists' attention and to obtain a meaningful response from them about the implied change in research productivity. 40. Examples include Otto (1981) and Zachariah, Fox and Brinkman (1989). Economic Surplus Measurement and Application 335 (5.4) where in year t, Qi,' and Pi,' are the quantity and price of commodity j and R,_, is expenditure on research in year r-years past.41 Then, assuming the past effects of research carry forward into the future, the future time path of proportional reductions in the marginal cost of the commodity (i.e., shifts down in supply relative to the commodity price) for a one-shot marginal change in research spending (say $1) in the current year, t, could be projected as kj,t+r = ~jr/(~j ~j,t+r) for r = 0 , ... , LR (S.Sa) where the "hats" denote projected values of variables and estimated values of parameters. Alternatively, the model could be used to simulate the impact of a permanent change in funding that would involve summing across research­ lag weights for the relevant years in which research had changed. For instance, to simulate the effect in the current year of research being reduced by $1 in all past years, the corresponding estimate of Ki" would be (S.Sb) Either equation S.Sa or S.Sb could be used to generate a stream of estimated values of K to be used in an economic surplus model for evaluating research. In each case, the directly estimated supply shift reflects the com­ bined effects of adoption and of supply shifts for those who adopted. The difference between the two is in the counterfactual experiment being carried out: a temporary or a permanent change in research funding by $1 per year. Production functions: For estimating aggregated rates of return for agri­ cultural research as a whole, a more common approach has been to estimate a production function, or productivity function, in which past expenditures on research (and extension) are included as arguments.42 In these studies, the research and adoption lag relationships are estimated jointly as part of the output response to research. Thus, it is possible to use the estimated production function and deduce (or simulate) an entire time path of research-induced supply shifts (reflecting either input savings for a given output or additional 41. This is similar to the corresponding model developed in chapter 3, but it is different in that (a) the "other" supply-shift variables are suppressed here for simplicity (subsumed in (Xi) and (b) the research variables have not been preaggregated. 42. Examples include Griliches (1963a, 1964), Bredahl and Peterson (1976), Evenson (1967), and Scobie and Eveleens (1987). 336 Economic Surplus Measurement and Application output due to research) associated with either the total research investment or marginal changes in it. For example, suppose the production function for commodity j was estimated, using a Cobb-Douglas model, as LR InQj,t = aj + ~j In Xj,t + 1: 0jr R t- r (5.6) r:=1 Then, in terms of percentage change, the time path of changes in output (or productivity) for a one-shot marginal change in research spending (say $1) in the current year, t, would be projected as Jj,Hr = Ojr. 43 This could be translated into a shift down in the price direction by dividing by the supply elasticity: ~,I+r =0 if,/4 It is important to be aware that, as discussed above (in relation to figure 5.1), this measure represents a supply shift only under restrictive conditions that might not be satisfied at the industry level (and assuming that the future mirrors the past).45 Cost and profit functions: Modem duality-based specifications of sys­ tems of equations representing output supply and factor demand permit the joint estimation of research impacts in multiple dimensions. There have been a small number of studies that have taken this type of approach to measure the size and bias oftechnical changes (e.g., Lopez 1980; Ray 1982; Ball and Chambers 1982; Zhang et al. 1993), but they have usually included one or more time-trend variables and have not included explicit research variables. One exception is Huffman and Evenson (1989), who estimated a multi-out­ put profit-function system, representing aggregate U.S. production of crops and livestock products on cash-grain farms and incorporating a range of research and extension variables. As with the supply or production-function models mentioned above, the parameters from the profit (or cost) function can be translated into measures of a research-induced supply shift. Productivity functions: A common alternative approach is to calculate a measure of productivity (or productivity growth) using index-number proce­ dures. Econometric models can be applied to estimate the relationship between productivity and past research (and extension) investments, among other 43. If In R were used instead of R in equation 5.6 ( a more typical approach in the Cob~Douglas model), then OJ, would measure the percentage change in output for a one-percent change ratrer than a unit change in research expenditure. 44. An equivalent approach, in many respects, has been to calculate the productivity growth attribut­ able to research using index-number procedures or assumptions about technology and to regress those estimates against distributed lags of research and extension expenditures. The results from those studies could be translated into supply shifts as was done here. See chapter 3 for further details. 45. The critical assumptions are (a) exogenous prices of "variable" factors (i.e., there are no variable factors at the finn level that are specialized at the industry level) and (b) exogenous quantities of "fIxed" factors (i.e., there are no allocatable fIXed factors). Economic Surplus Measurement and Application 337 things. The econometric estimates can be used to deduce values for K to be used to estimate streams of research benefits (chapter 3 provides details). In a recent article, Cooke and Sundquist (1993) proposed a procedure for measuring the "K-shift" in the supply function when new technology is introduced. They illustrated their procedure in an application measuring benefits from growth in U.S. soybean productivity. Their approach involves several critical explicit or implicit assumptions. In particular, it seems that their approach may be valid only under an assumption of constant returns to scale, and it might require an additional assumption that average and mar­ ginal costs are equal, which would imply a horizontal supply curve. The K-shift is defined by Cooke and Sundquist (1993) - drawing upon Lindner and Jarrett (1978), Rose (1980), Edwards and Freebairn (1984), and by implication, Mishan (1972) - as being equal to both (a) the proportionate reduction in average cost of production excluding rent, relative to initial aver­ age cost excluding rent, and (b) the proportionate shift down in the equilibrium supply curve (i.e., marginal cost), relative to the inital price (or marginal cost). This would seem to be equivalent to assuming that marginal and average costs are equal both before and after the supply shift. Cooke and Sundquist (1993) illustrate the problem of not being able to observe this K shift, the research-in­ duced change in costs, directly. They propose to measure it indirectly by first measuring an index of total factor productivity growth and then deriving a cost-efficiency index from that. This second step uses theory that relates an index of total factor productivity growth to an index of cost efficiency. Cooke and Sundquist (1993, p. 173) argue that "a Fisher cost efficiency index approximately equals the inverse of a Tornqvist total factor productivity index (Diewert, 1976, p. 124). Both indexes are 'superlative' in that they re­ flect second-order approximations of nonhomothetic production and unit-cost functions, respectively. Therefore, Ka or the proportionate reduction in average cost excluding rent is approximately equal to one minus the inverse of the Tornqvist index of total factor productivity." Thus the proportional reduction in average cost excluding rent (i.e., K) is measured by the proportional growth in total factor productivity. This seems to be remarkably simple and easy. Cooke and Sundquist do not discuss any restrictions on the production technology that are needed to make this measure valid. In chapter 3 we suggested that under the assumption of constant returns to scale - in addition to the assumptions of input-output separability, efficient and opti­ mizing producers, and disembodied technical change of the extended Hicks­ neutral type - the rate of change of total factor productivity (TFP) given by equation 3.29 would be equal to both the primal rate of technological change (or the shift of the production function) and the rate of dual technological change (or the change in cost of production). Similar assumptions are likely 338 Economic Surplus Measurement and Application to be required for validity in the Cooke and Sundquist approach, notwith­ standing their explicit suggestion, quoted above, that their approach would apply to nonhomothetic technologies. There are even more fundamental problems in using proportional changes in measured TFP as estimates of K-shifts that in turn are used to measure benefits attributed to R&D. As we also discussed in some detail in chapter 3, measured growth in TFP occurs for a good many reasons in addition to agricultural R&D or, more particularly, public-sector R&D. Some of the additional sources of growth in measured TFP include mismeasured or unmeasured improvements in the quality of inputs, such as land and labor, through the provision of irrigation, communication, and education services and the acquisition of these services by individuals, firms, and governments; unrecorded quality improvements in seeds, machinery, and agricultural chemicals due to private R&D; and economies of scale. Simply attributing all of this productivity growth to R&D will surely overstate the implied benefits from public-sector research and in some cases seriously so. Experiment and Industry Data Experiment data: Experiment data take several forms. A major distinc­ tion in kinds of experiment data is between the results from long-term trials over many years (say, monitoring crop yields with old and new varieties) and the results from one-shot or short-term studies (e.g., studies of particular cultural aspects, fertilizer trials, or specific varietal comparisons). The latter often allow an investigation of crop performance at a number of locations, whereas the former allow an investigation over time. Data from long-term trials allow response functions to be estimated for the specific alternatives included under a range of weather conditions; data from one-shot multilocational trials allow paired comparisons of response functions under a range of agroecological conditions. Thus, the different types of experiment data provide different types of information about differ­ ent types of research questions. Of course neither long-term trial data nor one-shot multilocational trial data are perfect because, for research evalua­ tion, we want to control for both site effects (i.e., over space) and weather (over time). Distinguishing the effects of uncontrolled factors is sometimes difficult with either type of experiment. Ex post studies46 might use experiment data and information on adoption rates and so on to deduce the pattern of past supply shifts attributable to a 46. Experimental yields have been used in ex post analysis as a proxy for industry yields or supply shifts in a number of more recent studies. Examples are provided by Echeverria, Ferreira and Dabezies (1989), Pardey et aI. (1992), and Palomino and Norton (l992b). Economic Surplus Measurement and Application 339 particular set of technological changes (e.g., Griliches 1957b). In ex ante studies, too, it is necessary to integrate information about the per unit cost saving for those who adopt the technology with information on the time pattern of adoption - often, in both cases, information that involves people's subjec­ tive judgments. For ex post work, results of on-farm yield trials and other experiment data for key types of research are essential for estimating per unit cost reductions or yield changes for farmers who adopted. For ex ante analysis, this information is also useful as background information for the scientists who are being asked to project future cost or yield changes.47 It is not always easy to translate increases in experimental yields into industry-level cost savings - even when we put aside the questions of adoption responses. Experimental yields are typically higher than commer­ cial yields (the so-called "yield-gap" phenomenon) and gains in experimen­ tal yields often exceed gains experienced on farms.48 But the sizes of these differences vary among places and technologies, and it is difficult to make empirical generalizations. There is some basis for scaling down experimen­ tal yield gains to better reflect likely on-farm gains - but it would probably be an overcorrection to scale down the gains in proportion to the difference between research-station yields and on-farm yields. In addition to differences between experimental and farm responses, and differences between individual farm and industry supply responses (discussed above), there are issues arising from the factor bias of the technical change interacting with factor cost shares, elasticities of factor substitution, and elas­ ticities of factor supply. Appendix A5.3 examines the relationships between the industry (final product) supply shift, experimental yields, and changes in industry yields for different types of technical changes. The relative increase in experimental yield, Y, will translate into an equal, proportional, rightwards shift of industry supply in the quantity direction (i.e., dYIY = E(l') = 1) under a neutral technical change with fixed factor proportions. To translate this into a measure of K (the percentage shift down of supply in the price direction), we divide by the elasticity of supply: i.e., K = lie = E(l')/e. If the technical change is not neutral (i.e., there is some factor bias, as is most likely) or if the factor proportions are not fixed (i.e., some factor substitution is possible, as is most 47. In many cases a detailed record of the results of an experiment is not available, and occasionally even summary statistics on experiments are not kept. In such cases there may be no more information available than avemge gains in experimental yields, perhaps only at one location. However, in some cases useful records have been maintained and it can be possible to develop measures of the site-specific effects of new technologies on the avemge outcome as well as the variability of outcomes. 48. For additional discussion, see Swanson (1957), Johnson (1957), Davidson and Martin (1965), and Davidson, Martin and Mauldon (1967). Scobie and Posada (1977, 1978) dealt appropriately with this issue when attempting to estimate returns to research on improving rice varieties in Colombia. 340 Economic Surplus Measurement and Application likely) then experimental yields will not translate so simply into industry supply shifts. The value for the industry supply-response elasticity, £, is a critical factor in converting an experimental yield into an industry-level, per unit, cost saving. When actual or hypothetical experimental yields are being used to deduce values for K and information on supply elasticities is lacking, it is often expedient to use a supply elasticity of 1.0.49 Industry data: Historical data on yields over time and in different locations are often more readily available and more complete than experiment data. These may be analyzed informally or using statistical techniques. Yields may have changed over time for a host of reasons in addition to the research-induced technical changes of specific interest. These reasons can include secular pro­ ductivity gains arising from other research carried out locally, including pri­ vate-sector research, or as a spill-in effect. They can also include the transitory effects of weather variations and of pest and disease factors. A careless use of statistical methods could be misleading. With appropriate care in model speci­ fication and interpretation of the results, best done in consultation with relevant scientists, the historical record can be very informative for providing a bench­ mark for current situations and potential future changes. Such data are partic­ ularly useful when combined with, and juxtaposed against, experimental data (e.g., Pardey et al. 1992; Constantine, Alston and Smith 1995). Eliciting Kfrom Scientists In some cases, a commodity enterprise budget can be developed with old and new technologies. The proportionate cost reduction can then be based on this information. More often, for priority-setting work, researchers are asked to project the percentage yield increase (or, in some cases, cost reduction) as a "best-guess" estimate.51) If scientists provide estimates of yield changes result­ ing from new technologies (as opposed to per unit cost reductions), they should be asked about changes in input requirements so that changes in input costs can be netted out in translating yield changes into cost changes. The effects of research expenditures on per unit costs or yields are unlikely to be in constant proportion, independent of the scale of expenditure, and research on some commodities is more expensive than on others. Thus, scientists can be asked about the effects of research at two or three levels of funding or staffing. 49. When using lIE to estimate K. clearly the value of the supply elasticity is critical. For instance a supply elasticity of 0.1 would imply a value for K 10 times the value of 1 and 10 times the value implied by a supply elasticity of 1.0. or one-hundredth that implied by a supply elasticity of 10. Thus. a 10% yield increase could be translated into a 100% cost saving or a I % cost saving. depending on whether the supply elasticity is assumed to be 0.1 or 10. 50. Scientists usually find cost changes m~h more difficult to estimate than yield changes. Hence. their cost estimates tend to be less precise than their yield estimates. Economic Surplus Measurement and Application 341 This estimation (or elicitation) process is not simply a matter of asking scientists to provide a number to be used as a value for K for each program. Deducing a value for K (or, more appropriately, a distribution of KMAX) from what scientists know involves combining various pieces of information intel­ ligently in a structured fashion. Consider the example of a research program for a particular crop. In order to elicit a meaningful value of K (and spillover effects) for a given program of expenditure, it is necessary (a) to look in some detail at the individual impacts of components of that program of expenditure (e.g., plant breeding versus agronomy versus pest control work), (b) to consider the substitution or complementary relationships among the components of the research program, and (c) to obtain a clear picture of the relevant alternatives (e.g., what would happen to yields in the absence of the research).51 Research program components: The different components of a research program are likely to have different potential effects on yield, creating a possibility for aggregation bias. But apart from the potential for aggregation bias, there is another reason for considering their disaggregated effects. Some scientists may not be able to give a sensible estimate of an entire program's impact on yield, even when they have very good information about its individu­ al components. Further, different scientists know about different components, and different program components can yield very different time profiles for cost savings. For example, research lags and the rate of uptake both differ between, say, developing a new, disease-resistant variety and developing new fertilizer recommendations. Eliciting information on the individual compo­ nents jointly with information on their aggregate effects provides a check on the consistency of the estimates. Such information can also allow a structured assessment of intraprogram allocations of resources to research. In all such work, it is necessary to be clear about what is being held constant, something that becomes potentially more serious when disaggre­ gated components are being dealt with.52 Even if disaggregated information 51. Even within relatively narrow "fields" of research there can be significant diversity and associated aggregation questions. For instance, plant breeders have many objectives in breeding new cultivars, including improving yield potential, pest and disease resistance, tolerance of adverse environmental conditions (e.g., cold and drought), and a number of different grain characteristics thatinteractto determine "quality." Apart from the determinants of quality, many other genotypic characteristics have their main expression through yield. Therefore yield is an extremely useful summary statistic, which reflects a diversity of objectives pursued by researchers. But yield changes may not reflect all the cost changes associated with the adoption of new technologies. Traxler and Byerlee (1993) show the economic trade-offs involved in the effects of grain yield versus straw in modem, semidwarf wheat varieties. See also Byerlee, Igbal and Fisher (1989) .. 52. This disaggregation could be taken beyond the point where it is useful. As the degree of disaggregation increases, the potential interactions aroong disaggregated components and the difficulties of obtaining meaningful estimates can quickly become overwhelming. 342 Economic Surplus Measurement and Application is not collected formally (say on the individual components of a research program or in relation to different agroecological zones) and it is decided to estimate an aggregate supply response directly for a given program of research, it can be very useful to talk through these points first with the scientists and to have some relevant past data at hand. Aggregation, substitution, and complementarity: Estimates of the yield­ enhancing impact of individual components of the research program can be added together so long as they can be regarded as independent. More often than not, they will not be independent. For instance, high-yielding "green-revolu­ tion" varieties were relatively responsive to new crop management and fertil­ izer regimes, so there was a complementarity between research on variety improvement and management. And the joint impact of the components of the technological package was greater than the simple sum. On the other hand, a new high-yielding variety will preclude the adoption in a particular selection of an alternative disease-resistant variety, and in this context the total potential effect of the research program will be less than the sum of the potential effects of its individual components. The problem of double counting mutually exclu­ sive technologies is likely to be more pronounced with programs that involve multiple institutions or mUltiple research sites conducting similar lines of research designed to produce technologies that can be adopted in the same places. These positive and negative interactions may be accounted for by drawing the different scientists together and eliciting their views on the com­ bined effects of the entire set of program components. 53 These considerations are also directly relevant to the notions of maintenance research and research spillovers discussed below. Maintenance research and benchmarks: When scientists' projections of the yield impact of research are being elicited, it is essential to be clear about the reference point. At a minimum we want two sets of projections: one in the absence of the research program being evaluated and the other if the research program is successful and fully adopted. From this perspective, there is no meaningful distinction to be drawn between maintenance research and any other type of research. 54 All research in this context is directed toward increas­ ing yield relative to what it would have been otherwise. Unless scientists are questioned carefully, they are only likely to give estimates of changes relative to current yields. But differences between current yields and future yields (without research) might reflect perceptions of a decline due to deterioration or 53. Of course, if detailed data were available on the yield effects of each component, and correspond­ ing specific adoption rates, the potential for double counting innovations could be avoided in a much more explicit way. 54. Maintenance research is typically defmed as the research required to maintain the status quo in terms of yields or costs of production per unit of output. Economic Surplus Measurement and Application 343 obsolescence, or a gain due to other innovations, that must be distinguished from future yield changes due to the research program of interest. One way to develop benchmark yield projections is to project historical time-series data forward statistically as a basis for discussion (see the discussion in this section on industry and experiment data). This could be done with spatially disaggre­ gated data when there are significantly different yield trends (or differences in anticipated yield response) among agroecological zones. Where it is believed there will be significant differences in yield response to research among zones, a disaggregated treatment of zones in the elicitation process may reduce spatial aggregation biases. Local and spillover effects in an international setting: In a multina­ tional market model, in the context of a NARS evaluation study, the supply­ and-demand functions for other countries must be projected for a baseline simulation. Again, ceteris paribus issues arise: what is being held constant when those projections are being made, and is there potential for double counting? For instance, a projection from the historical trends in wheat yields in Australia would be based on past yield growth that was driven in consid­ erable part by technology spillovers from CIMMYT (Brennan 1986). The analyst modeling the effects of future CIMMYT wheat research might want to project Australia's wheat production in the absence ofCIMMYT research, and the gross historical pattern may provide little guidance. Where spillovers are likely to be important in the future, they are likely to have been important in the past, and vice versa. Spillover effects are difficult enough to analyze when research results that are embodied in technologies (e.g., new machinery, production practices, or crop varieties) are being examined. Further difficulties arise when research results themselves are being transferred. For example, a new crop variety could be used directly by farmers, but it might, instead, be used as an input into an ongoing breeding program to produce further new varieties. Measur­ ing the potential spillover effects is a greater challenge in the latter case than in the former. Some studies have attempted to explicitly forecast own-country research effects while simultaneously allowing for the adoption of research results from other countries. In such work asking scientists to make forecasts they might be unqualified to make, or using rules of thumb related to spillover potentials, introduces a real potential for double counting (Pardey and Wood 1994). With a number of countries involved in such a model, all generating technologies that can spill over internationally, there is a greater potential for double counting research results than in the more common approach where only one country's R&D is explicitly included, albeit perhaps with multi­ country impact. 344 Economic Surplus Measurement and Application Multicountry studies might involve an analysis of the global or regional (e.g., west African or southeast Asian) consequences of research done by individual countries within the region, or they might focus on a country of interest, taking its multi country context into account. The regional conse­ quences of research will not be equal to the sum of the effects across countries within the region unless care is taken to define the individual country measures so that they are mutually consistent (i.e., to avoid double counting or other mismeasurement of spillover effects). International research spillovers are important for both multicountry stud­ ies of the effects of research from a multinational perspective (a perspective that we are not taking in this chapter) and for country-specific studies (the perspective we have adopted here), where research and technology spill­ overs in both directions can modify the domestic effects of the research undertaken by the country of interest. In either kind of study, it is important to be clear about which of these perspectives is being adopted before deciding what types of information to collect for analysis. Also, as in any economic analysis, it is important to be clear about what is being held constant and what is being allowed to vary between any pair of alternatives that are being compared. When one country's research is being evaluated in the context of a multicountry model with international price and technology spillovers, such ceteris paribus considerations assume particular importance. Explicit decisions must be made about whether research by other countries will be held constant at a baseline (with corresponding baseline spill-ins to the country of interest), held constant at zero (with zero spill-ins), or allowed to vary in response to research by the country of interest. Different choices here may involve eliciting different information for the analysis. Vagueness here may mean that the elicited information does not correspond to the information required for the particular analysis that is eventually carried out. Further questions arise concerning whether effects are additive or mutu­ ally exclusive. Consider, for example, a situation in which all countries directly experience an increase in productivity from locally conducted re­ search. If spillovers among all countries are presumed to be additional to their own research effects (Le., country i can simultaneously adopt, to some extent, all technologies developed everywhere else), then .,MAX MAX MAX .,MAX J<.T = Sil KI,l + ... + Sii Ki,i + ... + Sin J<.~~ (5.7) where IS~AX represents the maximum attainable local effect of country or region j's research and the Sij coefficients are multipliers that reflect the confluence of factors determining K~AX, the maximum potential impact of country j's research on supply in country i, given K'!/x, so that K~AX = Economic Surplus Measurement and Application 345 Oij K'tJAX.55 With this formulation, the Oij coefficients represent the cross-coun­ try or interregional "transferability" of research results to the extent that they reflect the maximum potential shift in supply when using region) technolo­ gies in region i relative to the local impact of region) technologies. Letting 0u = 1, equation 5.7 simplifies to n ~AX = K';jAX + ~ Oij KfJAX (5.8) jF.i However, a more plausible specification would adjust for the likelihood that some imported and domestic technologies are mutually exclusive (e.g., it is not possible to simultaneously adopt high-yielding domestic and foreign varieties on all hectares). In other words, we must account for differences between potential spillovers gi ven by 5.7, and actual spillovers, allowing for crowding-out effects. Avoiding the problems of double counting requires additional information about the local rate of uptake of these various new technologies. Introducing adoption parameters into equation 5.8 gives (5.9) where AU,' is the local (i.e., region i) rate of uptake of locally produced technologies in period t, A;J,' is the period t rate of adoption in region i of technologies developed in region), and K; / is the overall supply-shifting effect of local and non local technologies in period t.56 Obtaining plausible estimates of the n (2 + t) values for the A;J," O;j' and K'tJAX parameters required to estimate the K;"s is clearly a tall order. It is asking a lot (if not too much) of scientists and others to have meaningful views on the supply-shifting effects of both local and spill-in technologies that have yet to be developed, as well as the likely rate of uptake of these new technologies. It may be reasonable to seek opinions about the local effects of local research from scientists familiar with or likely to carry out the research. However, it is unreasonlj,ble to expect them to have enough knowledge of the current or planned research being done in other countries to be able to give plausible and highly disaggregated estimates of the spill-in potential of 55. Thus 9ij = K1fX IKf/X. The 9ijs usually range between oa nd 1 but may be greater than one if the research results are better suited to the region into which the research spills (i.e., region i) than the region where it was done (i.e., region)). One of the difficulties with this approach is that simply by disaggregating the world further, one can obtain bigger effects (e.g., by adding up a greater number of mutually exclusive technologies). This leads to the implication that, when in doubt, there might be "gains from aggregation" in terms of reducing the rest of the world to a single aggregate. 56. As elaborated below, to avoid double counting benefits, the Aj,j,ts and AjJ,ts must be defined appropriately so that the adoptions are mutually exclusive when that is appropriate. 346 Economic Surplus Measurement and Application technologies. At best, they may have some relevant knowledge about spe­ cific research being done in selected countries. Consequently, they may only be able to provide broad indications of the aggregate spill-in potential of research, based on past experience and a limited knowledge of related work being done by researchers in other locales. In the face of these constraints, there are few options here to reduce the information required to estimate the Ki./s. One approach is to presume, in the absence of compelling evidence to the contrary, that the rates of local uptake of all nonlocal technologies are roughly equal (i.e., A;J.r "" A;.r for all}:t:. i) so that (5.10) With this assumption, it is possible to preaggregate (or, more usually, form an estimate of) the local supply-shifting effects of all imported or spill-in technologies, (i.e., to estimate the overall spill-in effects, Lj;<; Elij K~AX), scale this aggregate by A:.r' and add it to the local impact of locally produced research to estimate the overall supply-shifting effects of research. A further simplification is to assume identical lag structures for all technologies, irrespective of their source, so that n (5.11) = A1·,·I, t LEl··K~l}A X IJ j=1 Clearly, adopting assumptions or rules of thumb along the lines used to form equations 5.10 and 5.11 is a recipe for double counting. 57 These procedures pay no regard to the types of technology being developed at each site and presume that the local impact of all these technologies are additive. In many cases this is an unrealistic assumption to make. If two countries develop new crop varieties, it is inappropriate to think of the combined supply-shifting effect of the two technologies to be a simple sum of their separate effects in a given locale. If any rule of thumb were to be applied, it might be more reasonable to 57. Davis, Oram and Ryan (1987) and Ryan and Davis (1990) used similar variants of these simplifying assumptions concerning the adoption aspects of local and spill-in technologies. Davis,Oram and Ryan appear to have assumed that the ceiling levels of adoption for local and all spill-in technologies were equal, while Ryan and Davis (p. 11) varied these ceilings in unspecified ways based on considerations of "rural infrastructure such as roads, fertilizer consumption, and so on." In both studies, the shapes of the local and spill-in adoption profiles were equal while the "mean adoption lag" for all spill-in technologies was taken to be 12 years, one year longer than the corresponding lag for local technologies. Economic Surplus Measurement and Application 347 assume that there are negligible spill-ins of nonlocal varieties (i.e., A;J,' ::: 0 for all j "# i) on the presumption that locally produced varieties (for local condi­ tions) will generally outperform varieties transferred from elsewhere. But even this presumption has been questioned recently regarding the international transferability of CIMMYT wheat varieties that are apparently widely adaptable (Maredia 1993; Traxler and Byedee 1994). In reality, farmers in a particular country or region within a country often use a mix of locally and non locally developed varieties at any point in time. This is partly because varieties do not generally perform uniformly well within the spatial aggregates commonly used for evaluation purposes, so wtihin a gi ven area, different varieties will find different niches in which they perform best. In any event, it is simply not enough to know the local supply-shifting effects of local and spill-in technologies to form an estimate of K;". Some notion of the types of technologies under development will help identify whether the technologies are potentially "complementary" (e.g., locally developed management practices and spill-in varietal technologies) or potentially "substitutable" (e.g., competing varietal technologies). If they are complementary, they may well have somewhat similar patterns of adop­ tion, but if they are potential substitutes, for instance, they may have dissim­ ilar, and indeed polar, patterns of adoption. So to estimate K;" requires knowledge of both the relevant values of K~AX and the corresponding A;J," In addition to the problems of translating potential effects into actual or realized effects (i.e., translating K~AX to K;,r), there is the problem of estimat­ ing the values of K~AX themselves. Rather than directly eliciting or estimating the values of K~AX (the maximum attainable supply-shifting effect of regionj technology in region i), an alternative and commonly used option is to jointly estimate the domestic impact of region j technology (i.e., ~~AX) with the performance of region j technology in region i relative to its performance in regionj (i.e., 9ij)' A consideration of the agroecological basis for variation in the spatial performance of past or potential technologies is helpful in forming these estimates. Ongoing developments in geographical information system (GIS) (in conjunction with elicitation techniques, crop simulation models, and so forth) will yield more structured and, it is hoped, more realistic estimates of the K~AX and 9ij parameters. These approaches reflect the influence of varia­ tions in agroecological conditions on the performance of many agricultural technologies. The potential yield superiority of a new crop variety is likely to vary less across areas that have similar edaphic, terrain, and climatic character­ istics than across dissimilar agroecological zones (AEZs). Using GIS procedures to overlay AEZs on geopolitical regions and existing production areas makes it possible to develop more refined estimates of the values of K;J by disaggregating region i into a series of agroecological zones. 348 Economic Surplus Measurement and Application Using this approach, J~~ (the output- or yield-enhancing response in zone z of region ito regionj's research) can be obtained and then summed across the relevant agroecological zones to form an estimate JWx so that (5.12) where Qj~z is the preresearch output in the zth zone of the ith region and (f! is the preresearch output in region i. Because supply curves are aggregated horizontally, the aggregate supply-shift effect is formed by horizontally sum­ ming across values for J;j,Z (the research-induced shift in output quantities holding output price constant), rather than using output shares as weights to vertically sum the cost-saving effects of research (i.e., the corresponding values of K;J',z s). This approach can improve the precision of the estimates of K~AIJX when there is substantial spatial variation in the effects of research because of agroecological diversity; it also provides information on the trade-offs in­ volved when research is targeted to different zones within a region.58 Local and spillover effects in a domestic setting: Similar problems can arise in within-country applications when research programs involve multi­ ple institutions or multiple sites within institutions. In order to minimize the potential for double counting, and in acknowledgement of the site-specific nature of many research results in which locally developed technologies locally dominate technologies developed elsewhere, Wood and Pardey (1993) suggest assuming no within-country spillovers between those locales where a national research program is simultaneously developing new tech­ nologies. Of course, when there is information to the contrary (such as where one site specializes in plant breeding and another in agronomy), the potential for spillovers should not be ruled out. Two options are (a) to conduct a spatially disaggregated am~lysis within a country, which may involve an explicit treatment of within-country spill­ overs or (b) to pre aggregate zones into the market aggregates to be used to evaluate the supply-shifting effects of research, obviating the need to mea­ sure spillovers. The latter approach runs a risk of aggregation bias but reduces the cost of information gathering in the process.59 58. Similar spatial aggregation techniques can be used to improve the estimates of the regional adoption parameters. In this case the basic area of analysis is defmed in tenns of uniformity with regard to adoption potential (in contrast to the agroecological zones used to partition regions into areas that have uniform supply-shifting potential). 59. The horizontal shift in an aggregate supply function (representing the sum of a number of competitive supply functions) is equal to the sum of the shifts of the individual curves in the quantity direction. This implies an approach to aggregation and to choosing weights for the aggregation - i.e" according to regional shares of preresearch output. Economic Surplus Measurement and Application 349 In the above model of spillovers in an international setting, the time profile of adoption of all the research results emanating from one region is com­ monly restricted to being equal among all regions. A more flexible approach would use separate parameters to measure the maximum spillover and the time profile of spillovers. As we discussed above, suitable information to support such disaggregation is unlikely to be available for any commodity research programs in an international setting, but it may be available for some commodities in a disaggregated domestic setting. The most sophisti­ cated analysis would disaggregate regionally within a country and measure the own-region effects and spillover effects in each region, properly allowing for different time profiles of adoption of own-research results and spill-in research results among the regions and allowing these parameters to vary among research programs. But the information requirements for such highly disaggregated studies are great, and simplifying assumptions are inevitable. It is incumbent on the analyst to make sensible simplifications and to test the sensitivity of the results to changes in the assumptions. Bias in subjective data: When information for calculating Ks for pro­ spective research is elicited from scientists, it is necessary to guard against unrealistic responses that would invalidate the analysis and the use of the results for evaluating research or setting priorities. A particular risk of bias arises because scientists know the purpose of the analysis and often have (or perceive) a vested interest in a high measured rate of return to research. Three ways of dealing with this potential bias are (a) using objective data on past experimental results, historical yield trends, and perhaps, total factor produc­ tivity growth rates to calibrate the elicitations, (b) creating an environment of peer review and, perhaps, a competitive process (e.g., Delphi methods) that will reduce the potential for personal incentives to bias estimates of technical parameters, and (c) creating an institutional setting in which scien­ tists will be held accountable for systematic biases in their estimates of research impact (e.g., by comparing actual achievements within programs over time against scientists' forecasts of what would be achieved). 5.3.3 Research Risk and Lags in Research, Development, and Adoption The temporal nature of the knowledge production function was described in chapter 3. It takes time to complete research, adoption takes time and is incomplete, and most research knowledge eventually depreciates. Thus, as described above, there are long lags in the process of research, development, and adoption. Pardey and Craig (1989) estimated that the effects of research on aggregate U.S. agricultural productivity persist for at least 30 years after 350 Economic Surplus Measurement and Application the research is begun. For research benefit calculations, the shape of the lag distribution in the earlier years is relatively important, and this shape, as well as the lag length, varies among research programs and among technological options within programs. In short, the time path of the flows of research benefits and costs is dynamic and uncertain - as discussed in section 2.1.3. Some of this uncertainty stems from the fact that the results from investing in research are inherently unpredictable, some from the fact that the industry response to the information is uncertain (so that K is uncertain), and some from uncertainty about technological and market parameters (so that translation of a known K into measures of benefits is uncertain). A further factor complicating the translation of a "known" K into benefits is that the underlying supply-and­ demand functions may involve dynamics (uncertainty and expectations) and leads and lags in relation to price responses as well as in relation to technical change. These dynamics mean that the elasticities, which we treat as constant parameters in economic surplus formulas, vary with length of run. In most studies of research benefits, these dynamics are put aside and the problem is treated as a comparative-statics exercise. The uncertainty is "managed" by conducting the analysis in terms of expected or, more pragmatically, most likely values and by carrying out some sensitivity analysis. Dynamics continue to be involved through variations in the size of the research-induced supply shift over time. A number of approaches may be used. In the typical approach for ex ante research evaluation, the first step is to estimate the proportional cost reduction, JCfAX, that would apply with successful research and full adoption of the resulting technology by the entire industry. That value is multiplied by the probability of success (treating "success" as an all-or-nothing outcome, rather than allowing a continuous range of degrees of success with corresponding probabilities occurrence) and by the likely rate of adoption. Then it is adjusted for any anticipated research depreciation to yield an annual value, K, , for inclusion in the research benefit formula for that year. An alternative shortcut approach is to calculate the flow of benefits, BMAX, corresponding to the maximum value, KMAX , and then to scale that flow of benefits according to the probability of success, adoption rate, and depreci­ ation rate. Gross annual benefits, B, are a quadratic function of the supply shift, K, but for small values of K, the quadratic term vanishes and the function is approximately linear. Thus, these two approaches are approxi­ mately equivalent for small values of K so long as the same formula (i.e., with the same parameters) is used to translate K, into a measure of B, for all values of t. However, in some cases, it may be desirable to use different parameters for different future time periods (reflecting, for instance, the effects of income and population growth on the underlying supply and demand or allowing different elasticities for longer run lengths). In such Economic Surplus Measurement and Application 351 cases, the two approaches will not be equivalent and the approximation may not be a good one. Hence, when dynamics are incorporated quantitatively in these models, there are three related components to estimate: • the research lag • the adoption or uptake phase • the depreciation or obsolescence phase In this section we deal with each of these components in turn. For ex ante evaluations, we also have to deal with the related question of the probability of research being successful. Research Risk6l.1 In ex post studies, we know what research was successful and what was not, at least from the point of view of meeting a scientific objecti ve. And we can find out whether the research led to a commercially successful new idea, method, technology, or input that was adopted by farmers or others. In ex ante evaluation, it is not known in advance whether research will be success­ ful in either the scientific or commercial sense. A measure of the odds of success will be required for each program alternative being considered. Success, however measured, will depend on the degree of aggregation within programs: highly disaggregated programs or individual projects might be highly risky; highly aggregated programs are more predictable if their outcomes are either uncorrelated or negatively correlated and if, as a result, uncertainty is reduced by pooling. While it is typical, and often convenient, to view success in absolute terms (i.e., the achievement of a particular result), a given program of research might be judged successful across a range of outcomes. For instance, it may be useful for some purposes to think of research as being successful if it generates a Z% increase in experimental yields. However, the same research would surely be a success if it led to a yield increase greater than Z%. Many (especially biological) research projects admit a continuous range of possible outcomes, and the outcome of research can be viewed in terms of the statistical distribution of the random variable used to indicate it (e.g., the experimental yield or gain in yield). In such cases the use of a discrete analogue to represent success or failure is an approximation for analytical convenience, and it might not be very convenient for the analysis. Consider a plant breeding program, for example, for which the outcome of the research is measured by the increase in experimental yields relative to 61.1. See Fishel (1970) for an early discussion of research risk issues related to research evaluation, planning, and resource allocation and Anderson (1991) for a more recent discussion of these same issues. 352 Economic Surplus Measurement and Application existing varieties (z = ~Y). 6 1 In figure 5.2, the distribution of experimental yield gains, as perceived before the research is undertaken, is represented by f(z), and z* is the minimum experimental yield gain that will lead to a commercially successful new variety. Research success is defined in the absolute sense as z ~ z·. The probability of research success is defined as Figure 5.2: Presumed probability distribution of experimental yield gains Probability fez) o z* Experimental yieldgain,Z Pr (z ~ z* ) = f f(z)dz (5.13) • Z The expected value of the experimental yield gain, given successful research, is ~(z I z ~ z* ) = f z f(z)dz (5.14) z. The expected yield gain attributable to research is often calculated by multiplying the expected yield gain, given successful research, i.e., ~( z I z ~ z'), by the probability that the research will be successful, i.e., Pr(z ~ z*): ~y 1\* = -z* = Pr ( z ~ z* ) x ~ ( z I z ~ z* ) = f f(z)dz f z f(z)dz (5.15a) z z• 61. The arguments could apply as well to conceptualizing a distribution of changes in industI)' yields due to reSearch. When an experimental yield distribution has been defmed, as discussed below, it is necessary to adjust for differences between experimental gains and commercial gains. Scientists are typically better able to judge experimental outcomes than outcomes in the field. Economic Surplus Measurement and Application 353 Alternatively, one could directly estimate the expected yield gain attribut­ able to the research by integrating the distribution of possible research outcomes over the entire range: ~y =z = s(z) = Jz f(z)dz (5.15b) This is the unconditional expected value of the yield gain from research. Either of these two alternatives (equation 5.15a or 5.15b) could be involved implicitly or explicitly when scientists are being asked to quantify the uncertain outcome from research. In many studies, the equivalent of equation 5.15a is used with estimates of the "probability of success" and the "conditional expectation of (experi­ mental) yield gains given successful research" that have been solicited directly. One drawback in using this approach is that the statistical meanings of the terms "probability" and "conditional expectation" might not be fully appreciated by the people providing the information. The meaning of "suc­ cess" has not always been clear, either.62 As a consequence, the validity of the measures of K may be questionable. An alternative approach is to solicit information on the distribution of experimental outcomes, instead - i.e., f(z) - and a definition of successful outcome - i.e., z· - and use that information to deduce a measure of the conditional (equation 5.15a) or unconditional (equation 5.15b) expected yield gain due to research. For instance, scientists could be asked to estimate a higher-bound (or maximum possible) experimental yield gain, Zh, a lower­ bound (or minimum possible) yield gain, z/ (which could be a negative number), and a most-likely value, Zm' Then, assuming a triangular distribu­ tion, as in figure 5.3, the complete distribution of experimental outcomes is defined.63 Alternatively, the scientists could be asked to estimate points on the cumulative distribution function (i.e., probabilites of experimental yields greater than various values - Pr[z ~ z;]). The results of this elicitation could be used either to define the entire distribution or as a direct estimate of the critical value associated with the definition of "success": Pr(z ~ z* ). 62 For instance, some scientists have interpreted "success" to mean meeting the stated objectives of a program - such as successfully completing experimental trials - which might not have any economic implications. Others might have a notion ofz * in mind that is well beyond what is necessary for commercial success. Unless success is dermed explicitly, and meaningfully, it is difficult to assess the validity or meaning of the information elcited from scientists and others. 63. In section 5.4.4 we show how to parameterize a triangular probability distribution and use it to calculate measures of the dispersion of the distribution to include in models accounting for research risk. 354 Economic Surplus Measurement and Application Figure 5.3: Triangular probability distribution of experimental yield gains Probability f(z) Zh Experimental yield gain. Z Knowledge of the experimental yield distribution could be used to gener­ ate either the summary statistics of the research outcome defined above or other summary statistics (such as a variance). Two drawbacks may be encountered when using this type of approach. First, the triangular distribu­ tion is a restrictive approximation, which, in some cases, may involve a loss of information compared with what is actually known about the potential research outcome - but the same approach may be used with more flexible distributions if necessary. Second, even using the simple triangular distribu­ tion, the information requirements are significant when only one level of research funding is being evaluated. Later we advocate varying the size of the research budget for each program in order to evaluate the shape of the research payoff relationship. It might be unrealistic, or simply too expensive, to attempt to parameterize an experimental yield distribution across a range of program budgets for each program. In what follows, we have adopted the conventional approach of combin­ ing an estimate of the probability of research success with an estimate of the expected yield gain (or cost saving), given successful research. This encom­ passes approaches that elicit the estimates directly, as well as those that use the preferred approach (when resources permit) of eliciting information on the distribution of outcomes from which the probability of success and expected values can be derived.64 Experienced research scientists are likely to be the best source of infor­ mation about the distributions of possible outcomes from alternative re- 64. Eliciting explicit details on the distribution of experimental outcomes makes possible the consid­ eration oft he joint distribution of the research outcome and the adoption of research results. In sophisticated studies it might be of interest to take this into account when the expected benefits are computed. It is likely that some type of Monte Carlo approach would be necessary to do this properly. Economic Surplus Measurement and Application 355 search programs, their expected outcomes, and their probabilities of research success. These scientists can be asked to provide the required information for each research program alternative and for a given program budget and time schedule to complete the research, taking into account the fixed re­ sources available for the research, previous research, information available in other countries, and so on. Time Required to Complete the Research Some types of research inherently require more time to complete (or to obtain partial results) than others. The sooner benefits are received, the more they are worth. Scientists should, therefore, be asked to estimate the length of time between incurring an expenditure on research and the release of new technologies. This is easier to do for disaggregated projects (e.g., wheat variety devel­ opment) than for an aggregative program (e.g., wheat research including a range of crop management projects and programs in addition to new variety work). The problem is that different components of an aggregated program are characterized by different research lags. What is required for ex ante evaluation of programs is a meaningful estimate for a profile of an "average" research lag for each program. Similar to eliciting K for an aggregate program, it might be best to ask scientists about the disaggregated compo­ nents along with the aggregate. Insights into the nature of the research production function (as well as the roles of diminishing returns and fixed factors in research) may be gleaned by asking scientists about the implications of changing the time taken to com­ plete a gi ven program of research. Often, scientists will have a clearer picture of the implications of speeding up or slowing down a particular line of work than the implications of changing the resources available for the work. For instance, they could be asked to consider the effects of finishing an ongoing project (say, with a five-year horizon) one year earlier than originally planned. Given the same annual budget, the same facilities and support staff, and the same expected result if the research is successful (i.e., KMAX ), what would be the effect on the probability of success of shortening the horizon? Or, holding the probability of success constant, what quantity of extra resources would be needed to complete the research one year earlier? This type of approach was used, for instance, by Scobie and Jacobsen (1992). Extent ofA doption The geographical spread of research results and the time path of adoption, both domestically and internationally, are important determinants of total 356 Economic Surplus Measurement and Application benefits of research and its distribution. These adoption rates depend on many geoclimatic and socioeconomic factors. 65 Because these factors differ by region and their importance differs by commodity, adoption will seldom approach 100% in a country as a whole, even if it is 100% within a village or local area. Ex post adoption studies: Ex post sudies of the adoption of research results have included (a) econometric studies where an adoption process is estimated jointly with other characteristics of the supply response of a particular com­ modity (or commodity aggregate) to aggregate research and extension and (b) studies using survey data on the uptake of specific new technologies (CIMMYT 1993). Lindner (1981), Feder, Just and Zilberman (1985), and Feder and Umali (1993) surveyed much of the literature in this area. The econometric approach predominates. It is of questionable use for extrapolating to an ex ante analysis of individual commodity research pro­ grams because (a) extrapolating from the past may be suspect,66 (b) aggregate (i.e., sectorwide) responses are not informative about dis aggregated (i.e., commodity-specific) responses, and (c) key aspects of the lag structure underlying the adoption curve are typically imposed rather than estimated in the analysis. The second approach, based on a survey (either formal or informal) of the past adoption of new technologies, has a better chance of yielding results that are relevant for ex ante evaluation of research programs but only as a basis for a benchmark. In addition, published data are some­ times available on the spread of varieties of a crop. These data may be used for ex post evaluation but may also provide a benchmark for forward-looking analyses of similar innovations. Ex ante models: For ex ante evaluation work, we have suggested treating the adoption process as a modifier that translates potential research effects, KMAX , into actual effects over time, K,. As discussed above, different compo­ nents of a commodity research program will generate technologies that have different adoption paths in different locations. Potential complementarities and substitution effects among components of technologies further compli­ cate the determination of likely adoption paths for individual components. Extension workers and others who have observed the adoption of previous research results are primary candidates from whom to elicit information on adoption. A useful approach for eliciting a synthetic adoption path applicable 65. The agroecological factors include rainfall, temperature, soils, topography, and the photoperiodic­ ity of crops; the socioeconomic ones include land tenure, farmer education, quantity and quality of extension, transportation, availability of credit, communications, market structure, religion, cultural differences in preferences, and incentives created or destroyed by output pricing and input subsidy policies. 66. It may be thought that the future will be like the past because, in the past, the future was like the past (Weinberg 1975). Economic Surplus Measurement and Application 357 to a broad program of research is to discuss the general set of questions and interdependencies and then infer adoption responses for the aggregate. This is similar to the approach for eliciting [(!'fAX and information about research lags, where we suggested considering individual components explicitly; there are complementary aspects of collecting the various types of informa­ tion on these disaggregated components together. To focus the discussion, it is often useful, at least for crops, to begin with research on new varieties, recognizing that the agronomic and pest and disease research results are likely to be adopted in concert with new varieties.67 To define the entire adoption curve corresponding to a particular technol­ ogy (or results from a particular program of research), it is necessary to choose a functional form. A linear form for adoption response is widely used as a component of a trapezoidal lag structure. Appendix A5.1.2 describes in detail how to specify a trapezoidal research lag structure, including a linear adoption phase (following an initial research lag) and a linear decline. The main alternative is an S-shaped (usually logistic) curve that involves a similar number of parameters. The logistic curve can be specified as68 (5.16) where AMAX is the maximum adoption rate (commonly expressed as a fraction of the total area ultimately planted to a crop), At is the actual adoption rate t years after the release of the new technology, and a. and 13 are parameters that define the path of the adoption rate that asymptotically approaches the maximum. Thus, a logistic adoption curve can be defined completely by three parameters: AMAX , a., and 13. The entire curve can be generated, also, by defining any three points on the curve (preferably with two near-extreme values). A variety of approaches has been used to elicit values that will define the parameters of a logistic adoption curve. A wise choice would be dictated by judgments about which points on the curve are easier to guess. It is usually reasonable to assume very low adoption in the year of release (say, Au = 0.01, or 1% ) to define one point on the curve. It is also reasonable to try to elicit an estimate of the ceiling rate of adoption. One more point is needed. The scientists and extension workers could be asked either (a) to estimate the most likely adoption rate in a particular year, say seven years after release of 67. Some studies have considered piecemeal adoption of components of a technological package (e.g., Byerlee and de Palanco 1986), but these studies have usually concerned ex post evaluation of relatively disaggregated research programs. In ex ante evaluation of aggregate research programs, it is usually difficult to predict such details. 68. For useful discussions of the use of the logistic adoption curve in this context and its derivation, see Griliches (1957b) and Lekvall and Wahlbin (1973). 358 Economic Surplus Measurement and Application the technology or (b) to give a best estimate of the number of years required after release of the technology before adoption reaches 50% of the maximum (e.g., if it takes 11 years, All = 0.5AMAx). Using such information, it is easy to parameterize the curve as follows. Taking logarithms of equation 5.16 yields an equation for ~ as a function of a, AMAx, At, and t: (5.17) We know AMAX and two combinations of At and t. Substituting those values into equation 5.17, we can solve for values of a and ~. 69 Research Depreciation Many types of research results depreciate over time for various reasons. Crop varieties become susceptible to insects and diseases and eventually begin to yield less. Particular pesticide practices may become less effective as insects and diseases develop resistance to pesticides, and so on. Economic deprecia­ tion may not be the same as physical depreciation. Even if a technology doesn't physically depreciate, it can become obsolete and be superseded as a result of changes in market conditions or other changes in technology?' Consequently, estimates of research depreciation should be made to adjust expected research impact downward a few years after use of the new technology begins. The effects of discounting mean that early years are weighted much more heavily than later years so that an accurate estimation of research depreciation is less important than an accurate estimation of the time needed to complete the research and of the adoption rates, but it can still be important. For ex ante analysis, scientists and extension workers can be asked to estimate how rapidly they expect the results of the proposed research to degenerate. For ex post analysis, survey results may be available on when the use of particular varieties or other technologies slowed or stopped. 69. Scobie and Jacobsen (1992) show how to do this in a particular setting. See also Pardey (1978), who describes the Pearl-Reed method of fitting logistic culVes as given in Pearl (1924, pp. 576-81) and Davis (1941, pp. 216-8). 70. The relevant notion of economic depreciation oftechnology depends on the particular technolog­ ical alternatives being considered - that is, it depends on the ceteris paribus conditions. The effects of a one-shot research investment in a particular year are defmed given fixed values for the investments in all other years. When considering the effects of a program of research spending over several years, the baseline simulation will refer to a baseline of no investment in the several years in question in the particular program. An extreme example is when the program is regarded as pennanent - applying over the indefinite future. The relevant notion of research depreciation will differ between these two types of investments (i.e., one-shot versus multiyear or permanent). Economic Surplus Measurement and Application 359 As for other forms of capital, the economic depreciation of the output from successful research (be it new machinery, new genetic material, or new ideas) is commonly represented in one of three ways: • "one-hoss-shay" depreciation, in which the technology holds its value (i.e., K, = KMAX) until a point in time, LR , when it becomes worthless instantly (i.e., KLn +r = 0, r ~ 0) • "straight-line" depreciation, in which, after some point, the value of the technology declines linearly to zero (as described in detail in appendix AS. 1.2) • "declining-balance" depreciation, in which, at some point, the value of the technology begins to decline by a constant proportion each period - i.e., Tyears after release of the technology it begins to depreciate according to KT+n = (1 - o)nKMAX Variations on each of these approaches to delineating the profile of the agricultural research lag can include specifying the depreciation as applying to the gross measure of K (avoiding the problem of modeling the disadoption process explicitly) or as applying to the technology itself (with endogenous disadoption decisions). In the latter case, for example, it would be necessary to estimate both the decline in yields and the share of production affected. The former approach is more common and is illustrated in appendix AS.1.2 for the straight-line depreciation approach. Questionnaires For ex ante analysis, interview questionnaires can be used to obtain information from scientists, extension workers, and research directors on expected yield increases or per unit cost reductions, probabilities of research success, time to complete the research, adoption rates, and research depreci­ ation. It is impossible to define a questionnaire or elicitation process that will be generally applicable. In appendix AS.4 we provide some illustrative examples. Combining the Information Box S.l shows how to deduce a time path for the research-induced supply shift for a program of research using elicited information on the expected yield gain, adoption rates, probability of success, additional input use, and research depreciation. 360 Economic Surplus Measurement and Application Box 5.1: Combining Information to Calculate k and K Suppose the following information has been collected for a research program on commodity j: (a) The research has a probability pj of successfully leading to a new technology that when fully adopted, will result in a 100 E(Yj)% increase in commercial yields (after allowing for the optimization of the input mix when switching from existing technology to new technology and allowing for differences between changes in commercial yields and changes in experimental yields) - e.g., E( Yj) = 0.30 so that 100 E(Yj) = 30%. (b) The fraction of the total industry (area or output) adopting the new technology is defined in relation to years, t, from commencement of the research as Aj,t (where the particular values might be derived from a logistic curve or some other model) and there is a declining-balance rate of depreciation in the new technology, OJ. (Here we treat the depreciation rate as applying from the point at which the project commences, but in many cases, it will be preferable to defer the commencement of depreciation until later, say, T years after maximum adoption is achieved.) (c) The supply elasticity is Ej and the current producer price is PPj,O per tonne. (d) The increase in commercial yields involves an additional cost of purchased inputs (e.g., fertilizer, fuel, or pesticides) of !lCj (or 100 E[Cj]%) in costs per hectare that can be translated using preresearch yields, Yj,O, to a change in cost per tonne of output of !lCjI{[l+E(Yj)] Yj,o}. This could be a positive or negative number. The percentage change in costs per tonne - obtained by dividing by average costs per tonne (C/Yj,o) - is equal to E(Cj)/[I+E(Yj)]. (e) The increase in commercial yields involves a 100 E(Fj)% increase in the use of allocatable fixed factors (e.g., land or operator labor and managerial inputs) per tonne of output. And quasi-rents to allocatable fixed factors account for a fraction, Sj, of preresearch costs per tonne. E(Fj), too, could be a positive or negative number. Given this elicited information on potential yield changes, adoption rates, and so on, values for the absolute reduction in costs per tonne, kj,t, for all future years can be projected as follows. First, assuming the use of variable or quasi-fixed inputs does not change in order to bring forth the projected yield increase ! 0 (AS.3b) 388 Economic Surplus Measurement and Application where 1tf = the growth rate of demand (e.g., population growth rate + income elasticity x income growth rate) 1tf = is the growth rate of supply (e.g., area growth rate + yield growth rate not attributable to research) Now we have sufficient information to parameterize the supply-and-de­ mand equations for each region in each year under the no-research scenario. Research-Induced Supply Shifts Local effect of research: Let region i undertake a program of research with • probability of success Pi' which, if the research is successful and the results are fully adopted, will yield • a cost saving per unit of output equal to ci percent of the initial price, PPi•O in region i, while • a ceiling adoption rate of A~AX percent holds in region i Then it is anticipated that the supply function in region i will shift down (in the price direction), eventually, by an amount per unit equal to Jti.fA-X P- i c AMAX j i PPi ,O>- 0 (A5.4) The actual supply shift in any particular year is some fraction of the eventual maximum supply shift, ! AR + AA + AM + AD) Figure AS.4.2 in appendix AS.4 shows the trapezoidal adoption curve and shows how the parameters above (AR• AA' AM' and AD) may be used to define the entire curve. Options for deriving S-shaped adoption curves are dis­ cussed in section S.3.3 and appendix AS.4. Spillover etTects of research: 84 The spillovereJtects from region i to other regions,}, are parameterized in relation to the supply shifts in region i, implicitly assuming the same adoption curve applies in every region. kj ., = 9ji ki,t for all i and } (AS.S) where 9ji = supply shift in} due to research-induced supply shift in i (9ii = 1) With-Research Supply and Demand To model the with-research case (denoted by superscript R on all relevant variables and parameters), we take the intercepts from the without-research case (but include the effects of exogenous supply growth), add the effect of the supply shift to them, and include the result in the supply equation: (AS.6) The models for supply and demand that reflect the local and spillover effects of research are Qf" = a~ + /3i PPf, (AS.7a) (AS.7b) The only substantive difference from the corresponding without-research equations (AS.la and AS.1 b) is in the supply intercept, but as noted above, the prices and quantities are labeled differently (the R superscript) to distin­ guish them from the without-research values: • quantity consumed in each region - C~ • quantity produced in each region - Of, 84. 1lte spillover coefficients. Sji. are defined as if they were constant for all types of research-induced supply shifts and. for a given technological change, constant over time, implying that the relative shifts always occur in fixed proportion. 1ltese might not be reasonable restrictions for all problems. Between agroecological zones i and}, the spillover relationships are very likely to differ among commodities, among types of technological changes for a given commodity, and over time for a given technological change and a given comroodity. For some problem.<;, it might be necessary to redefine the spillover matrix for different types of technologies and different times after release of a technology. 390 Economic Surplus Measurement and Application • producer price in each region - PPfl • consumer price in each region - PCfl Market-Clearing Rules For all of the scenarios to be considered, there is an overall quantity clearing rule to the effect that the sum of quantities supplied equals the sum of quantities demanded in each year. Considering n regions, QI= (QI.I+ Q2.1 + ... + Qn,l) = CI = (CI,I + C2,1 + ... + Cn,l) (AS.S) All of the market-clearing rules express policies in terms of price wedges that permit differences between consumer and producer prices within and among regions consistent with clearing quantities produced and consumed.85 Free trade: The easiest case is that of free trade, where • with-research prices: ppR1,1 = PCI1!,1 = PCR = ppR },1 },I = pRI • without-research prices: pp;., = PC;.I =P Cj,1 = PPj.t =P , are defined for all regions i and j and for any year t. Making this substitution into each of the n regional supply-and-demand equations and then substituting them into equation AS.8 yields a solution for the equilibrium price for each year. To simplify, let us define the following aggregated parameters for each year, t: • YI = Yit + Y21 + ... + Ynl • al = all + a2t + ... + anI • a: =a fl + a~1 + ... + a~1 • 01= 0 = Ow + 020 + ... + 0nO < 0 • PI = P = PIO + P20 + ... + PnO > 0 Then the without-research and the with-research market-clearing prices under free trade are given by PI = (YI - al)l(p - 0) (AS.9a) P: = (YI - a:)I(p - 0) (AS.9b) These are always positive numbers, with PI> P:, because the intercepts 85. Transportation costs influence trade among countries and should theoretically be incorporated into the analysis if possible. However, accurate calculation of these costs is often difficult because it requires knowing the transportation differentials for each commodity between the home country being studied and each of its rnajor trading partners, as welI as the pattern of commodity flows. If international research spillovers are included in the analysis, either (a) information should be colIected on the likely destinations and transportation costs between the horne country and each other country involved so that a relatively accurate assessment can be made of the price wedge to be driven between the excess-supply and excess-demand curves or (b) transportation costs should be ignored. When regional analyses within a country are being conducted, regional transportation costs also may be needed. Economic Surplus Measurement and Application 391 oh the quantity axis satisfy 'Yt > a~ > at - unless we make a mistake such as letting supply grow too fast relative to demand. 86 We can substitute the results for prices from equations A5.9a and A5.9b into the regional supply-and-demand equations to compute regional quanti­ ties produced and consumed with and without research and, as we shall see later, then calculate the regional consumer and producer welfare effects. Generalized taxes and subsidies: We can define a general solution for a large variety of tax or subsidy regimes by setting out a general model in which a per unit tax is collected from consumers in every region and from producers in every region. · If = per unit consumer tax in region i • Tf = per unit producer tax in region i Different policies can be represented as different combinations of taxes and subsidies • consumption tax in region i at T; per unit: If=T;; 1¥=O • production tax in region i at T; per unit: If=O; 1¥=T; • export tax in region i at T; per unit: If = -T;; 1¥ = T; • import tariff in region i at T; per unit: If = T; ; 1¥ = -T; A subsidy is a negative tax, so it is also possible to use these to represent subsidies on output, consumption, imports, or exports. One way to think about this is to imagine a region with no taxes or subsidies in which the prices to producers and consumers are Pt = PCt = PPt and P~ = PC~ = PPf. Thus, P, (expressed in common currency units, either local currency or $US) is the border price for an exporter or an importer whose internal consumer or producer prices will be equal to that price in the absence of any domestic distortions. The arbitrage rules are that the prices in all regions are equal to • PP;,t == Pt - Tf • PC;,t == Pt + Tf • ppR ==pR_ T9 J,t t t • PCR =pR+ yc l,t t I for all regions i andj and for any year t. Making this substitution into each of the n regional supply-and-demand equations and substituting them into equation A5.9 yields a solution for the equilibrium price for each year. As for the case offree trade, let us define the following aggregated parameters for each year: 86. For instance, we could violate this condition by setting the autonomous growth rate of supply so much greater than the autonorrwus growth rate of demand that in some set of future projections a point would be reached where supply and demand did not cross in the positive orthant - i.e., the quantity intercept of supply would become greater than the quantity intercept of demand in either the with- or without-research case. 392 Economic Surplus Measurement and Application • Y, = YI, + Y2t + ... + Yn, • a, = a lt + a2, + ... + ant • a: =°a f, + ~, + ... + a! • 0, = = 0 10 + 020 + ... + 0nO < 0 • ~, = ~ = ~]() + ~20 + ... + ~nO > 0 In addition, we can define the following aggregated demand-and-supply shifts in the quantity direction because of consumer and producer taxes: • Tf = 7ft 010 + ~ 020 + ... + ~ 0nO • Tfl = 1Vr ~]() + 1'£ ~20 + ... + T£ ~nO P,= (y,+ 'If + Tf - a, )/(~ - 0) (AS. lOa) P: = ( Y, + 'If + Tf - a: )/( ~ -°) (AS. lOb) To check the signs intuitively, taxes on production in all regions will raise the equilibrium world trading price P, - which is equal to the producer price in any country or region with no producer taxes and the consumer price in any country or region with no consumer taxes. Taxes on consumption in all regions will lower it.87 Of course this hypothetical price, P" might not actually apply anywhere. To compute the actual consumer and producer prices in any region, the results of equations AS.lOa and AS.lOb are substituted into the arbitrage (market-clearing) rules given above (under the heading "generalized taxes and subsidies"). Then the individual prices can be used in the individual supply-and-demand equations (equations AS.I and AS.7) to compute quan­ tities with and without research, and from there to compute surplus effects. Notice that this set of results includes the free-trade model as a special case (i.e., when all ofthe taxes and subsidies are zero). The small-country case: The small-country case can be represented in this model without modification. However, to do that requires getting information - that is not useful otherwise - on quantities produced and consumed in the rest of the world. The alternative is to define the market-clearing price for equations AS.IOa and AS. lOb as an exogenous parameter: P,=~=P, It might require defining a growth rate for P, to obtain a series of exogenous world prices based on a starting value for PO" Then corresponding quantities 87. A positive value of 19 for all i leads to a positive value of Tf1 and thus a higher value of P,. A positive value of ~ for all i leads to a negative value of r; (because the demand slopes 0;.0 are negative numbers) and a lowering of the price, P" Economic Surplus Measurement and Application 393 can be obtained by substituting into the relevant supply-and-demand equations. Other policies: Quantitative restrictions on production or trade can be treated approximately as tax/subsidy equivalents with a little care to distribute "tax revenue" as quota rents. The approximation is somewhat unreliable in a dynamic model, but it might suffice for our purposes. A target price, defi­ ciency-payment scheme might involve more work. Conceptually, the approach is to define target price and allow it to determine output in regions where it applies. Then, with that supply as exogenous, supply equations in the other regions and demand equations in all regions would interact to determine price. Welfare Effects The following equations for welfare effects should be correct for most (if not all) types of policies (i.e., market-clearing rules). fl.PSj,I = (kj,l + PP'J.t - PPj.t) [Qj" + O.S (Qf,t - Qj,t)] (AS. 11 a) fl.CSj,t = (PCj" - PCf,,) [Cj,t + 0.5 (Cf,,- Cj,t)] (AS. 11 b) fl.GSj,l = Tji (Cf" - Cj,l ) + 1J! ( Qf.t - Qj" ) (AS. 11 c) where fl.PSj,/ = producer research benefit in region) in year t fl.CSj,t = consumer research benefit in region) in year t fl.GSj,/ = government research benefit in region) in year t Aggregation over Time and Interest Groups The model so far is capable of generating an indefinitely long time series of prices, quantities, and economic surplus measures for the regions of interest for a range of tax or subsidy policies. The remaining problem is to aggregate those measures into summary measures of research benefits. For a given policy scenario, we have the measures of benefits - MSi,/, fl.CSi,/, fl.GS;,/ - for each region in each time period. The real discount rate must be defined for the computation of the present value ofthe stream of benefits. A reasonable approach is to fix a single value for all regions, interest groups, and years so that We need to define a relevant planning horizon. Thirty years should be adequate for most purposes if we are using discount rates of S % per year or greater. The present values of benefits to interest groups are then defined as 394 Economic Surplus Measurement and Application VPS; = l:~ t..PS;/(l + r Y =t ..PS;.o + t..PS;/(l + r ) (AS.12a) + t..PS;/(l + r)2 + ... + t..PS;,301(1 + r)30 VCS. =l :30 t..CS. 1(1 + r)' I r=() I.' = t..CS;.o + t..CS;/(l + r) (AS.12b) + t..CS;/(l + r? + ... + t..CS;.30/ (l + r)30 VGS; = l:~ t..GS;/(l + rY =t ..GSi.O + t..GS;/(1 + r) (AS.12c) + t..GS;/(l + r? + ... + t..GS;.30/ (l + r)30 If a longer planning horizon is appropriate, we can either simply increase the number of years from 30 (probably the best way) or approximate the effects beyond 30 years using an infinite series (which is risky when the markets are growing and research effects are depreciating). Then we are free to add these present values up across the different producing and consuming groups in whatever fashion we find useful. Economic Surplus Measurement and Application 395 Appendix A5.2: Selected Formulas for Calculating Research Benefits The Dream© model is relatively general, and for some situations that generality may be a disadvantage. Sometimes it is considered unnecessary, for instance, to go to the trouble of allowing for exogenous growth in supply and demand or to do a full capital-budgeting analysis. Some studies implicitly assume that the time structure of costs and benefits is constant across research alternatives, and therefore they compare the gross annual research benefits after full adoption. These simplifying assumptions, whether made implicitly or explicitly, are very restrictive. It is therefore questionable whether the results obtained under such assumptions would be valid generally. However, in some situations such assumptions will be made either because they are believed to be an appropriate approximation or because the analysis is necessarily a shortcut one. In order to facilitate the analysis in those cases, we provide some formulas for measures of the total gross annual research benefit (GARB) and its distribution in some typical market-structure scenarios. All of these shortcut formulas may be obtained as special cases of those in the model above; that is, they are based on an assumption of approximately linear supply and demand with a parallel research-induced supply shift in the home country (or region). A5.2.1 Simplified, Two-Country Model Dropping redundant time subscripts (i.e., to consider only a one-shot, comparative static analysis) and considering only two countries or regions, the variables in the model are, for regions) = A (the home or innovating country or region) or B (the rest of the world or the "other" noninnovating country or region) • producer price in regionj - PPj • consumer price in regionj - PCj • production in region} - Qj • consumption in region} - Cj • production tax in region} - 'If > 0 ('If < 0 for production subsidy) • consumption tax in region) -TJ > 0 (TJ < 0 for consumption subsidy) • producer research benefit in regionj - PSj • consumer research benefit in region j - CSj • government research benefit in regionj - GSj • national research benefit in region) - NSj Research-lnduced Supply Shifts As a special case of equation A5.4, we assume that the supply function in the home country (region A) will shift down (in the price direction) by an 396 Economic Surplus Measurement and Application amount per unit equal to k == kA == K PP A (AS.I3) where K is the percentage shift down of supply relative to the initial producer price, PPA• Research-Induced Changes in Prices and Quantities In order to measure the welfare impact, what remains to be defined are the changes in prices (I1PPj and I1PCj for j =A and B) and quantities (I1Qj and I1Cj forj =A and B) due to the research-induced supply shift in terms of the magnitude ofthe supply shift, kA' the parameters of supply and demand, and the policy wedges. The policy wedges may be represented (using the per unit consumption and production taxes, ]C and 7'2, respectively, and the hypo­ thetical price, P, as before) as PPA = P- TfJ..; PCA =P+ 'ri.; PPB == P-Ii; PCB =P+ ~ (AS.I4) Then, we may write the supply-and-demand equations, including the tax wedges and the supply-shift parameter, to reflect research-induced technical change, as QA == a A + ~A (P - TfJ.. + k) (AS.ISa) CA =" fA + 0A (P+ 'ri.) (AS.ISb) QB =' a B + ~B (P -Ii ) (AS.ISc) CB == "fB + 0B (P + ~ ) (AS.ISd) Market clearing in this system of four equations is enforced by setting aggregate supply equal to aggregate demand (i.e., QA + QB == CA + CB)' The without-research case is obtained by setting k equal to zero in equation AS.ISa. The general solution for the research-induced change in price is given by (AS.I6) It is an interesting and useful feature of the linear and additive structure of this model, with fixed per unit taxes/subsidies, that the research-induced changes in all producer and consumer prices are equal. That is, (AS.l7) Thus, the research-induced changes in quantities are given by I1QA == ~A (k + I1P) ; I1CA == 0AI1P ; I1QB = ~BI1P; I1CB = 0BM (AS.I8) Economic Surplus Measurement and Application 397 Welfare Impact The following equations for welfare effects are special cases of equations A5 .11 a through A5 .11 c, where the welfare effects in the home country or region, (i.e., country A) are given by I:!PS A = (k + I:!P) (QA + O.5I:!QA) (A5.19a) I:!CSA = -M (CA + O.5I:!CA) (A5.19b) I:!GS A = ~ I:!C A + ~ I:!QA (A5.19c) /lNS A = MSA + I:!CS A + I:!GS A (A5.19d) and the welfare effects in the other country or region, or the rest of the world (i.e., denoted B) are given by I:!PSB =M (QB + O.5I:!QB) (A5.1ge) I:!CSB = -I:!P (CB + O.5I:!CB) (A5.19f) I:!GSB = 1ii I:!CB + ~I:!QB (A5.19g) /lNSB = I:!PSB + I:!CSB + I:!GSB (A5.l9h) In these equations, I:! denotes the difference between the with-research and without-research values of variables. Total regional (national) benefits in regionj are defined as /lNSj' and global benefits, I:!WS, may be obtained by adding up regional (national) benefits. Examples Small, open economy with no distortions: The simplest case of all is that of a small country in trade for which the export (or import) price is exoge­ nous and unaffected by the research-induced supply shift. In this case, taking the limit as I3B ~ 00, I:!P =0 so that welfare of the ROW in total, and welfare of domestic consumers, are unaffected. Also, in the absence of price-distort­ ing policies, taxpayer welfare is unaffected. In short, only domestic producer welfare is affected, and the measure of gross annual producer (and national and global) research benefits is MSA = /lNSA = I:!WS I:!PSA = k (QA + O.5I:!QA) = k (QA + O.5I3Ak) (A5.20a) If we choose to express the supply shift, k, as a fraction of the initial price, P A' i.e., let K = kiPA ' then we may express the result above in terms of the 398 Economic Surplus Measurement and Application percentage supply shift and the elasticity of supply, EA, as APSA = KP A QA (l + O.5KEA) (AS.20b) where the values of PA a nd QA refer to the preresearch equilibrium and the elasticity of supply has been evaluated at that point. Small, open economy with a border distortion subsidy or tax: In the case of a small country in trade with a border tax (or subsidy), the welfare of the ROW in total and of domestic consumers is still unaffected by domestic supply shifts. However, in this case, taxpayer welfare is affected. The measure of producer research benefits is, as before, (AS.21a) =K ' PA ' QA' (1 + O.SK' E/) where K' = k P / defines the supply shift as a proportion of the distorted preresearch price. Here, the values of P A' and QA' refer to the distorted preresearch equilibrium and the elasticity of supply, EA', has been evaluated at that point. In addition, there is the welfare impact on taxpayers to consider. An import tariff of T per unit (or an equivalent ad valorem tariff at a rate 't = T / P A) on an imported good is equivalent to an output subsidy of T per unit and a consumption tax at the same rate. Thus, for an imported good with a tariff, as a result ofthe research-induced increase in supply, taxpayers lose an amount equal to the per unit tax multiplied by the reduction in imports (i.e., the increase in output): (AS.21b) The same formula would apply for an export good with an export subsidy, while the sign would be reversed in the case of an import subsidy or an export tax. Net domestic benefits from the research-induced supply shift are obtained by adding effects on taxpayer welfare and producer welfare. Recall that with linear supply and demand and a fixed per unit tax or subsidy, in the small-coun­ try case, national research benefits are unaffected by the presence of the distortion. That is, producer research benefits are higher under a tariff by an amount exactly equal to the research-induced reduction in tariff revenues. This result (shown by Alston, Edwards and Freebaim 1988, and generalized by Alston and Martin 1992) can be verified using the equations above. Small, closed economy with no distortions: In a small country, price is exogenous. The simplest case with endogenous prices is that of a closed economy with no market distortions. In that case, the change in producer and consumer price is equal to Economic Surplus Measurement and Application 399 !1P = -k~A/(~A - OA ) = -kEA/(EA + llA ) =-KP ( A EA/(EA +l1A) =-ZP AS.22a) A where Z =- KEA/(EA + llA) is the fall in price relative to its initial value, and llA is the absolute value of the price elasticity of demand in country A, so that the corresponding change in quantity is (AS.22b) and the welfare effects are given by !1PSA = (k + !1P) (QA + O.5!1QA) = (K - Z) P AQA (l + 0.5Z11A) (AS.22c) !1CS A = -/1P (CA + O.S!1C A) = ZP AQA (l + O.SZl1A) (AS.22d) !1NS A = k (QA + O.S!1QA) = KP AQA (1 + O.SZl1A) = !1PS e) A + !1CS (AS.22 A Small, closed economy with distortions: As for the small open-economy case, the introduction of market distortions (in the form of per unit taxes or subsidies) with linear supply and demand does not change the impact of research on any of the prices. The introduction of distortions does, however, change the quantities produced and consumed, to which those price changes are applied in order to evaluate the welfare impact. Hence, the formulas for changes in prices and quantities, and for the welfare of producers and consum­ ers, are the same as for the undistorted closed economy (i.e., equations AS.22a­ d). The measured welfare effects will differ, however, because the base quantities for the calculation differ between the distorted and undistorted cases. In addition, there is the impact on taxpayer welfare to consider, and the formula for taxpayer welfare is as given in equation AS.19c for any combination of consumer and producer taxes or subsidies. Since output and consumption are equal, the taxpayer welfare effect is simply !1GS A = TA!1QA' where TA is the net per unit tax rate on consumption (or production). Therefore, !1PS A = (k + !1P) (QA' + O.S!1QA) = (K' - Z') P A'QA' (l + O.SZ' (AS.23a) llA') !1CS A= -/1P (CA ' + O.5!1C A) = z' P A' QA' (1 + O.SZ' llA') (AS.23b) (AS.23c) 400 Economic Surplus Measurement and Application IlNSA = k (QA' + 0.5.1QA) + TA.1QA = K' (A5.23d) P A' QA' (1 + 0.5Z' lh') - TA Q/ Z' 11A' As in the other cases shown here, the effect of the distortions is to change the incidence of the benefits from research but not to change the total benefits (Alston, Edwards and Freebairn 1988). It may be important to note that in this case, unlike the small-country cases, the preresearch price is affected by the tax so that the value of K in equations A5.23a-d (i.e., K') will be different from that in equations A5.22a-d (i.e., K) for a given value of k. Large, open economy with no distortions: The case of a large, open economy is significantly more complicated than the closed economy or small, open economy because it includes complications arising (a) from the fact that prices are endogenous and (b) because the quantities consumed domestically differ from the quantities produced domestically. Equation A5.16 defines the research-induced price change, and that equation may be transformed into elasticity form as follows: .1P=- k~A ~A + ~B - 0A - 0B (A5.24a) where SA = QA/(QA + QB) is the domestic share of global production, dA = CA/(CA + CB) is the domestic share of global consumption, and Z is the decline in the domestic (and world) price, which depends on overseas as well as domestic supply-and-demand parameters. The corresponding changes in production and consumption are as defined in equation A5 .18, and these may be expressed in terms of elasticities as follows: .1QA = eA (K - Z) QA; .1CA = 11A ZCA ) .1QB = - eB ZQB ; .1CB = (A5.24b 11B ZCB Then, formulas for domestic welfare effects can be obtained by making the appropriate substitutions in equations A5.19a-h. In elasticity form, these equations are .1PSA = (k + .1P) (QA + 0.5.1QA) A5.24c) = (K - Z) P AQA [1 + ( 0.5eA (K - Z) ] .1CSA = -.1P (CA + 0.5.1CA) = Z PACA (1 + 0.511A) (A5.24d) The net domestic benefits in this case would be equal to the sum of the Economic Surplus Measurement and Application 401 producer and consumer benefits thus obtained. It would be relatively straight­ forward to compute welfare effects in the other country and global effects as well. However, in order to deal with distortions in the large-country case, another layer of complications is introduced, and it is significantly more complicated to derive analytical solutions for the welfare effects on the different groups in terms of elasticities and so on when the different groups face different prices. That is, when different consuming and producing groups are involved that face different prices, the formulas do not simplify as readily as when quantities produced equal quantities consumed and buyers and sellers face the same prices. Further problems may arise unless care is taken to be consistent in the definition of the analysis. In particular, results may depend on whether a supply shift is defined as a proportion of the undistorted preresearch price (i.e., K) or as a proportion of the distorted preresearch price (i.e., K'). These issues are illustrated next in the case of a small, open economy. A5.2.2 Alternative Formulasfor a Small, Open, Distorted Economy In the analysis of research benefits with distorted markets, we assume linear supply shifting in parallel by a given amount, k, in the price direction, against a horizontal demand equation (in the small-country case, there is no price change). Subscripts 0 and 1 denote quantities with and without the research-in­ duced supply shift and the prime (') is used to denote quantities produced and consumed in the presence of policy (defined as a price wedge corresponding to a tax equal to T per unit). In this illustration, a negative value for T could represent the effects of an output subsidy or an import tariff or a floor price scheme, and a positive value could represent the effects of an output tax or a price ceiling, for example. Policy-induced change in quantity is equal to - /3T = Q/ - Qj = Qo' - Qo' where /3 is the slope of the supply function (i.e., when /3 and T are both positive, a tax causes a fall in output from Qo to Q / O )' The research-induced change in quantity (ilQ = Qj' - QO' = Qj - Qo = /3k) is unaffected by per unit susbidies or taxes. This means that the effect of price policy on producer research benefits arises only through a change in the initial quantity, upon which the given per unit benefit, k, is received (i.e, the base of a rectangle), from Qo to Q / O The triangle of producer benefits is unaffected by the pres­ ence of policy (because the height of the triangle, k, and the slope of supply, /3, are unaffected). In every case, then, the research benefits are given by the following: 402 Economic Surplus Measurement and Application with policy !:iPS = kQo' + O.5k!:iQ =k (Qo - /3n + 0.5k2/3 (A5.25a) =k Qo - k/3T + 0.5k2/3 !:iGS = T!:iQ = k/3T (A5.25b) !:iTS =! :iPS + !:iGS =k Qo + O.5e/3 (A5.25c) =k Qo + O.5k!:iQ without policy !:iPS = kQo + O.5MQ = (A5.26a) k Qo + O.5e/3 !:iTS = !:iPS (A5.26b) The difference in producer surplus, between the with- and without policy cases, is equal to (Qo' - Qo)k =- T/3k which is equal to the research-induced increase in subsidy cost. Thus, as can be seen above, the total (net) benefit is unaffected by the policy. In the formulas that use elasticities and percentage changes, these relationships can become obscured and they will hold in practice only when care is taken about the initial price (from which the supply shift is defined as a percentage change) and the initial quantity (to which the supply elasticity is applied to deduce the induced change in quantity - i.e., the implied value of /3 depends on the initial price and quantity combined with the supply elasticity). General Formulas in Terms of Elasticities The next step is to convert these formulas into formulas involving elastici­ ties and price wedges (i.e., T = tPw ) and percentage research-induced supply shifts. We confine attention to the with-distortion formulas from above and show the effects of different assumptions about whether the percentage supply shift, K, applies to the undistorted or distorted supply price [i.e., k = KP w or k = K(l - t)Pw, respectively]. Supply Shift Defined Relative to Undistorted Price (k = KP w) Producer surplus in terms of the undistorted quantity: From the formulas above, !:iPS = kQo - k/3T + 0.5k2/3 (A5.27a) Economic Surplus Measurement and Application 403 Substituting for k and T gives llPS = (KP w)Qo - (KP w) ~ ('tP w) + O.S(KP wP~' or (AS.27b) Then, using the definition that £ = ~P w / Qo is the supply elasticity at the undistorted equilibrium, Pw, Qn, the formula for producer benefits simplifies to tlPS = KPwQo (1- 't£ + O.5K£) (AS.27c) Producer surplus in terms of the distorted quantity: Alternatively, suppose we begin with the formula tlPS = kQo' + O.SktlQ (AS.2Sa) Substituting for k gives tlPS = (KP w)Qo' + O.S(KP w)2~, or (AS.28a) tlPS = KP wQo' [l + O.5K (~P wi Qo')] Then, using the definition that £' = ~(1 - 't)P w / Qo' is the supply elasticity at the distorted equilibrium, (1 - 't)Pw, Qo', the formula simplifies to (AS.28b) This is equivalent to the previous result but expressed now in terms of the distorted equilibrium quantity and the corresponding supply elasticity. Government revenues: The effects of research on government revenues are computed according to tlGS = TtlQ = k~T (AS.2Sb) Substituting for k and T gives tlGS = (KP w) ~ ('tP w) =K P wQo 't(~P wi Qo) (AS.29a) =K PwQo't£ Alternati vely, for a measure in terms of the distorted quantity, tlGS = (KPw) ~ ('tPw) =K P wQo 't [~(l - 't)Pw i Qo']/( l - 't) (AS.29b) =K PwQo''t£'/(l - 't) Net welfare: In the small-country case, the net (or total) welfare effect is equal to the sum of the effects on producer surplus and government revenues. 404 Economic Surplus Measurement and Application Beginning with the formulas defined in terms of the undistorted quantity, 6.TS =A PS + AGS =K PwQo [1- te + O.SKe] + KPwOote (AS.30a) =K Pw Qo [I + O.5Ke] Alternatively, using the formulas defined in terms of distorted quantities, ATS =A PS + AGS =K P wQo' [I + O.SKe'/(1 - t)] + KP wOo'te'(1 - t) (AS.30b) =K Pw Qo' [l + te' (I -1) + O.5Ke'/(I - t)] This formula is equivalent to the one above defined in terms of the undis­ torted quantity. Supply Shift Defined Relative to Distorted Price (k = K'(l - t)Pw ) Producer surplus in terms of the undistorted quantity: From the formulas above, 6.PS = kQo - k~T + O.S~~ (AS.2Sa) Substituting for k and T gives APS = [K' (I - t)Pw]Qo - [K' (I - t)p w] ~ (tPw) + O.S[K(l - t)Pw F~, or (AS.3Ia) llPS = K' (1 - t)Pw Oo [1 - t(~P wlQo) + O.5K(I- t)(~PwlQo)] Then, using the definition that e = ~PwlQo is the supply elasticity at the undistorted equilibrium, Pw, Qo, the formula simplifies to APS = K'(I - t)PwQo [1- te + O.SK'(l - t)e] (AS.31b) Producer surplus in terms of the distorted quantity: Alternatively, suppose we begin with the formula APS = kQo' + O.SkAQ (AS.2Sa) Substituting for k gives APS = [K'(l - t)p w]Qo' + 0.5[K(1 - t)pw ]2~, or (AS.32a) APS = K'(l - t)PwQo' [I + O.5K(~(l - t)PwIQo')] Then, using the definition that e' = ~(1 - t)Pw i Qo' defines the supply elasti- Economic Surplus Measurement and Application 405 city at the distorted equilibrium, (1 - 't)Pw, Qo', the formula simplifies to IlPS = K(1 - 't)Pw Qo' [1 + O.5K'E'] (AS.32b) This is equivalent to the previous result but expressed now in terms of the distorted equilibrium quantity and the corresponding supply elasticity. Government revenues: The effects of research on government revenues are computed according to IlGS = TIlQ = kPT (AS.2Sb) Substituting for k and T gives IlGS = [K'(l - 't)P w] P( 'tP w) = K'( 1 - 't)Pw Qo' t(PP wi Qo) (AS.33a) =K'(I-'t)PwQo'tE Alternatively, for a measure in terms ofthe distorted quantity, IlGS = [K'(l - 't)P w] P( 'tP w) =K 'P wQo''t [P(l - 't)Pw i Qo'] (AS.33b) = K' P wQo''tE' Net welfare: Beginning with the formulas defined in terms of the undis­ torted quantity, !!.TS = !!.PS + !!.GS =K '(l - 't)Pw Qo[l - 'tE + O.SK'( 1 - 't)E] + K' (1 - 't)Pw Qo't E (AS.34a) = K'(1 - 't)Pw Qo[l + O.SK'(1 - 't)E] Alternatively, using the formulas defined in terms of distorted quantities, !!.TS = IlPS + !!.GS =K '(1 - 't)P wQo'(l + O.SK'E') - K' P wQo''tE' (AS.34b) =K'(1-'t)PwQo'(1-'tE' +O.SK'E') This formula is equivalent to the one above defined in terms of the undis­ torted quantity. Implications In many studies, the most readily available information relates to the distorted equilibrium, so the natural place to begin is there. The formulas for producer surplus and government revenues using the quantity-and-supply elasticity at the observable (distorted) eqUilibrium are reasonably straightfor- 406 Economic Surplus Measurement and Application ward. The analyst is free to choose whether the K percent supply shift refers to current marginal cost (i.e., the distorted value) or the shadow value of output. The summary formulas for total welfare effects are somewhat more cumbersome and it might be simpler to compute the components separately and add them up. On the other hand, when the interest is only in the total (or net) effects, one can either deduce the prices and quantities at the undistorted equilibrium and apply the K shift to the standard formula (since we know that in this model the presence of distortions of these types does not affect the total benefit), or alternatively, one can use the formulas developed above. In some cases, it might be simpler, and less prone to errors, to parameterize the linear supply-and-demand curves themselves, rather than rely on using elasticities, and to use a per unit k rather than a percentage K to calculate welfare effects. Selected Formulas Some examples of formulas that can be used in common cases for measuring research impact on the welfare of producers, consumers, and government revenues are given in table A5.2.1. Table AS.2.1: Selected Formulas for Computing Changes in Economic Surpluses o-1:> 0. Model Formula "'-I Corresponding chapter 4 figure Closed-economy models 1. Basic closed economy flCS = PoQc7 (1 + O.5Z1'\) Figure 4,1 MS = PoQo(K - Z)(1 + 0.5Z1'\) flTS = PoQoK (1 + 0.5Z1'\) 2. Closed economy with flCS = Po'Qo'Z (1 + O.5Zr!) Figure 4,7 exogenous demand shift MS =P o'Qo'(K' -Z)(l + O.5Z1'\) flTS =P o'Qo'K' (1 + 0.5Z1'\) Open-economy models 3. Small open economy flCS=O Figure 4.5, panels a and b (home-country effects) MS=flTS=PwQoK(1+0.5KE) =PoQoK(1 +0.5KE) 4. Large open economy (no technology spillovers): Z = - (PI - P ~/Po 4a. Home-country flCS A = Po C A.c7(l + O.5Z1'\A) Figure 4,2 and 4.3 (i.e., country A) effects MSA = P0 Q A.O (K - Z)(1 + 0 .5 2£A ) box 4.2 flTSA = flCSA + MSA 4b. Rest-of-world flCSs = Po Cs.c7(l + 0.5Z1'\s) Figures 4.2 and 4.3 (i.e., region B) effects MSs =- Po Qs,c7(l + O.5ZEs) flTSs = flCSs + M5s Table A5.2.1: Selected Formulas for Computing Changes in Economic Surpluses (contd.) -I:>.. ~ Model Formula Corresponding chapter 4 figure 4c. Global (i.e., world) effects ACSw =A CS A + ACSB Figures 4.2 and 4.3 APSw = APS A + APSB ATSw = ATSA + ATSB =A CSw + APSw 5. Large open economy (with technology spillovers): Note that Z here is necessarily a different value than Z with no spillovers 5a. Homecountry ACSA =P OCA,O Z(l +O.5ZTlA) Figure 4.4 (i.e., country A) effects APS - P Q (K - Z)(l + 052£ \ A - 0 A,O A,A • B' ATSA = ACSA +APSA 5b. Rest-of-world ACSB= P OCB,o 2(1 +0.5Z11B) Figure 4.4 (i.e" region B) effects APSB =P OQB,O (KBA - Z)( 1 + O.5Zta) ATSB =A CSB + APSB 5c, Global (i.e., world) effects .6.CSw = floGS A + ACSa Figure 4,4 APSw =A PSA + APSB ATSw =. 6.TSA + ATSB =A CSw + APSw 6, Small open economy with output price supports (home-country effects): PM IN = (1 - t)Pw where t < 0 6a. Proportionate supply shift ACS =0 Figure 4.15 defined relative to .6.PS =K '(1 - t)pw Qo' (1 + O.5K'e') distorted market AGS = tK'(1- t)P Q 'e' equilibrium (k =K ' PM IN) .6.TS = ACS + .6.PS : A~S =K '(l-t)Pw Qo' (1 +O.5K'e') +K'(l-t)Pw Qo'te' =K'(l-t)pw Qo' (1 +te' +0,5K'£,) Table AS .2.1: Selected Formulas jor Computing Changes in Economic Surpluses (contd.) "~" " Model Formula Corresponding chapter 4 figure 6b. Proportionate supply shift ll.CS=O Figure 4.15 defined relative to ll.PS = KPw Qo (I - 'tE + 0.5KE) undistorted market ll.GS = 'tPw QoKE equilibrium (k = KPw ) ll.TS = ll.CS + ll.PS + ll.GS = Pw QoK(1 + 0.5KE) 7. Small open economy with output price ceiling (home-country effects): PM AX = (1 - 't)pw and 't > 0 7a. Proportionate supply shift ll.CS = 0 Figure 4.16 defined relative to ll.PS = K'(I - 't)Pw Qo' (I + 0.5K'E') distorted market ll.GS =' tK'(l - 't)P Q 'E' equilibrium (k = K' PM AX) ll.TS = ll.CS + ll.PS : ll.~S = K'(I -'t)Pw Qo'(1 + 0.5K'E') + K'(l- 't)Pw Qo''tE' =K '(l- 't)Pw Qo'(l + 'tE' + O.5K'E') 7b. Proportionate supply shift ll.CS=O defined relative to ll.PS = KPw Qo [1 - 'tE + O.5KE] undistorted market ll.GS = 'tPw QoKE equilibrium (k = KPw) ll.TS = ll.CS + ll.PS + ll.GS = Pw QoK(1 + O.5KE) Table AS.2.I: Selected Formulasfor Computing Changes in Economic Surpluses (contd.) c.".".. ".. Model Formula Corresponding chapter 4 figure 8 Small open economy with import tariff (home-country effects), i.e., 't < 0 8a. Proportionate supply shift des = 0 Figure 4.17 defined relative to MS = K(l - 't)Pw Qo'(1 + 0.5KE') distorted market dGS = 'tK( 1 - 't)Pw Qo' E' equilibrium(k=K(l +npw) dTS = deS + MS + dGS = K(1- 't)Pw Qo'(l + 0.5K'E') + K(1- 't)Pw Qo''tE' = K(l - 't)Pw Qo'(l + 'tE' + 0.5K'E') 8b. Proportionate supply shift des = 0 Figure 4.17 defined relative to MS=KPw Qo(1-'tE+0.5KE) undistorted market dGS =' tPw QoKE equilibrium (k = KPw ) dTS = deS + MS + dGS = Pw QoK(1 + O.5KE) Note: See text on pages around the figures in chapter 4 that are cited above for definitions of variables. For notational simplicity, country and regional subscripts have sometimes been suppressed. Unless otherwise indicated, all surplus changes refer to home-country effects. In all of these formulas, the demand elasticity is in terms of absolute values (i.e., TI > 0), K defines the proportional shift down of supply (not change) (i.e., K> 0 for a decrease in marginal cost) and Z is the proportional reduction (not change) in price (i.e., Z> 0 for a decrease in price). Differences between the measures from models 6a and 6b (or 7a and 7b or 8a and 8b) arise in part because (a) a different porportionate supply shift is implied (i.e., KPM IN versus KPw ) and (b) the elasticity of supply (E' versus E) in the formula applies to a different point on the supply curve (i.e., Q; versus Qo' where Qo' = (1 - 'tE) Qo when E is defined at Qo and Qo' > Qo when 't < 0), so that a different supply slope and a different research-induced change in quantity is implied unless care is taken to adjust the elasticity accordingly. Note, also, that formulas for distorted markets are the same for the minimum or maximum price policy or tariff (models 6, 7, and 8) once the policy is expressed as an "equivalent" ad valorem tax, 'to The signs of the effects vary depending on whether't is a tax ('t > 0) ora subsidy ('t < 0). Thus, 't > oi mplies an increase in government revenue due to research and smaller research benefits for producers. Economic Surplus Measurement and Application 411 Appendix A5.3: Estimating K Using Industry and Experiment Data Studies of returns to research using economic surplus models require an estimate of the shift in supply due to research-induced technical change. In the literature, the supply shift has been represented either as a vertical shift down, K, or a horizontal shift to the right, J. To estimate the value of J or K for a particular technical change - or for a particular research investment - a number of approaches have been used. For example, experimental yields have been used in ex post analysis as a proxy for industry yield or supply shifts. Echeverria, Ferreira and Dabezies (1989), Pardey et al. (1992), and Palomino and Norton (1992b) present recent examples. In ex ante evaluations, scientist questionnaires and other mechanisms for eliciting information about potential J or K have been employed, as discussed earlier in this chapter. The purpose of this appendix is to clarify the relationships between the industry (final product) supply shift (J or K), experimental yields, and industry yield changes for different types of technical changes. Neutral and biased technical change are considered in a two-factor linear­ elasticity equilibrium-displacement model. The results illustrate the import­ ance of the supply elasticity and the elasticities of substitution among factors of production in determining the interrelationships among changes in exper­ imental yields, industry yields, and final product supply. AS.3.] A Two-Factor Equilibrium-Displacement Model Following Alston (1991), we can model the effects of technical change on the market equilibrium of a competitive industry producing a homogeneous product using two factors of production in terms of the following six linear elasticity equations (equivalent to text equations 4. 14a' to 4.140: E(Q) = 11E(P) (AS.3Sa) E(P) = s] E(W]) + S2 E(W2) (AS.3Sb) E(X]) = 11]] E(W]) + 1112 E(W2) - 0] (AS.3Sc) E(Xz} = 112\ E(W\) + 1122 E(W2) - °2 (AS.35d) E(X]) = £] E(W]) (AS.3Se) E(X2) = £2 E(W2) (AS.3Sf) where E denotes relative changes (i.e., for a variable Z, E[Z] = dZtZ = d[InZ]), 11 is the elasticity of demand for the product measured in natural 412 Economic Surplus Measurement and Application units (i.e., 11 < 0), Si is the cost (and revenue) share of factor i (i.e., Si = WXIPQ) and in this two-factor case, SI + S2 = 1, 11ij is the uncompensated cross-price elasticity of demand for factor i with respect to price of factor j, 01 and ~ are factor-demand-shift variables reflecting changes in technology, and £i is the elasticity of supply of factor i. The endogenous variables in the model are industry output, Q, the amounts of the two factors used by the industry, XI and X2, the price per unit of the final product, P, and the factor prices, WI and W2• Equation AS.3Sa is the demand schedule for the industry's output, equation AS.35b is a zero-profit condition reflecting a constant-retums-to-scale industry production function, equations AS.3Sc and A5.3Sd are derived factor-demand equations, and AS.3Se and AS.3Sf are the factor-supply equations. The solutions to this model are E(WI ) =- (~- 1122)01/D - 111202/D (AS.36a) E(W2) = - 1121011D - (£1 - 1111 )O/D (A5.36b) E(XI ) =- £1(Ez - 1122 )o/D - £111 120/ D (AS.36c) E(X2) =- £21121 0/ D - £2(£1 - 1111)02ID (AS.36d) E(P) =- [sl(Ez - 1122) + s211 21 01 + si£1 - 11 11 ) + sl111202]/D (AS.36e) E(Q) =- [SI(£2 - 1122) + s21121 01 +S2(£1 -1111) + SI111202]11ID (A5.36f) where D = (~ - 1122)(£1 - 1111) - 112111 12 > 0 Using Slutsky symmetry and the homogeneity of the cost function, we can define the four own- and cross-price elasticities of factor demand using factor cost shares, Si' the final demand elasticity, 11, and the elasticity of substitution between XI and X2 ( i.e., eJ) 1111 = - S2eJ + sl11 1112 = S2eJ + S211 ( 1121 = SleJ + SI11 1122 =- AS.36g) SleJ + s211 Substituting equation AS.36g into equations A5.36c, AS.36e, and A5.36f gives E(XI) =- £1(£2 + SleJ - s211)0/D - £lsieJ + 11)O/D (AS.36c') E(P) =- [SI(£2 + eJ)OI + S2(£1 + eJ)02]/D > 0 (AS.36e') E(Q) =- [SI(£2 + eJ)OI + S2(£1 + eJ)02]11ID > 0 (AS.36f) Economic Surplus Measurement and Application 413 where D = a (Sl EI + S2 E2 - T) - T) (SI Ez + S2 E1) + EI Ez> 0 for T) < 0 and s, E(, ~ > 0 The relative change in yield (Y =Q IX) is E(Y) == E(QIX1) =E (Q) - E(X1) (AS.37a) Substituting equations AS.36c' and AS.36e' into equation (AS.37a) gives E(Y) = [(Sla[E1-T)] - T)[SI E2 + S2E(] + E1E2)01 (AS.37b) + S2a(Ez -1)02 ]/D Experimental Yield Data Neutral technical change: Consider a neutral technical change that we can model by setting O( = O2 = 0 in the equations above. The results are E(XI) = - EI(a + E2)0/D > 0 (AS.38a) E(P) = - (SIE2 + S2EI + a)OID < 0 (AS.38b) E(Q) = - (s lE2 + S2El + a) 8T) I D > 0 (AS.38c) E(Y) == [a(E1- T) - T)(SI~ + S2EI) + E1E2] olD> 0 (AS.38d) To represent an experimental setting in which both input quantities are held constant we make the factor supply functions perfectly inelastic (i.e., E( = E2 = 0). Under these conditions, the change in yield is equal to the change in output (E( y) =E (Q». A neutral technical change, 0, would increase output and experimental yields by E(Y*) =- T)a&D =0 (when E( = Ez =0 , D = - T)a). This would be the observed change in experimental yields due to the change in technology. In a firm or industry setting (i.e., allowing factor­ supply response), the change in yield might be quite different, depending upon the extent to which inputs are variable and substitutable for one another. The shift in the industry supply function might be proxied by the increase in experimental yields, but that might not be a good idea. To measure the vertical shift in industry supply, K, due to the neutral technical change, we set T) = 0 in equation AS.38b and get K == - E(P) (given Q ) = 0 (s1E2 + S2EI + a )/[a (S(EI + S2E2) + E1E2] > 0 (AS.39) To measure the horizontal shift in industry supply, J, due to the neutral technical change, we set T) =- 00 in equation AS.38c to get 414 Economic Surplus Measurement and Application J = E(Q) (given P) = 0> 0 (A5.40) Notice that J = £rK, where Er is the elasticity of supply of the industry product, Q, and is given by (A5.4l) Thus, for neutral technical change (as defined) the increase in experimental yields is a good measure of J, the horizontal supply shift at the industry level. However, it might not be a good measure of K, the corresponding vertical supply shift. The correct measure of the vertical supply shift is K = O/~and to get K =° re quires restrictions on elasticities of factor supply, factor substitu­ tion, or factor cost shares such that the elasticity of product supply is 1.0. Biased (land-saving) technical change: Let us suppose that XI is land. To model a biased (land-saving) technical change, set 02 =0 in equations AS.36c', AS.36e', AS.36f, and AS.37b to get E(X1) =- £1(£2 + Slcr - s211)0/D < 0 (A5.42a) E(P) = - sl(Ez + cr)o,ID < 0 (AS.42b) E(Q) =- S'(£2 + cr)O,11ID > 0 (A5.42c) E(l') = [s,cr(£, -11) -11(S'£2 + S2£) + £'£2 ]o,ID > 0 (AS.42d) where D = cr(s,£, + S2£2 -11) -11(S'£2 + S2£') + £IEz > 0 for 11 < 0 and cr, e" Ez> O. The effect of this technical change on experimental yields (holding factor quantities constant by setting £, = Ez = 0) is given by E(Y*) = E(Y) (given XI and X2) = E(Q) (given X, and X2) = s,O,. To measure the vertical shift in industry supply due to the biased technical change, we set 11 =0 in equation AS.42b to get K =- E(P) (given Q ) =S l ( £2 + cr) O/[cr (SI£1 + S2£2 ) + £1£2] > 0 (AS.43) Setting 11 = - 00 in equation A5.42c, the corresponding horizontal shift in supply is J = E(Q) (given P) = Sl (£2 + cr) O/[cr + SI£2 + S2£1 ] = £T K > 0 (AS.44) Thus, with biased (land-saving) technical change, the increase in experi­ mental yields (E(Y*) =s ,O,) is not likely to be a good measure of either the Economic Surplus Measurement and Application 415 vertical or horizontal shift in industry supply. Parametric restrictions are needed to make experimental yield increases a good measure of the vertical or horizontal supply shift (either K or 1) resulting from land-saving technical change. For instance, when the factor-supply elasticities are equal (£1 = £2)' J = 0)\ = E(Y*) and K = SIOI€r= E(Y*). This situation seems unlikely. Biased (land-using) technical change: To model a biased (land-using or Xz-saving) technical change, set 01 = 0 in equations AS.36c', AS.36e', AS.36£', and AS.37b and get (AS.4Sa) iff cr + Tl < 0 (i.e., inputs 1 and 2 are gross complements) (AS.4Sb) (AS.4Sc) E(Y) = sp (£1 - Tl )O/D > 0 (AS.4Sd) where D = cr(SI£1 + sz£z - Tl) - Tl(sl£z + SZ£I) + £I£Z> 0 for Tl < 0 and cr, lO l, £z> 0 The effect of this technical change on experiment yields (holding factor quantities constant by setting lOl = £z = 0) is given by E(Y) (given Xl and Xz) =E (Q) (given Xl and Xz) = szoz· To measure the vertical shift in industry supply due to the biased (XI-sav­ ing) technical change, we set Tl =0 in equation (AS.4Sb) and get K = - E(P) (given Q) =S z (£\ + cr) 0zl[cr (s\£\ + sz£z) + £\lOz] > 0 (AS.46) The corresponding horizontal shift in supply due to the biased technical change (i.e., with Tl =- 00 in AS.4Sc) is J = E(Q) (given P) =S z (£\ + cr )ozl[cr + s\£z + Sz£\] = lOr K> 0 (AS.47) The experimental yield increase from a land-using technical change is analogous to that from the land-saving technical change: E(Y*) = szoz. Again (as with land-saving biased technical change), the experimental yield data are unlikely to provide an accurate measure of either the vertical or horizon­ tal shift of the industry supply function. 416 Economic Surplus Measurement and Application Industry Yield Data An alternative approach to measuring J and K is to look at the growth in industry yields as a consequence of the introduction of new technology. The corresponding measure is E(y) in the equations above. That too may lead to misleading results. The algebraic results for the effects of neutral and biased technical changes on experimental yields, E(Y*), vertical supply shift, K, horizontal supply shift, J, and industry yields, E(y), are summarized in table AS.3.1. AS.3.2 Summary ofA lgebraic Results In table AS.3.2 it can be seen that a neutral technical change, 0, will lead to an increase in experimental yields by the same percentage, and the increase in experimental yields, E(Y*), will be the same as the consequent horizontal shift of supply, J. However, the increase in experimental yields will be a good measure of the vertical supply shift, K, only when the output-supply elasticity is unitary; otherwise (and more generally) it is necessary to divide the increase in experimental yields by the supply elasticity to get the vertical supply shift. The increase in industry yields will be a good measure of the increase in supply, J, when the factor proportions are fixed (i.e., the elasticity of factor substitution is zero: a = 0); otherwise it might not be a very good measure. In the case of a biased (land-saving) technical change, 01' the increase in experimental yields, 0lal, is a good measure of the horizontal shift of supply, J, only when the factor-supply elasticities are equal, EI =£ :t. Again, it is necessary to divide the horizontal supply shift, J, by the output-supply elastic­ ity, e" to estimate the vertical supply shift, K. As with the neutral technical change, the change in industry yields will be an accurate measure of the horizontal supply shift when factor proportions are fixed. In the case of a biased (land-using) technical change, 02' the increase in experimental yields, 02a2' is a good measure of the horizontal shift of supply, J, only when the factor-supply elasticities are equal. Again, it is necessary to divide the horizontal supply shift, J, by the output-supply elasticity, ET , to estimate the vertical supply shift, K. The change in industry yields, E(y), will be zero in the case of fixed proportions. Theoretical conditions under which changes in experimental yields and industry yields are accurate measures of the shift of industry supply are shown in table AS.3.2. Table A.5.3.l: Consequences ofN ew Technology for Yields and Supply Shifts ...t...,.. "'l Type of technical change Effect Neutral (0) Land-saving (o}) Land-using (02) E(Y*) 0 OISI 02S2 J 0 OISI(~ + 0') 02S2(£1 + 0') 0'+SI~+S2£1 0' + Sl~ + S2£1 K 0(0' + SI~ + S2£1) OISI(~ + 0') 02S2(£1 + 0') O'(SI£I + S2~ + £I~) O'(SI£I + S2~) + £I~ O'(SI£I + S2£2) + £I~ E(Y) o[ 0'(£1 -ll) -ll(SI~ + S2£1) + £I~] 01[SIO'(£1 -ll) -ll(SI~ + S2£1) + £I~] 02S20'(£1 -ll) D D D Note: D = O(SJEJ + S2(2) + EJE2 -11(0 + SJE2 + SZ(2) . ] = the percentage horizontal (rightwards) shift of product supply K = percentage vertical (downwards) shift of product supply E(Y*) =t he percentage change in experimental yields (QIXJ) holding factors fixed E(y) = the relative change in industry yields In every case, K=]IET, where ET= [O(SJEJ +S2(2) +EJE2]/(0+SJE2 +SZEJ) 418 Economic Surplus Measurement and Application Table A5.3.2: Sufficient Conditions under Which Yield Changes Accu­ rately Reflect Supply Shifts Type of technical change Neutral (0) Land-saving (01) Land-using (02) E(Y*) =J always when 101 =102 when 101 = 102 E(Y*) = K when€r= I when 101 = 102 when 101 =102 and £r= I and £r= 1 E(Y) = J when 0= 0 when 0 = 0 and when 101 = 102 = 0 s ,(I - 102) = (1 - S1)€1 E(y)= K when 0=0 when 0 = 0 and when 10, = 102 = 0 and €r= 1 s1(1 - 102) = (1 - s1)€1 and £r= 1 and £r= 1 Economic Surplus Measurement and Application 419 Appendix A5.4: Data for Estimating the Supply-Shifting Effects of Research In chapter 5, the basic data required to implement a research-evaluation priority-setting study were described. Both the market- and research-related data used in these types of analyses were reviewed and some of the main conceptual and empirical issues were also discussed. This appendix builds directly on that discussion and develops in more detail a set of procedures and guidelines for eliciting information and compiling data to estimate the research-induced shift in supply. In compiling these data, it is important to proceed in an organized and structured fashion. This facilitates estimating the economic consequences of research. And well-constructed primary data themselves are useful decision­ making aids. Moreover, institutionalizing the evaluation process demands a structured and documented approach to data gathering and processing, given the relatively frequent turnover of analysts and decision makers in many agricultural research institutions, particularly those in developing countries. Accumulating primary market- and research-related data over time on a con­ sistent basis can also confer significant economies of size and scope regarding the data acquisition and processing aspects of this work. It also builds the basis for the whole evaluation cum priority-setting exercise to take on a monitoring function as well. Systematically cross-checking scientists' best guesses about the future impact of their current work against the current impact of their past work is valuable from various vantage points, be it scientists thinking about the prospects of socially valuable impact coming from their planned research or those individuals operating at more strategic levels in the decision-making hierarchy who are reviewing the likely consequences of reallocating resources between or within different programs. While the data requirements are much the same for ex post and ex ante analyses, the emphasis here is on compiling estimates of the economic consequences of research that is yet to be done. By explicitly linking these likely research effects to the planned deployment of research resources, it is possible to gain a structured appraisal of the implications of alternative resource allocation decisions, information that is valuable for choosing among alternatives. The topics dealt with in the elicitation forms and guide­ lines on data compilation described below include • Research resources - research personnel (commodity focus, research program areas) - research costs (total resources, research program details) • Research impact 420 Economic Surplus Measurement and Application - aggregate, local yield effects - factor bias and additional input costs - spatial variation in yield effects • Research dynamics - research and development lags - adoption parameters (including research depreciation) • Research risk - sensitivity analysis • Reconciliation The topics are grouped and ordered in a sequence that should prove acceptable for, or that can be adapted to suit, most evaluation studies.88 However, it is simply not feasible to design a questionnaire or elicitation process that is of direct use in all applications. Nevertheless, these prototypi­ cal forms and their accompanying notes should be relevant in many circum­ stances, particularly when it is the economic consequences (at the farm level) of broad programs within a national agricultural research system or agency that are being evaluated and prioritized. In any case it is a relatively straight­ forward job to modify these forms so that they can be used in other contexts, e.g., when dealing with research that affects the costs of production at various stages in the production-marketing chain or when applying the procedures to a multicountry, regional priority-setting exercise. Benchmarking: Simply asking scientists and others to estimate likely yield increases or unit cost reductions, the time required to complete the research, the riskiness of the research, and adoption and research depreciation rates will not suffice. The approach adopted here is to assemble both historical and projected or elicited data, so that analysts can combine information from various sources in an intelligent fashion, drawing on the ideas and arguments discussed in chapter 5 to derive estimates of parameters. Wherever possible, historical data on past experimental and industry yield gains, prior rates of uptake of new technologies, the current cost structures of production, and so on should be explicitly incorporated into the elicitation process to condition and thereby calibrate the responses of scientists and others. Alternative research scenarios: Research evaluation involves assessing and comparing alternatives. When eliciting research-related data and evalu­ ating the economic consequences of the research, it is incumbent on the analyst to be clear about just what research scenarios are being assessed and compared. In particular, it is necessary to be completely clear about what is being held constant and what is being allowed to vary between any pair of alternatives that are being compared. 88. These illustrative elicitation fonns are based on Pardey et al. (1992). Scobie and Jacobsen (1992). Dey and Norton (1993). and Bantilan and Lantican (1994). Economic Surplus Measurement and Application 421 In many ex ante studies, it is an ongoing rather than a new program of research that is being evaluated. This has direct consequences for the de(ini­ tion of the with- and without-research scenarios. When evaluating an on­ going research program, the with-research scenario often implies or explicitly uses a baseline that presumes an indefinite continuation of the current program of research. The corresponding without-research situation usually implies that none of the baseline research is conducted. But other interpretations are possible. Some scientists might think that without-re­ search means without any research by anyone, rather than simply eliminating the specific program being discussed. Or, a scientist might make the com­ mon mistake of confusing information about changes over time (i.e., before and after) with the information being sought about changes attributable to changes in a particular factor, research (i.e., with and without). Having established the baseline with and without cases, the implications of deviating from this baseline (e.g., plus or minus 15% of the baseline research budget) can be established to give some indication of the marginal trade-offs in­ volved in reallocating research resources. Generally, when doing this type of sensitivity analysis, the without-research situation continues to be a case in which none of the baseline research is undertaken. Alternatively, the baseline could inv,olve a smaller temporary or perma­ nent change in the current program of research. For example, the baseline (with-research) scenario may involve a continuation of the existing pattern of research investments while the "without"-research case refers to a one­ shot, permanent, decrease in funding from the baseline. Or, the "without"­ research situation could involve a change in funding that persists only for a finite period, say one round or cycle of research that runs for the time it takes to develop a new crop variety. Alternatively, the baseline could be defined as a sequence of research program expenditures that differs from the current program, and the alternative could be different in any way thought to be relevant for comparison. One of the most difficult, and potentially useful parts of a research evaluation cum priority-setting project is to define mean­ ingful, relevant alternatives to be assessed. A closer correspondence between the alternatives defined for assessment (on which explicit information is elicited) and the real options being considered enhances the possibility of a useful outcome. Exactly which scenario is being evaluated has important implications for what data to collect or elicit and how to interpret the results of the subsequent analysis. Many studies are unclear on this point, so the elicited parameters (particularly regarding depreciation and adoption) used to estimate a re­ search benefit stream involve a confusion of scenarios that often don't properly correspond to the cost stream used as the basis for analysis. 422 Economic Surplus Measurement and Application Joint determination of research parameters: Both in practice and in their elicited responses, scientists may well trade off smaller maximum yield gains for shorter R&D lags, or longer R&D lags for a higher probability of successfully completing the research, or various other combinations of these research-related parameters. And most adoption studies find that profitabili­ ty is a critical determinant of adoption rates: the rate of uptake of a new technology is positively related to the size of the yield-increasing or cost­ reducing effects of research. These trade-offs should be dealt with explicitly in the research evaluation framework. A preferred option may be to benchmark the elicitation of all these parameters on a baseline deployment of research resources, making it clear to those from whom the information is elicited that the parameters are to be viewed as jointly determined. This makes it possible to consider in a structured way the effects of changes in research resources on all the parameters that are relevant for estimating the likely benefits from research. An important aspect of the elicitation-cum-parameter estimation process is reviewing and reconciling the various estimates to ensure that they are internally consistent for a given research scenario and meaningfully compa­ rable among alternative research options. An alternative approach, and one adopted by Davis, Oram and Ryan (1987), is to fix the (maximum potential) unit cost reduction at 5% of current production costs for each commodity being analyzed and to elicit, or esti­ mate, the R&D lags, probabilities of research success, and adoption parame­ ters compatible with the presumed unit cost reduction. One difficulty with this approach is that the baseline 5% unit cost reduction may lie well outside the range of previous experience or future prospects, making meaningful estimates of the remaining research-related parameters unlikely. Another, and perhaps more fundamental, difficulty is that this approach treats kMAX as an exogenous parameter. Estimates of the R&D lags, probabilities of research success, and adoption parameters are elicited, conditional on a 5% /(MAX but without explicitly linking those estimates to the change in research resources thought necessary to achieve such a supply shift. It seems much more natural to think of the research resources themsel ves as the exogenous (or decision) variable and to analyze the likely supply-shifting effect, given various research scenarios that are of interest to decision makers. Modeling the changes in the research program that are thought necessary to generate a 5% decrease in unit costs across all commodity research programs may provide little in the way of useful information. Indeed, if an option being considered were to close down one research program and redeploy the resources among other programs, for instance, it would be much more natural and meaningful to model that reallocation. Economic Surplus Measurement and Application 423 Guidelines: The following guidelines are developed with the typical research evaluation study in mind. The usual sequence of events is to (a) compile the relevant data on research input and resource deployment to benchmark the cost side of the analysis, (b) assemble and process pertinent data, such as historical yield trends, current cost of production, past adoption practices, as well as relevant agroecological data (to both benchmark sci­ entists' responses and help structure the elicitation of the technical parame­ ters used to estimate the benefit side of the analysis), and (c) jointly elicit the set of research-related parameters corresponding to clearly defined research program alternatives, using the current program as a benchmark. Naturally, not all studies proceed in such a linear fashion. In some cases a pilot evaluation exercise for a limited number of commodity research programs will be undertaken to familiarize scientists, analysts, and decision makers with all the steps in the process before proceeding with a more comprehensive priority-setting exercise. A careful evaluation exercise involves a good deal of interaction with scientists and decision makers. Elicited research-related pa­ rameters are often recalibrated and revised when new information becomes available, often as a consequence of the evaluation process itself, or as various sensitivity analyses are performed to provide decision makers with relevant information on which to assess the implications of alternative research options. A well-integrated evaluation-cum-priority-setting effort will be cycling through various versions of data elicitation, data processing, and result presen­ tations as new questions and new opportunities for doing worthwhile analyses present themselves. 424 Economic Surplus Measurement and Application AS.4.1 Elicitation Form Cover Sheet Table AS.4.1: Example of an Elicitation Form Cover Sheet Institute Name: Institute Address: Respondent details Elicitor details Commodity! Research research area Name Position specialty Name Date Note: To expedite data reviews and revisions, it is helpful to keep track of who performed and who participated in the data-elicitation exercise. Economic Surplus Measurement and Application 425 A5.4.2 Research Resources Personnel Table A5.4.2: Research Resources: Human Resource Information by Commodity Year: Scientistsa Technical support staff Commodity! research area Number Share Number Share (fte) (%) (fie) (%) Other Institute total x 100% x 100% Note: Research personnel data that are stratified by commodity OJ,research area for the current year or, preferably, the past several years (or, even more desirable, a lengthy time series of past years) provide a useful indication of the current and likely future focus of a research institute, organization, or system. For evaluation purposes at least, it is recommended that the resources devoted to factor-oriented research (e.g., research on soil, water, and so on) be identified as a component of a specified commodity program whenever it is possible and appropriate to do so. Much, if not most factor-oriented research is done to improve the value of a natural resource that is (or is likely to be) used to produce a known output. With this approach, only the resource management and conserva­ tion research that is difficult to allocate in this way (because, for example, it jointly affects multiple outputs) is classified in a residual, "noncommodity," or "other" research category. Explicit use of vertically disaggregated market models may be required to model and measure the effects of this type of research. aIt is often easier to allocate full-time equivalent (fte) researchers and associated support staff, rather than research expenditures, to specific research areas. If the program-specific fte researchers are difficult to obtain from published records or by elicitation, it may be useful to identify the fte total for the institute or system being evaluated, determine the share of total ftes working for specific commodity programs, and then prorate the fte total to specific programs using the corresponding share figures. 426 Economic Surplus Measurement and Application Research Program Area Table A5.4.3: Research Resources: Human Resource Information by Research Program Area Commodity: Year: Scientists Research program areasa Number Share (jte) (%) Plant breeding Plant protection Soil management and fertilizer Crop production practicesb Other ----------------------------- ------------------- -------------------- Post-harvestC Commodity total x 100% Note: Having identified the commodity orientation of the research being assessed, it is useful for evaluation and resource-allocation purposes to identify the research program areas of each commodi­ ty program. This information could be compiled from existing management and accounting data, perhaps drawn from databases developed using the INFORM program (Gijsbers 1991). It is also useful to elicit this same information from scientists as a check against data obtained from other sources published and unpublished) or as a substitute for such data when they are not available. aCorresponding categories for livestock research are animal breeding (or genetic improvement), animal health (including the vetwinary sciences), livestock management, and perhaps, a separate category for feed and nutrition (which may include research on pasture management). blncludes research on planting densities and timing, cropping patterns and rotations, irrigation practices, and so on. cPostharvest areas of work include a wide range of research related to on- and off-farm issues that need to be explicitly identified in each case. If a significant share of the research is postharvest in nature, then vertical-market models and commensurate data will need to be assembled to preform the evaluation. See Scobie and Jacobsen (1992) for details of such an evaluation-cum-priority-setting exercise. Economic Surplus Measurement and Application 427 Research Costs Detailed information should be gathered on the total costs and, where possible, on the form of expenditure (e.g., capital versus labor) at the level of the relevant alternatives to be assessed, and with a view to total constraints (table A5.4.4). Thus, for instance, if the analysis pertains only to a particular research institute, information would be gathered on the total expenditure by that institute over recent years, the mix of expenditures on different types of research inputs, and the mix of expenditures on different research programs carried out by the institute that are to be considered separately in the analysis. Details are usually not available on the expenditure mix by major research programs, but such information may be available in some cases and could be useful. Alternatively, a NARS might want to collect disaggregated details on research costs by institute as well as by program, depending on the alterna­ tives to be assessed in the analysis. Research Program Details To benchmark the analysis around the current deployment of resources, it is helpful to get a structured view of the research orientation of the current program. To do this, it could be useful to compile a list of current (and, if relevant, proposed) projects, identifying each project or research theme (consisting of a logical grouping of projects or areas of research) by its • commodity focus (see table A5.4.2) • research program area (see table A5.4.3) • spatial (agroecological) focus • research problem focus • linkages among projects within the program and to projects in other areas Table A5.4.4 Total Resources and Disposition "N Year Labor" 0'" 0 No I I Maintenance I I I Totallabor Total research I Total extension Date Capital costs costs Operating costs fte Cost per fte costs costs costsb I 1990 x x x x x x x x 2 1991 x x x x x x x x 3 1992 x x x x x x x x 4 1993 x x x x x x x x 5 1994 x x x x x x x x 6 1995 7 1996 8 1997 9 1998 10 1999 II 2000 12 2001 13 2002 14 2003 15 2004 Note: For an ex ante evaluation, data files from management and accounting, or, in their absence, elicited data, can be used to establish the baseline deployment of research and extension funds. The baseline is usually taken to be the current spending pattern, or, preferably, the average spending pattern for the past three years. A time series of past expenditures in conjunction with elicited data from research administrators, scientists, and others is usually the best basis for projecting forward the baseline spending scenario. Pardey and Roseboom (1989) and Roseboom and Pardey (1993) discuss data compilation and processing issues while Pardey et al. (1993, appendix 2) provide substantial details on the methods used to compile commodity-specific, time-series, research spending data for Indonesia. See also Pardey, Roseboom and Craig (1992) for further details on deflation and currancy conversion issues related to aggregates of agricultural research spending. "It might only be possible to distinguish between capital and noncapital costs, in which case a labor cost series could be estimated by scaling up estimates of cost per fte using the number of fte researchers. Likewise, non capital costs might be projected forward using current and projected noncapital costs per researcher, scaled by the corresponding fte researcher series. "Extension agencies are generally more labor intensive than research operations. Projecting baseline labor costs forward per fte (perhaps using extension personnel data) and making suitable adjustments for anticipated noncapital costs and changes in personnel could generate an acceptable cost series for extension. Economic Surplus Measurement and Application 429 A5.4.3 Research Impact - Estimating kMAX The effects of research on shifting supply functions - assuming full adoption of the results, the /(.MAX parameters - can be estimated using experiment or industry data for ex post analysis. In ex ante studies, it is necessary to use estimates of the effects of hypothetical research alternati ves; past experiment or industry yield data can be used to ensure realism in the elicitation of those estimates. Aggregate Industry Yield Effects In order to estimate a value of JcMAX for use in computing research benefits, it is necessary to define two scenarios between which yields (or costs) can be compared. A baseline scenario is commonly defined as a basis against which one or more alternative programs are to be compared; the JcMAX estimate refers to the shift in supply from the baseline due to the difference in research expenditure under the alternative. For ex post evaluation of past programs, studies have treated the actual program as the baseline and the alternative is a counterfactual scenario of no expenditure on the research program of interest. This wiII yield a measure of average returns. For forward-looking analysis, actual expenditure is not known and the baseline is hypothetical, along with the alternatives. A reasonable approach is to project the current (or recent past) forward as the baseline and to consider variations from that baseline (say plus or minus 15%) as relevant alternatives. This latter approach will generate an estimate that is closer to a marginal than an average rate of return. For forward-looking analysis, for both the baseline and alternative scenar­ ios, scientists and others are usually asked to estimate yields if the research is successful and the results are fully adopted, and the estimated yields are used to deduce measures of /(.MAX. It is likely that elicited estimates will be more accurate if past yields are used to benchmark the projections. One approach is to graph past yields and ask scientists and others to juxtapose their estimates of "lowest," "most likely," and "highest" future yields - given a particular research program alternative - against the historical record, as shown in figure A5.4.1. In figure A5.4.l, information on a single research option is shown. Alternatively, and perhaps preferably, a table such as table A5.4.5 could be used, in conjunction with the historical yield plot in figure A5.4.1, to elicit information on the distribution of potential outcomes from each of several research alternatives. A moving average of yields over several recent years may be used to adjust for weather effects. Alternatively, the elicitation form might be constructed to simultaneously collect additional information, such as research lags, on a single research alternative. 430 Economic Surplus Measurement and Application Figure AS.4.1: Benchmarking elicited yield effects for a single research option Commodity: AEZlRegion: Research scenario: .4 highest most likely x lowest x x .0- x x x XX x x .6- x x x x x x 2 .2 x x x x x .8 1970 1980 1990 Current R&D lag Table AS.4.S: Sample Data Sheetfor Recording the Yield Effects of Various Research Options Commodity: AEZl Region: Research Yield Scenario 1.0. No. Lowest Most likely Highest (t/ha) (t/ha) (t/ha) Baseline 1 No research 2 Baseline + 15% 3 Yield gain Research Alternative Lowest Most likely Highest (%) (%) (%) I vs 2 I vs 3 Estimates of industry yield effects might also be based on experimental data. In such cases, care must be taken in the elicitation to contrast experi­ mental and industry yields and to account for differences in experimental and industry conditions, cultural practices, and input use. Economic Surplus Measurement and Application 431 Factor Bias and Additional Input Costs As discussed in section S.3.2 and in box S.l, the elicited estimates of increases in industry yields identified in table AS.4.S may involve a change in the cost of purchased inputs (such as fertilizer, fuel, and pesticides) or a change in the use of allocatable fixed factors (such as land or operator labor) per tonne of output. Such induced changes in input costs must be accounted for when elicited yield changes are translated into measures of the per unit cost reductions attributable to research (i.e., J10 Research program area S;l 2 3 4 5 6 7 8 9 10 (specify) I Plant breeding Plant protection I Soil management & fertilizer I Crop production practices Other --------------------------------------------------------------------------------------------------------------------- Postharvest ~erall commodity _ ---_._---------- ---- Economic Surplus Measurement and Application 435 Adoption Parameters Before attempting to elicit parameters describing the time path of adoption, it is usually necessary to assume a particular type of adoption process. Elicita­ tion forms are given below for parameterizing two typical adoption profiles. These curves include the growth phase (during which technologies are taken up) and the decline phase (during which technologies depreciate or become progressively abandoned). Figure A5.4.2: Trapezoidal adoption profile Adoption rate 100% ----------------------------------------------------- AM~ __________________ ~--------------~ '---v---''---y----'~~-----.,v------~~ Years AR AA AM AD 436 Economic SurpLus Measurement and Application Table A5.4.8: Sample Data Sheet for Adoption Parameters for the Trapezoidal Adoption Profile Adoption parameters Period from Total Period from maximum Period from the research and Ceiling initial adoption adoption to beginning to the adoption lag, adoption rate, to maximum eventual end of the A.T= A.R Commodity AMAX adoption, A.A decline, AM decline,A.D + A.A + A.M + A.D (%) (years) (years) (years) (years) Note: The research lag, A.R, is obtained from table A5.4.7. The ceiling adoption rate, AMAX, is best measured as the proportion of output produced using the new technology, but it is usually more readily approximated as the maximum area sown to a new crop or cropped area produced with a new technology, or the maximum proportion of farmers adopting a new technology. See appendix A5.1.2 for details on calculating the trapezoidal adoption profile At, t = 0 , ... , LR using the data in this table. Economic Surplus Measurement and Application 437 Figure A5.4.3: Logistic adoption profile Adoption rate 100% --------~---------------- ,, , , , , , , , , , , , , , , , ... v Table A5.4.9: Sample Data Sheet/or Adoption Parameters/or the Logistic Adoption Profile Adoption parameters Period from initial Period from initial adoption to attain 50% of Ceiling adoption adoption to maximum maximum adoption, Commodity rate, AMAX adoption, AA AO.5A MAX (%) (years) (years) Note: See section 5.3.3 and the references cited therein for details on constructing a logistic adoption curve from these data. The latter part of the adoption profile can be approximated using the AM and AD parameters from table A5.4.8. 438 Economic Surplus Measurement and Application A5.4.5 Research Risk Mean or Most Likely Effects Table AS.4.5 is suggested for eliciting three values -lowest, most likely, and highest - to parameterize the potential distribution of yield outcomes for a given research scenario. This information could be used in a number of ways. First, by assuming a particular functional form for the probability distribution - we recommend triangular - an expected value can be deduced (a variance and other moments could be computed as well). For some studies, this expected value may be all that is required: the expected value for the triangular distribu­ tion is equal to the simple average of the three elicited values. In addition, however, the triangular distribution contains information that can be used with the methods described in section S.3.3 to jointly define the expected value of the yield, given a successful research program, and the probability of achieving success defined in terms of a particular yield outcome or better. Figure AS.4.4 shows the triangular probability density function elicited for a particular research scenario. Figure A5.4.4: Example of a triangular probability density function Economic Surplus Measurement and Application 439 Sensitivity Analysis The availability of data on distributions of research outcomes is intrinsi­ cally valuable because it conveys information on the degree of dispersion around the scientists' estimates of the most likely research outcome - a measure of their confidence - and it is convenient for conducting informal or formal sensitivity analyses. For instance, section 5.4.4 discusses the use of a triangular distribution in a Monte Carlo simulation for sensitivity analysis. Also, some studies might want to take explicit account of research risk in the objective function (section 5.4.5), which can be parameterized using the variance from the probability density. It is important not to confuse the sensitivity analysis, which measures variations in the effects of a given program of research, as discussed here, with measures of the effects of variations in research programs. The former holds the research program constant; the latter holds everything else constant and varies research. A5.4.6 Reconciliation The purpose of the elicitation is essentially simple: we want an estimate of the time path of research-induced supply shifts (i.e., k,) for each program alternative, which can be used to compute corresponding benefit streams using the methods described in preceding sections of this appendix. And for both evaluation and priority-setting purposes, estimates of the streams of research costs that correspond directly to the benefit streams (i.e., that are consistent with the underlying k, estimates) are required. But estimating these parameters is not easy. It involves combining esti­ mates of underlying parameters. Several hazards arise. Each component is uncertain, and errors in one can corrupt the whole. And scientists might knowingly or inadvertently provide biased estimates. In some cases the biases may be offsetting, but in others they will not be - in particular, scientists seem more likely to overestimate research impacts than to under­ estimate them, both because they are too optimistic and because they have a stake in a favorable analysis. Perhaps the best that can be done is to attempt to minimize the hazards by two expedients. First, since scientists compete with one another for re­ sources, peer review at the elicitation stage could provide useful checks on scientists' estimates, and in an ongoing program of research evaluation and priority setting, monitoring actual research performance against initial claims can lead to increased incentives for accuracy. Second, scientists (and others involved in the elicitation) should be pressed to be sure that the various parameter estimates are mutually consistent: is the expected yield gain from a particular program of research consistent with research program 440 Economic Surplus Measurement and Application expenditures and the R&D (and adoption) lag'profiles that have been tabu­ lated? This involves collecting all of the information together at the end of the elicitation and reviewing it. Such a review could involve not only the original sources of the estimates, but also other knowledgeable people, and it could use information available from other sources, such as studies con­ ducted elsewhere or earlier in the same place. 6 Mathematical Programming The research evaluation methods described in chapter 5 can be used to provide information on research benefits by program, as well as a ranking of program alternatives. However, these methods do not indicate the amount and share of total resources to allocate to each program. In chapter 5 we examined how estimated NPVs can be used in informal decision-making processes. Now we examine ways to use the same information - on the estimated benefits and costs, objectives, and resource constraints - in more formal modeling and analysis of resource allocation decisions. Optimization subject to constraint is a fundamental part of economics. Results have been derived to characterize the solutions to general classes of problems and to identify the important features of those solutions. Several procedures are available to obtain optimal numerical solutions. The purpose of this chapter is to provide some guidance about the use of some of those procedures for optimizing research program portfolios. Which procedures are most appropriate depends on the characteristics of the optimization problem in terms of (a) the objective function to be maximized (i.e., some function of the costs and benefits of research, and perhaps other impacts), (b) the relationship between changes in research activities and the value of the objective function (i.e., combining the research production function[s] that relate research output to operating expenses and other inputs and the functions that relate research output to its economic impacts), and (c) constraints (on total resources avail­ able for research, on particular inputs, and on the research portfolio itself). Specific situations sometimes satisfy general conditions for optimality, obviating a need for using empirical optimization methods. For instance, the unconstrained maximization of the total anticipated social payoff from 441 442 Mathematical Programming research would call for funding all research programs having positive antic­ ipated NPVs. Maximization of the same objective subject to a fixed research budget would call for choosing the portfolio that maximizes the anticipated NPV per unit of constraint. This portfolio could be found by ranking pro­ grams according to NPV per unit of budget and moving down the list until the budget was exhausted. But this method for allocating resources to research only works for a set of discrete alternative investments. Suppose the size of individual research programs is flexible. If there is a linear relationship between NPV and program size, the benefits from the total portfolio will be maximized by allocating all resources to the program with the highest NPV per unit of constraint - a comer solution. An interior solution (i.e., a diversified research portfolio) requires either (a) multiple constraints on inputs to research programs and different input requirements for different programs (i.e., a linear programming problem), (b) a nonlinear (diminishing­ returns) relationship between the sizes of programs and their NPVs, (c) a nonlinear relationship between the NPV of a program and its contribution to the objective function (where multiple objectives are involved), or (d) some combination of these features. In all of these situations, an explicit optimization procedure is needed to establish the empirical trade-off among programs where their marginal contributions to the various relevant objectives are equated. Mathematical programming is an optimization procedure that can be applied to such problems. I A mathematical-programming model can • include multiple objectives and be used to quantify the nature of trade-offs among objectives (e.g., the economic efficiency sacrificed to meet a distributional objective) • incorporate a research response function that exhibits constant or diminishing returns to research so that, for a given objective, the mix of research programs can be optimized • relate the marginal research benefit to the amount of funds going into research and their deployment • examine the implications of changing facility, human resource, and financial constraints on research • identify both short- and long-run priorities by considering changes in constraints on resources that may be fixed in the short run but variable in the long run I. Several texts are available that document the theory underlying mathematical programming models. their practical application. and the computer programs that can be used. These include Hazell and Norton (1986) and Paris (1991). Simulation models have also been proposed to assist with agricultural research prioritization (Pinstrup-Andersen and Franklin I Cf77 and Bosch and Shabman 1990). These models lend themselves to building in risk components and mayor may not include an optimization algorithm. Mathematical Programming 443 • provide information on the benefits foregone due to short-run fixities in human resources and facilities • examine the sensitivity of research priorities to estimated changes in research funding, market conditions, per unit cost reductions due to research, and other assumptions2 In this chapter we discuss the use of mathematical-programming models in making decisions about allocating resources to research.3 We examine issues to consider in model design and suggest possible formulations of a multiple-objective programming model for allocating research resources. Mathematical programming has its greatest potential for assisting with research resource allocation when it is combined with measures of benefits derived from economic surplus analysis. Hence, a decision to apply a programming procedure in this context implies that adequate resources are also available for implementing the economic surplus approach. Computing the measures of economic surplus changes required for the mathematical­ programming models described below is no small task. However, once the economic surplus calculations have been made, a mathematical-program­ ming model offers the possibility of utilizing the information more effec­ tively. Even if formal optimization is not undertaken, thinking through such models can provide additional insights into the problem of allocating re­ sources to research. This chapter begins with a review of the basics of mathematical-program­ ming models. Then it considers some specific aspects involved in applying those models to agricultural research portfolios. The third main section goes into practical implementation issues, including data, solution procedures, and possible extensions. 6.1 Mathematical-Programming Principles 6.1.1 Basics of Mathematical-Programming Models Several variants of the basic multiple-objective, mathematical-program­ ming model are available for obtaining a weighted "optimal" solution or a set of feasible solutions that trade off the various objectives. The basic 2. Mathematical progranuning does assume separability among activities in the model, and the analysis is partial equilibrium like the methods described in chapters 3 to 5. 3. Mathematical programming models have been fonnulated for agricultural research resource allocation by Russell (1975), de Wit (1988), and Scobie and Jacobsen (1992). Russell applied his model to a set of research projects in the United Kingdom; de Wit applied his to a hypothetical set of data for the CGIAR system. Scobie and Jacobsen provide guidance for allocating research resources across research programs supported by the Australian Wool Corporation. 444 Mathematical Programming multiple-objective decision-making model can be represented in its general form as max z(x) = G [ z\(x), Z2(X), ... ,zix) ] (6.1a) subject to XE X (6.1b) X~O (6.1c) where z(x) is the objective function with k objectives, G is the goal operator (which defines the functional form of the objective function being maxi­ mized), x is the n-dimensional vector of decision variables (i.e., in this context these will be the research programs to which resources are commit­ ted to achieve the stated objectives), and X defines the decision space (in this context, defined by the set of research resources available and any other constraints on the choices of x) so that equation 6.lb is the set of m constraints for the problem. Equation 6.1c is a set of non-negativity condi­ tions that constrain the problem so that the values of the decision variables or activities (e.g., in this context the research resources invested in any program) cannot be negative. Together, equations 6.1b and 6.1c determine the feasible region. Each feasible solution implies a value for each objective Zj(x), i = 1,2, ... ,k.4 Basic Solution Approaches The two basic means of solving this general model are (a) to define and apply a set of decision-makers' preferences or weights before optimization, so as to obtain a unique "optimum" solution, or (b) to generate a set of non-inferior solutions that illustrate the tradeoffs among objectives rather than provide only a single optimum solution (the decision makers must choose then from a, sometimes large, set of possible solutions).5 In the latter approach, non-inferior solutions are generated without prior specification of preferences by parametric variation (varying by increments) of either the weights on the objective function or constraints on the solution (figure 6.1 ).6 This more generally adopted approach, of varying the weights or constraints, amounts to defining empirically the benefit transformation curve or surface which shows how the (maximum) value of the objective function varies with changes in activities (i.e., how the total research benefit varies with changes in combinations of research programs - chapter 2). The first approach 4. This fonnulation of the problem follows Willis and Perlack (1980). 5. A noninferior solution is a feasible solution to the problem, x E X, such that no other feasible solution, x* E X, exists for which Zp(x*) > Zp(x) for some p = I, 2, ... , k, and Z j (x*);:: Z j (x) for all i '# p. 6. Cohon and Marks (1973), Cohon (1975), and Willis and Perlack (1980) compare these techniques. Mathematical Programming 445 Figure 6.1: Illustration of a set of noninferior solutions generated by parameterizing weights on the objective functions Contribution to objective 2 set of non-inferior solutions inferior solutions Contribution to objective I amounts to specifying the slope of a particular indifference curve and the constraints, and then finding a point on the benefit transformation curve. Goal-programming approach: Following Willis and Perlack (1980), one formulation of the problem, called goal programming, involves specify­ ing weights, Wi' and positive deviations, d;, and negative deviations, e;, ofthe objectives from their targets, T;. In the case of research resource allocation, the targets might represent the maximum value that would be achieved if all research resources were devoted to satisfying a particular objective.? k min L W; (d; + e;) (6.2a) i=1 subject to XE X (6.2b) X~o (6.2c) z;(X) - d; + e; == T;, i == 1, ... , k (6.2d) (6.2e) ? Chames and Cooper (1961) provide an early discussion of the fonnulation of the goal-program­ ming model and Kornbluth (1973) surveys the subsequent literature. Lee (1972) and Neely. North and Fortson (1977) provide examples of applications. 446 Mathematical Programming In goal programming, the solution is constrained to minimize weighted deviations from the goals but is not constrained to achieve the goals. The Wi are the penalties attached to deviating from the targets, and if di is nonzero, ei will be zero, and vice versa. For instance, if there were two objectives, efficiency and equity, one penalty would be attached to any deviations from maximum efficiency benefits and another to any deviations from maximum equity. The solution to the model would minimize the sum of those weighted deviations. Parametrically varying weights in the objective function: The formu­ lation of the multi goal problem in which G is linear and weights on the objective function are not specified a priori but are parameterized to generate a set of non-inferior solutions can be represented by k max I. WiZi(X) (6.3a) i=1 subject to XE X (6.3b) X~o (6.3c) where Wi ~ 0 for all i and is strictly positive for at least one of the Wi s. The initial Wi s are arbitrarily set and then varied parametrically. Parametrically varying constraints on the solution: The formulation of the problem in which weights on the objective function are not specified but constraints on the solution are varied by the analyst to generate a set of non-inferior solutions can be represented by max Zj(x) (6.4a) subject to XE X (6.4b) Zi(X) ~ bi , i"#j (6.4c) x~O (6.4d) where bi are the lower bounds on the k-l objectives. The lower bounds are set by maximizing equation 6.4a for each of the k objectives individually, subject to equations 6.4b and 6.4c, substituting the values of x for each of the k optimal solutions into z;(x), and then selecting for each Zi (x) the lowest of its k values to be its bi' The set of noninferior solutions is then generated by solving equations 6.4a to 6.4d with parametric variation of bi and substitution of each Zi(X) into equations 6.4a, for all i "# j. In essence, this is goal programming as well, because targets for each goal are being set. Mathematical Programming 447 Several comparisons can be made among these three alternatives. First, the approach in which weights are specified a priori can be converted to the approach in which weights are varied parametrically.s Second, the formulation in which weights on objectives are varied gives the same noninferior solution set as the formulation in which constraints are varied, as long as the objective space is strictly convexY Third, the formulations vary in degree of quantifica­ tion of trade-offs and in the ease of presenting results to decision makers. Willis and Perlack (1980) compare these and other criteria for evaluating these approaches. They note that, if the weights were varied six times and if there were four objectives, the set of noninferior solutions would be 216 (and 1,296 for five objectives). They argue that the parameterizing approach is likely to be impractical for more than four objectives. However, it is likely that the 36 solutions generated by only three objectives also provides too much informa­ tion to be of much help to research decision makers. Hybrid Programming Approaches Compromise programming: A hybrid approach can narrow down the noninferior set of solutions. Various hybrid approaches have been suggested in the literature. One is based on a technique called compromise programmingYI With compromise programming, the ideal point, the coordinates of which are gi ven by the optimum amount for each objective in isolation, is established first (figure 6.2). However, because of conflicting objectives, this ideal point may not be feasible. Compromise programming then defines the best compromise solution as the feasible solution that is closest to the ideal point. Closeness is measured by a weighted sum of deviations from the maximums (ideals) for each individual objective. Closeness for an individual objective is measured by dj = z/(x) - Zj (x), where z/(x) is the ideal value for objective}." Compromise programming uses the function (6.5) to measure the distance between each solution and the ideal solution. In this formulation, each Wj weights the importance of the difference between the 8. See Willis and Perlack (1980). 9. Cohon and Mari<:s (1973) and others have used the possibility of inferior solutions as an argument against parameterizing weights in the objective function. While concave portions of the objective space seem unlikely for the objectives described earlier for the model, strict convexity might be violated. 10. See Zeleny (1973), Cohon (1975, 1978), and Romero, Amador and Barco (1987). II. See Romero, Amador and Barco (1987) for more details. If objectives are in different units, the dj can be converted to a proportion of the maximum possible deviations from the ideal for each objective. When thejth objective is minimized, dj = zj(x) - Zj (x). 448 Mathematical Programming Figure 6.2: Trade-off between objectives and illustration of the ideal point and compromise set Contribution to objective 2, z2(x) zi(x) __,-, -:...:-~-- - - - - - - - - - - - - - - - - - - - - - - .... Ideal point I Illustrative zj(x) Contribution to objective I, zl(x) jth objective and its ideal value, and p is a parameter that is set equal to either one or infinity in order to bound the solution. Bounds can be placed on the solution set by first setting p = 1.0 such that the following linear program­ ming problem is solved: k min LI = 1: Wj [Zj (x) - Zj (X)] (6,6a) FI subject to XEX (6.6b) X::?:O (6.6c) and then, by setting p = 00, the maximum of the individual deviations is minimized (i.e., only the largest deviation counts and the algorithm mini­ mizes it) to obtain the other bound min C=d_ (6,7a) subject to WI [zj(x) - ZI(X)] :s; d_ (6,7b) Mathematical Programming 449 XE X (6.7c) X~O (6.7d) The compromise programming procedure narrows down the efficient set, but it does so at the cost of requiring the weights on the deviations from the ideal to be specified. Furthermore, the set of efficient solutions could still be large. A filtering technique could be used to remove some of the solutions that are not very different from the other efficient solutions already calcu­ lated, but the problem of weighting deviations would remain. 12 Iterative procedures: A more practical approach to the multiobjective, decision-making problem facing research decision makers may be to com­ bine their opinions about weights on objectives with the generation of a reduced set of solutions and to follow an iterative procedure. 13 First, weights could be elicited from the decision makers for the various objectives using the procedure described in chapter 7, and equation 6.3a could be maximized subject to (6.3b) and (6.3c) to generate an initial solution. At the same time, equation 6.4a could be maximized for each of the k objectives subject to (6.4b) and (6.4c). This would generate the maximum and minimum bounds for each objective. These k + 1 solutions would be shown to the decision makers who then would be asked if they would like to change their weights on any objectives, given the trade-offs illustrated by the solutions used to generate the bounds. If they revise their weights, the model is rerun with the new weights and the process is repeated. An alternative iterative procedure would be to generate the bounds but not to elicit the initial weights. Instead, the k solutions would be shown to the decision makers who would be asked the objective for which they would most like to see the lower bound raised. The model would be rerun after raising that lower bound and at least one of the objectives would achieve a lower maximum than before. This process would be repeated several times until the decision makers were satisfied with the trade-offs made among the objectives. If the process were entirely iterative with no specification of initial weights, it might require a larger number of iterations to arrive at a similar solution than if initial weights were elicited. The final, and perhaps the recommended, option would be to begin by generating a benchmark solution (e.g., maximize equation 6.3a subject to [6.3b] and [6.3c] with a weight of 1.0 placed on the efficiency objective and zeros placed on other objectives). Then, a weight of say 0.1 could be added to each of the other objectives. These two solutions could be presented to 12. See Steuer and Harris (1980) and Romero, Amador and Barco (1987) for a discussion of a filtering technique. 13. See Candler and Boehlje (1971). 450 Mathematical Programming decision makers. Their views could then be solicited on any desired changes in weights. We recommend this option because (a) it is difficult to discuss and evaluate with decision makers more than two or three alternative re­ search portfolios at once, (b) research is a relatively blunt instrument for meeting nonefficiency objectives and, hence, an initial situation with all weight placed on economic efficiency is a reasonable base for comparisons, and (c) in practical experience with eliciting weights in several agricultural research systems, decision makers experimented with different weights in an informal process until nearly all the weight was placed on the efficiency objective (chapter 7). This prior information can be used to provide a useful point of departure for reaching consensus on a research portfolio. 6.1.2 Formulations/or Research Resource Allocation We begin with a simple model in which (a) research resources are allocated across alternative programs under the assumption that increased economic efficiency is the only objective, (b) the research production func­ tion is nonlinear, exhibiting diminishing marginal returns (e.g., Scobie and Jacobsen 1992), and (c) there are n research programs and three levels of spending for each program. The model is then modified to incorporate objectives associated with income distribution and risk. Finally, multiperiod optimization is added. The basic structure of the model is described in figure 6.3. We begin by presenting a single-period,linear-programming model. The research benefit per unit of research input is a discounted sum of economic benefits for several years (although the benefits would accrue over several years, they enter in present value form only once, creating a one-decision-period prob­ lem). These discounted totals per unit of research input are allowed to vary with changes in research expenditures so that the research production func­ tion exhibits diminishing returns to research inputs. Maximizing Total Research Benefits The activities in the model, i.e., the cij' are research program alternatives, where i =t he program and j =l evel of funding. For example, the activities Cll' C!2' and Cl3 in figure 6.3 represent aggregate research program 1 under three ranges of financial support. The constraints, Rij, are the maximum funding levels for each program, i, and range of support,j. For example, Rll , R1 2, and R 13 are the three altemati ve maximum levels of funding for program 1 (Rll < RI2 < R I3 ). This support includes money for facilities and personnel as well as operating costs. In addition, total expenditures on all the research Figure 6.3: Single-objective, linear-programming model/oral/ocating resources to agricultural research ~ Research programs Commodity I Commodity 2 Commodityn Equation Description ell el2 e13 ell ell el3 enl enl enJ RHS I objective function all al2 a13 all all a23 ani anl an3 MAX 2 total resource limit I I $R 3 program resource limit $Rli $RI1 $R13 $Rl i $Rl2 $R13 $Rnl $Rn2 $Rn3 452 Mathematical Programming programs must be less than R (it is not possible to spend beyond the total research resources on any individual program), and the sum of all maximum individual Rijs must be greater than R (it must be possible to profitably spend up to the total available) - i.e., R is a binding constraint on total research spending so that 1:.; cij::; R ::; 1:.; Ri). The units for the activities are units of research expenditures (dollars or some other currency). For example, for program C I, a unit of CI2 might be a dollar of research expenditure for spending up to the current level of research spending, a unit of CII would be a dollar of research expenditure for the low-expenditure option (say up to 75% of current), and a unit of C13 would be a dollar of expenditure for the high-expenditure option (say, up to 125% of current program expenditures). The three levels of support could vary independently by program; the options considered should correspond to the levels of funding assumed when the research benefits were calculated. The all' a 12 , ... , an3 represent the contributions to the efficiency objective (discounted economic benefits) associated with the n research programs and their corresponding levels of support. For example, under the low range of research support for program 1, all represents both the average and marginal products of research in terms of NPV per dollar of research spending, and these average and marginal products are equal within the low range of support. Under the intermediate range of research expenditures (i.e., RII < CI ::; R 12), a 12 is the marginal product of research measured in NPV per dollar for expenditures between RII and R12 • In this range, the marginal product is less than the average product because it is lower than the initial marginal and average product. Because all> al2 > a l3' the three levels of research benefits per unit of input represent a three-step, research-response function with diminishing marginal returns. l4 The model as formulated requires that suc­ cessive increases in spending on any research program result in smaller gains to productivity (i.e., it requires that the research production function for each program be concave). Also, if, when questioning scientists, it appears that current levels of research spending are in the range of increasing returns (i.e., all < a 12), additional opinions can be solicited from those scientists about the implications of higher spending to obtain information on the diminishing returns portion of the function. Research support for each commodity cannot exceed the maximum for the highest-cost program for that commodity (i.e., Ri3)' The program resource limits in figure 6.3 force the model to move up the research production 14. While research-response functions may exhibit increasing, constant, or diminishing returns, the assumption of diminishing returns is probably reasonable in most situations because it is a commonly held view that "nature is increasingly niggardly." Mathematical Programming 453 function for a commodity, once a specified resource limit is reached (e.g., from CII to C l2 after Rll is reached and from Cl2 to CI3 after RI2 is reached). Because R < (RI3 + R23 + R33 + ... + Rn3), the research system may not move up to the high level of research spending for some (or even any) of the commodity research programs. Multiple Objectives The model in figure 6.3 can be extended to include multiple objectives as illustrated in figure 6.4. For ease of exposition, we assume that the research decision makers identify two objectives: (a) an efficiency objective, to maximize the well-being of all citizens and (b) a distributional objective, to provide additional benefits to a particular group (e.g., small-scale farmers). The all' a12, ... , an3 represent the marginal contributions of changes in research expenditure to the efficiency objective; bll , b l2 , ... , bn3 represent the same contributions to the distributional objective. The model contains an objective function with an initial weight of WI =1 .0 placed on the efficiency objective, ZI(X), and a weight of W2 placed on the distributional objective, zix). The NPVs for each objective are summed and transferred to the objective function for weighting. Portfolio Risk Since the early work on portfolio analysis by Markowitz (1952) and the quadratic programming model suggested by Freund (1956), several different approaches have been suggested for incorporating risk into mathematical-pro­ gramming models. Many of these procedures are reviewed by Anderson, Dillon and Hardaker (1977), Boussard (1979), and Hazell and Norton (1986). Over the past 20 years, the two most common approaches for incorporat­ ing risk into agricultural applications of mathematical-programming models have been quadratic risk programming, in which a variance-covariance matrix of net returns is incorporated in the model, and the MOTA D approach, in which the mean of the total absolute deviations is minimized. 15 Another technique is the focus-loss or safety-first approach that is designed to limit the "risk of ruin" by establishing some maximum admissible loss or mini­ mum income level. Mathematical-programming models for allocating re­ sources to agricultural research can take advantage of this vast literature when a procedure for incorporating risk is selected. In chapter 5 we presented a Monte Carlo approach for deriving measures of dispersion of estimated NPVs for particular research programs (e.g., Scobie and Jacobsen 1992). 15. For risk in a quadratic programming model see Freund (1956) and for risk in a MOTAD linear programming model see Hazell (1971). Figure 6.4: Multiple-objective, linear-programming model for allocating resources to agricultural research ~ ~ Research programs Objectives Commodity 1 Commodity 2 Commodityn Equation Description XI X2 Cll CI2 CI3 C21 C22 C23 Cnl Cn2 cn3 RHS objective functions W2 MAX 2 contributions to objectives -I all al2 a13 a21 a22 a23 ani an2 a.3 =0 -1 bll b12 b13 ~I ~2 ~3 bnl bn2 bn3 =0 3 total resource limit I 1 I I I I ~R 4 program resource limits ~ Rll ~ R12 ~ R13 ~ R21 ~ R22 ~ R 23 ~ Rnl ~ Rn2 ~ Rn3 Mathematical Programming 455 Production and Income Risk Production or income risk can be influenced by research in two ways. First, some commodities are inherently riskier than others in terms of yields and prices in a particular location. Therefore, research that in itself does not affect yield or price risk can still influence production or income risk simply by altering the relative amounts of different commodities produced. Second, research itself may be directed at influencing yield or price risk (e.g., it may produce a new pest-management practice that lowers the risk of pest infes­ tation; it may produce a more drought-resistant crop variety; it may suggest a policy change that reduces price risk). When policymakers speak of a social objective of reducing risk, they usually are referring to a desire to reduce these types of production or income variability and not to the risk associated with the research process itself or with the adoption of research results. Their concern for production or income risk is often derived from a concern for food security. In principle, the ultimate influence of research expenditures on the variability of national agricultural income could be represented by a joint probability distribution that reflects each type of risk. 16 However, in practice, if the public research investment is small relative to total income in the country, as argued above, it may be sufficient, when reduced risk is a social objective, to include in the model the effects of research on production or income risk. The security goal, particularly as related to the objective of reducing income risk, has been incorporated into numerous mathematical-program­ ming applications, at least for farm-level applications. Other security objec­ tives of research, including sustaining environmental quality, are more difficult to quantify but could also be included. 17 Environmental research benefits are typically multidimensional, and at least some include nonmarket activities, which implies that extensive and relatively costly data gathering and analysis would be required in order to generate the environmental NPVs required for the environmental objective. The model shown in figure 6.4 could be extended to include a social objective of reducing the income risk associated with the variation in pro­ duction and price. A variance-covariance matrix could be added, thus con­ verting it to a quadratic programming model. Alternatively, the linear nature of the model could be maintained by converting it to a MOT AD representa­ tion l8 in which mean absolute total deviations are incorporated to approxi- 16. Finkelshtain and Chalfant (1991, 1993) discuss some issues that arise when dealing with multivariate risk. 17. See, for example, Miller and Byers (1973). IN. See Hazell (1971). 456 Mathematical Programming mate the effect of research on the income risk of production. For the portfolio of commodity research programs developed, inclusion of this risk compo­ nent implies that research will be skewed toward increasing the production of those commodities with lower production and price risk and away from those with higher risk. 19 One difficulty, however, with incorporating income risk is that the easiest income-risk measures to calculate may capture income risk to producers but not to society as a whole. For example, it is relatively easy to calculate a variance-covariance matrix for producer gross income using historical data. Production and price risk may be negatively correlated, however, so that low production years coincide with high price years, reducing the income risk compared with that implied by production variability alone. A second complication is that measured variability in income associated with production of any particular commodity must be translated somehow into a measure of research-induced reduction in total risk. This involves paying attention to the effects of research in changing the mix of commodi­ ties produced (not necessarily a simple task) and also adjusting, where appropriate, for covariance effects among commodities in the national port­ folio. Finally, it must be noted that it is not clear what meaning should be attached to this particular measure of research-induced risk reduction. A more relevant risk measure might be risk as perceived by individual farmers, an entirely different concept than variability of total income to the agricul­ tural sector. Or, the government might be interested in the risk of famine, a concept that may bear little relationship to the variance of gross agricultural income. Thus, it is much less straightforward to include risk in a mathematical­ programming model for allocating resources to research than it is to incor­ porate the efficiency and distributional objectives - primarily because of conceptual and measurement problems related to the measure of risk and the effect of research on it. 6.2 Mathematical Programming In Practice Applying a mathematical-programming model for research resource allo­ cation involves four steps: (a) designing the mathematical-programming model, (b) compiling information and calculating the coefficients in the model, (c) running the model, and (d) using the results to develop plans. 19. If the linear programming model were structured around disciplinary programs aimed at one or several commodities, a different approach would be required that would incorporate the effects of specific types of research on risk. Mathematical Programming 457 6.2.1 Model Design Designing a mathematical-programming model of an agricultural re­ search portfolio involves choosing (a) the research programs to include as activities in the model (e.g., commodity- or noncommodity-based research programs, components within research programs, or research projects), (b) the objectives to be incorporated in the model (e.g., economic efficiency, income distribution, security, and research portfolio risk), and (c) the partic­ ular procedure to use for weighting the objectives and for generating trade­ offs among the objectives. Research Programs (Activities) The research program alternatives to be considered must be defined well before the stage of optimization in order to obtain measures of the perfor­ mance of the alternatives, as discussed in chapter 5. The research programs could be defined according to the broad commodity aggregates (e.g., crops and livestock) or to particular commodity or noncommodity subjects to which the programs of research are directed. The degree of aggregation, or number of potential activities, is not constrained by the procedure itself. For example, Russell (1975) designed a model for selection of research projects; Scobie and Jacobsen (1992) focused on broader program areas defined by the Australian Wool Corporation (A We) within wool research funded by the AWe. The real constraints on the choice of activities to be included in the model are (a) the types of decisions the analysis is intended to support and (b) the quality and quantity of meaningful information that can be obtained on the economic effects of the alternatives. It often becomes difficult to get mean­ ingful measures of research benefits and costs as programs are disaggregated (chapter 5). For strategic decisions on agricultural research resource alloca­ tion, it may be necessary to structure activities around commodity programs and components within commodity programs (e.g., plant breeding in rice, plant breeding in corn, or sheep nutrition). However, at a practical level, in a NARS or other large research agency, it may be too expensive to compile the information needed to evaluate such highly disaggregated research pro­ grams when defining the activities in the model. Consequently, in a typical developing-country NARS, it may be reasonable to structure a model around aggregate commodity research programs. A concern with, say, regional priorities, might be addressed by structuring the same type of model around program components within a limited set of individual commodity pro­ grams. 458 Mathematical Programming Objectives In chapter 5 we discussed practical approaches to measuring a research program's contributions to three of these objectives - efficiency (NPV of total domestic economic surplus), distribution (NPV of economic surplus accruing to particular domestic groups), and total portfolio risk (variance of the NPV oftotal surplus). Here we review those measures and alternatives for use in mathematical-programming models. Efficiency: The objecti ve of improving the average level of well-being in society can be represented by net total research benefits in terms of net present value, measured using the economic surplus and capital-budgeting techniques described in detail in chapters 4 and 5. Measuring improved economic efficiency provides a yardstick against which the trade-offs asso­ ciated with attempts to maximize other objectives can be measured. Changes in the net present value of economic surplus due to research can also indicate the efficiency costs associated with alternative research resource allocations, even when efficiency is the only objective. Distribution: The goal of improving the well-being of particular groups in society may be represented by one or by several objectives, as described in earlier chapters. Examples of distributional objectives include the im­ proved well-being of producers on small farms or of people living in certain regions.20 Measures of the contributions of alternative research programs to these objectives can be calculated using measures of the net present value of the economic surplus accruing to the particular group(s) as discussed in chapters 4 and 5. These are not the only income-distribution objectives that may be relevant, and where other effects, such as benefits to the urban poor, are to be emphasized, it will be necessary to take a measure of those effects for each research program. Research portfolio risk: The return from any particular research portfo­ lio will have an expected value and a variance associated with it. The expected return and variance for the complete portfolio depends on the expected value and variance of research benefits (NPV of economic surplus) per unit of research for each research program. It also depends on how the research budget is allocated across programs. As discussed in chapter 2, in most cases, an agricultural research system should be risk neutral and, hence, research portfolio risk can be ignored. However, as discussed in chapter 5, unless the distribution of expected benefits is symmetric around the most likely value (mode), it may be necessary to elicit information on more or less likely values to obtain a good estimate of the mean effect of research. Also, 20. Cohon and Marks (1973) provide an example of a mathematical-programming model in which regional income is traded off against national income. Mathematical Programming 459 decision makers might like to know if two research programs or research portfolios with similar expected economic benefits differ widely with respect to the range of possible outcomes, even if they make risk-neutral decisions. 21 Security: The risk associated with research - variability of returns to the research portfolio - must be distinguished from the risk associated with producing particular commodities. Changes in the research portfolio may affect both these types of risk. If one objective for the research system is to reduce production or income risk, that objective may be included with a weight that reflects the importance of reducing risk. However, the compara­ ti ve advantage of research as opposed to other policy instruments (e.g., crop insurance, investing in infrastructure such as irrigation) in reducing risk or mitigating its effects should also be considered. Weighting Objectives Once the relevant objectives and the quantitative indicators of the contri­ butions of research to attaining these objectives have been defined, proce­ dures for weighting objectives and generating trade-offs among them can be considered. We have discussed measures of the contributions of research to objectives of efficiency, income distribution, and risk. If the decision is made to consider multiple objectives and to apply a set of weights to the objectives prior to the analysis, then these weights would also need to be established (perhaps using the procedures described in chapter 7). However, we recom­ mend an alternative weighting procedure, which is described below. The problem with trying to define weights in advance is that typically, people do not have a clue as to what a particular set of weights implies for the trade-offs among different objectives. This is the case, even when the contributions to two (or more) objectives are measured in the same units (e.g., NPVs of benefits to different groups). !tis even more of a problem when the units are not the same (e.g., one is a measure of income variance and another is the NPV of economic surplus) or even comparable (e.g., one is cardinal, such as the NPV of economic surplus, and one is ordinal, such as a score from one to five representing the "small-farmer" focus of a program). Thus we suggest an empirical approach to eliciting the weights, using the model in which the weights are to be applied. As discussed earlier, we recommend that for models with multiple objec­ tives, the model be run first with all weight placed on the efficiency objective 21. Of course, the decision makers need not be risk neutral from a personal standpoint, even if they should be risk neutral when defining an objective from society's standpoint. That is, when eliciting weights from decision makers, it is not their personal preference but their professional judgments about preferences that are being sought. 460 Mathematical Programming and then be rerun with small weights on other objectives. In the example in figure 6.4, the second run might place a weight of 0.1 on the distributional objective. The two solutions could be presented to decision makers to indicate the opportunity cost associated with placing such a weight on the distributional objective in order to determine whether they would like to see additional solutions with different weights. An alternative way of presenting information that would require more runs would be to maximize each objective individually while placing zero weights on the other objectives. This procedure establishes lower bounds on each objective, as described earlier. The benchmark solution can be reviewed with decision makers and compared with the other solutions. Then weights can be modified or lower bounds raised on the particular objectives as desired to examine the trade­ offs among objectives more thoroughly. 6.2.2 Compiling Data and Calculating Coefficients The mathematical-programming analysis uses as basic data the NPVs com­ puted at three or more levels for each research program in a research evaluation exercise, as described in chapter 5. Some additional data may be required, depending on the other objectives being considered (e.g., a security objective implies a requirement for measures of the contributions of programs to security objectives) and the degree of detail of the set of constraints to be considered (e.g., data will be required on the use of land and labor by individual programs if land and labor are to be included as specific constraints in the analysis). The aij and bij coefficients defined in the models above for each research program, associated with the efficiency and distributional objectives, can be estimated using the economic surplus calculations and capital budgeting procedures described in chapter 5. A single-period model will require a discounted sum of net research benefits (i.e., an NPV) corresponding to each level of research support for each commodity research program. Research benefits per unit of research expenditures are likely to vary with the level of expenditures, reflecting a nonlinear research production function (chapter 2). The NPVs of research benefits corresponding to particular levels of research costs can be incorporated in the mathematical-programming model, and the model can be constructed so that the optimal solution reflects the nonlinear nature of the research production function, as described above. 6.2.3 Running the Model The steps discussed above permit a benchmark solution and sensitivity analysis to be generated according to weights on multiple objectives. Mathematical Programming 461 Components of Research Programs The model can be redefined for a disaggregated treatment of the compo­ nents of the most important research programs or regions. As discussed in chapter 5, it is difficult to apply any optimization procedure to some compo­ nents of research programs because of the inherent difficulties in obtaining aij and bij coefficients (measures of contributions to NPVs of economic surplus), particularly for certain areas such as agricultural economics or agroclimatology. However, economic surplus and mathematical-program­ ming models can be applied to some research program components, partic­ ularly when only a few commodity programs are involved, as demonstrated by Scobie and Jacobsen (1992). They applied a nonlinear mathematical-pro­ gramming model to five research program components, defined by five stages in the production and manufacturing process for wool. Their model was nonlinear because it included net revenue functions as polynomials in research spending, incorporated a cost for adjusting research programs away from their current budgets, and added a variance-covariance matrix to account for research portfolio risk.22 Resource allocation models need not represent all research program components in a nonlinear framework, and some, such as plant breeding, crop management and protection, and animal nutrition, can be incorporated in models of the types presented earlier in this chapter. One suggestion is to solve a mathematical-programming model, incorporating the components for which economic surplus estimates can be made. Then, priorities for other components (e.g., social science, agroclimatogy, or soil science) can be developed through structured discussion. Short- and Long-Run Plans The model can be run with different sets of resource constraints to develop short- and longer-run priorities. Using current resource constraints, the results of the mathematical-programming analysis will lead to short-run research priorities. Relaxing the constraints on human resources and facili­ ties leads to longer-run priorities. Plans for investment in training and facilities can then be developed. 22. While it made sense for Scobie and Jacobsen (1992) to include research risk in their model for the Australian Wool Corporation, as discussed in chapter 2, it makes less sense for a public agricultural research system to consider research risk (even though they may be concerned about a social objective to reduce production risk). 462 Mathematical Programming 6.3 Conclusion The above discussion is a brief summary of the use of the mathematical­ programming approach as a guide to allocating research resources. Mathe­ matical programming is a potentially useful tool to provide information for allocating research resources. To be meaningfully applied, the model re­ quires an economic surplus model as well as the mathematical-programming analysis. It is perhaps for this reason that unlike the other models described in this book, mathematical-programming models have seldom been applied to assist with agricultural research priority setting at the strategic level. However, as Scobie and Jacobsen (1992) have demonstrated, these models can be practically applied. From a practical standpoint, it is impossible for research directors to explore all the trade-offs implied when there are several objectives. Never­ theless if sufficient resources are available for research priority setting, a NARS might want to explore the broad options it faces using a procedure that facilitates comparisons of a limited number of alternative research portfolios while readily providing information on the opportunity costs associated with the allocations. As with any research priority-setting procedure, the degree of detail and sophistication should be subject to the criterion that the addi­ tional economic benefits associated with the more complete procedure ex­ ceed the additional costs of its implementation. However, the marginal costs of incorporating the estimated economic surplus changes into a mathemati­ cal-programming model may not be that great. Most of the costs are incurred in the prior step of generating the economic surplus estimates. 7 Scoring and Other Shortcut Approaches The primary rationale for the research evaluation and priority-setting principles and practices described in this book is to provide information to enable strategic research priorities to be formed and to support resource­ allocation decisions that follow from those priorities. A research evaluation study that employs economic surplus measures provides a basis for setting research priorities and making resource-allocation decisions - perhaps using mathematical-programming methods. But this ideal approach is not always feasible. Certain types of research programs (e.g., socioeconomic research or basic research) are not easily amenable to the economic surplus approach, and for such programs some other measures of performance may be required (chapter 5). Moreover, a mathematical-programming model requires a relatively detailed economic-surplus analysis, and in many cases sufficient resources will not be available (chapter 6). In such cases, shortcut procedures might be called upon both to obtain measures of performance (perhaps as approximations of economic surplus) and to assist in setting priorities and making decisions. Perhaps the most common, formalized approach to making decisions on allocating research resources is to rank a set of research program alternatives according to multiple criteria, without appealing to economic surplus and without involving formal optimization. I Most of the approaches that have been used combine shortcut procedures for both measuring program perfor­ mance and making decisions about resource allocations. Often this is done without explicitly identifying the relative importance of the different criteria l. Chapter 5 (section 5.4) discusses rankings as decision aids and considers the types of decisions that can be addressed by rankings and the types that cannot. 463 464 Scoring and Other Shortcut Approaches used for making decisions.2 In recent years, however, relatively formal, structured, weighting procedures called scoring methods have become more prevalent. 3 Scoring methods are viewed as means of reconciling multiple objectives with less information than is required by a mathematical-programming algorithm. In addition, scoring methods can use simple approximations of economic surplus measures when constraints on data or resources for the analysis prohibit a more complete analysis.4 The term "scoring" has been associated with highly subjective methods that lack rigor. However, even when simple approximations are used, the basic economic principles dis­ cussed elsewhere in this book should not be abandoned. These principles concern both measuring the impact of agricultural research and using the measures to set research priorities. In this chapter, we discuss how the elements and results of simplified scoring methods relate to basic principles, and we identify common mistakes that cause those principles to be violated. Steps involved in implementing scoring methods are described as well. In addition, some other "shortcut" methods - including congruence analysis, peer review, and precedence - are reviewed. This chapter places simple, shortcut methods in the context of economic theory. It is inevitable and obvious that these methods generate less precise, and less informative, estimates than would be obtained using the more demanding methods described in earlier chapters. However, while simplified scoring methods require minimal quantitative skill, their proper application and interpretation does require a more complete understanding of basic economic principles than is sometimes recognized. Unless scoring methods are applied carefully, they will readily produce nonsensical results. One purpose of shortcut methods is to foster the development of an institutional­ ized "economic way of thinking" about research; but if economic principles are absent from the process, even this purpose will not be served. The same basic economic principles, concepts, data, and measures are relevant for all serious approaches to research evaluation and priority setting. While each approach may lie at a different point along a spectrum of varying 2. Examples include Binswanger and Ryan (1977), Drilon and Librero (1981), Idachaba (1981), Jahnke and Kirschke (1984), and Norton and Ganoza(1986). 3. Examples include Mahlstede (1971), Williamson (1971), Shumway and McCracken (1975), paz (1981), Chaparro et aI. (1981), Von Oppen and Ryan (1985), Moscoso et aI. (1986), Venezian and Edwards (1986), ESpinosa, Norton and Gross (1988), Ferreira, Norton and Dabezies (1987), Moscardi (1987), Cessayet aI. (1989), Teri, Mugogo and Norton (1990), KARl (1991), Medina Castro (1991, 1993), Palomino and Norton (1992a), Gryseels et aI. (1992), and Dey and Norton (1993). 4. Scoring methods can also be (and have been) applied when more complete swplus measures are used to obtain more precise estimates of the contribution of research to a range of objectives. See Lima and Norton (1993), for example. Scoring and Other Shortcut Approaches 465 detail and effort, they all rest on a single theoretical foundation. More elaborate approaches involve more sophisticated ideas and more complete empirical analysis. Simpler, shortcut methods, may be different in practice but ought not to differ in principle. In keeping with that idea, the main data for well-conceived scoring and other shortcut approaches are essentially the same as the data required for an economic surplus analysis with a balancing of multiple objectives. Indeed the details on data collection and measurement for all of these approaches are provided in chapter 5. 7.1 Scoring 7.1.1 Common Practice versus Basic Principles Common Practice in Scoring Methods Many scoring studies make little, if any, appeal to a meaningful concep­ tual framework, a feature that limits their usefulness for any purpose. Often, they also lack a sound methodological basis, and do not proceed logically; they have usually been conceived and executed in an ad hoc fashion. To focus discussion, and provide a foundation for critical review of common practice, we define a hypothetical scoring study that is comparatively coher­ ent, drawing on elements from the best studies. It should be emphasized that most scoring studies have not involved all of the following steps: • Identify objectives: Several quantifiable objectives, including eco­ nomic efficiency, distributional, and security objectives, are defined in discussion with the clients of the study. Often research policymakers do not perceive the objectives in this fashion at the outset and, as discussed in chapters 2 and 5, the objectives are commonly derived from a set of broader development goals. • Identify program alternatives: Depending on the institutional con­ text of the study, commodity and noncommodity research programs are usually listed. Often the list is long and includes relatively disag­ gregated "alternatives." Sometimes alternatives overlap. Often they are not represented quantitatively (i.e., the current funding of pro­ grams or alternative amounts of support is not identified). • List criteria (grouped by objectives): "Criteria" replace objectives as performance measures to be used to assess program alternatives. As we show below, criteria are often incompatible with one another and are poor proxies for achievement of objectives. This need not be the 466 Scoring and Other Shortcut Approaches case, at least to such a great extent. For publicly funded research investments, criteria that relate closely to measures of efficiency or distributional impact could be identified by drawing on the relevant economic theory. Different criteria, corresponding to private objec­ tives, may be used for private research or for producer-funded research organizations. • Score programs according to criteria: Scientists and policy makers are asked to "score" each of the alternative programs. Often, in the scoring process, little or no distinction has been drawn between (a) scoring as a measure of contribution to an objective and (b) scoring as a weight to be attached to different objectives. As a result, "scores" have been either weights or measures or a hybrid of both. Because there is no formal framework, the analyst must choose units for scores on some arbitrary basis. Often the scores used as measures are ranks (say on a five-point scale from I to 5), and the scores used as weights are fractions (from 0 to 1) or percentages (from 0 to 1(0). Higher scores are usually taken to indicate either a greater contribution to an objective or a contribution to a more important objective, but sometimes the opposite is true. As seen later, unclear or inappropriate definitions of what scores are meant to represent can lead to serious errors. • Rank programs according to scores: Each program alternative (e.g., each commodity program) can be scored according to each criterion and then ranked from highest to lowest. This provides a separate ranking of all programs according to each criterion. • Calculate overall scores for program alternatives: To produce a summary, overall ranking, each program alternative can be scored overall by adding across criteria. The scores on individual criteria might be weighted for aggregation or simply summed. This step might be absent from some studies, which use multiple rankings in decision analysis without reducing them to a summary, overall ranking. • Dialogue: The rankings of program alternatives (either overall or according to individual criteria) are presented to research policymak­ ers and reviewed. The review process is intended to derive im­ plications for resource allocation and, at the same time, to "validate" the results. Thus, in the dialogue, the analyst draws on advice from scientists and policymakers to review the rankings and reevaluate scores, weights, ranks, and implications. How this happens in practice is not always clear. In summary, problems arise with respect to (a) how objectives are defined and measured, (b) how criteria are defined and what the scores associated with them are supposed to reflect (either a measure of the contribution of Scoring and Other Shortcut Approaches 467 research to an objective or a weight reflecting the importance of an objective or elements of both), (c) how scores on criteria are traded off in developing a ranking, (d) how the rankings are validated, and (e) how the results are used to support decisions. Principles for Scoring If the economic approach is to be followed when using scoring for evaluating research and for setting priorities, four operational principles are especially pertinent: Identify meaningful objectives: Setting priorities for agricultural re­ search requires that clear objectives for the research system be identified from discussions with research policymakers (i.e., the research directors, agricultural research boards, or other policymakers who are the "clients" for the work). It is essential that objectives not be confused with means and measures of achieving them. For example, increasing production and em­ ployment and improving nutrition are means of improving economic and physical well-being. If the research policy makers list means and measures, rather than meaningful objectives, additional work is needed to elicit the fundamental and analytically more meaningful efficiency, distributional, or security objectives. Distinguish weights from measures: Weights on objectives should re­ flect the clients' value judgments about the trade-offs among objectives. Determining weights is different from calculating the measures used to assess the research programs' contributions to each objective. Various cri­ teria - such as value of production, probability of research success, and expected adoption rates - can be combined to provide a simple measure of the contribution of research to the efficiency objective. While research policymakers may have views about the relative importance of different objectives, they often have little understanding of how the criteria should be combined to generate meaningful measures of the contributions of research to those objectives. 5 The opinions of scientists, extension workers, and others are needed to specify values for technical criteria, while economic theory provides a guide for how to combine these criteria into a useful measure of the contribution of research to the stated objectives. In what follows we use "criteria" to refer to performance measures: thus an efficiency criterion is meant to measure the contributions of programs to 5. They might not be too good at defining weights, either, as discussed in chapters 5 and 6. Often it is necessary to review the weights in the light of a sensitivity analysis that shows the implications of varying the weights, in order to obtain weights that reflect the preferences of policymakers in relation to trading off the various objectives of the research system. 468 Scoring and Other Shortcut Approaches an efficiency objective. Where possible, we use "scores" to purely reflect weights on objectives. However, we acknowledge that as criteria become increasingly remote from explicit measures of objectives, so too must their corresponding scores stray further from pure measures of weights on the objectives. For instance, the multiplier for area planted to a crop, used as one efficiency criterion, ought to be different from the multiplier on the NPV of research on that crop, used as an additional or, more appropriately, alterna­ tive efficiency criterion. Recognize that research is a blunt instrument: If research policymakers treat research as the only policy instrument available for meeting social objectives, they might select a research portfolio that trades off substantial efficiency for additional equity (or some other objective). If they recognize the presence of other policy instruments, however, they might be less willing to trade off efficiency for, say, equity in selecting a particular research portfolio. Other policies may be more efficient than research at contributing to equity.6 Attempt to approximate economic surplus measures: When criteria that relate to the total research benefit and its distribution are being devel­ oped, where possible they ought to be combined in ways that correspond to the economic surplus measures described in chapter 5. For instance, an efficiency index that corresponds relatively closely to economic surplus measures may be calculated according to equation 7.1. In this equation, the benchmark or baseline value of production for each commodity (Pj x Qj for commodity i) is multiplied (a) by the anticipated proportional reduction in per unit costs, or proportional yield increase, E( ~AX), that would follow if the program were fully successful (at the particular funding level in question, usually the current level, and given other factors such as time to complete the research and so on) and the results were fully adopted, (b) by the estimated probability of success, Pj (treating success as an all-or-nothing outcome), and (c) by the proportion of farmers likely to eventually adopt the new technol­ ogies, A~AX. The result is a gross efficiency index for each commodity research program, G 7 j : G =A Mj AX pjE( y'lAX j j )PjQj (7.1) 6. In chapter 2 we discussed the fact that research is but one instrument of social policy and is often a high-cost way to attain nonefficiency objectives. In chapter 5 (section 5.4.5), the concepts of opportunity costs and weighting objectives in the presence of multiple policy instruments were related to the discussion of benefit transformation curves, BTCs, and indifference curves, ICs, introduced in chapter 2. 7. It can be seen that if this efficiency index is divided by the quantity of output, Qj, the resulting expression corresponds closely to the measure of the per unit research-induced cost saving, kj, defined in chapter 5 (see box 5.1). The main difference is that the expression for k was time-subscripted, reflecting the time path of adoption and research depreciation, /)j. Scoring and Other Shortcut Approaches 469 This is a proxy for gross annual research benefits. Many factors are excluded, such as the costs of research and effects of agricultural policies. A net efficiency index, Nj' can be calculated by dividing the gross efficiency index by the research costs, Rj' that were assumed when questions pertaining to research benefits were asked (e.g., costs of the current research program over the next five years), as in equation 7.2.8 These net efficiency indexes can be ranked from highest to lowest to provide what is best thought of as an ordinal ranking of commodities for the efficiency objective. While the efficiency indexes do provide a rough cardinal ranking as well, their imprecision should be kept in mind - as noted, many factors are not explicitly considered.9 MAX y'lAX N. = Gj = Aj Pj E (i ) Pi Qi I R R (7.2) j j This index is an improvement over the gross efficiency index in that it takes some research costs into account, but there is still no accounting for differences in the timing of flows of benefits and the fact that benefits accrue over many years. 10 A similar approach will yield indexes of contributions to distributional objectives - for instance the total benefits can be apportioned roughly between consumers and producers using information about elasticities of supply and demand. Thus, an index of producer benefits, Np,j' might be derived by multiplying the commodity-specific efficiency indexes by the ratio of the corresponding consumer demand elasticity (in absolute value terms, 1'\j> 0) to the sum of the supply-and-demand elasticities (T1; + E;): II 8. An alternative would be to subtract the research program costs. The problem is one of scale. The gross index Gjis a one-shot measure of the peak annual flow of research benefits for program i, while the research cost is closer to the present value of costs of the program. In many cases the difference between the two will be a negative number even though the rate of return would be positive, and the difference will be larger for larger programs. Thus the ranking of (Gi - Ri) will tend to favor smaller programs (smaller Rj) and could be entirely unrelated to the ranking according to the NPVs or IRRs obtained using the procedures in chapter 5. 9. Expected yield changes, probabilities of research success, and adoption rates together reflect the technical feasibility and usefulness of the research. Elasticities, agricultural policies, and trade patterns are of secondary importance in that they affect the distribution of benefits more than the size of the total benefits. 10. In chapter 5 we discussed the use of the net present value per unit of research resources(Nj* = NPVI Rj) as the conceptually correct criterion forranking research programs when the total research budget is limited. The use of approximate economic surplus measures rather than the net present value of economic surplus measures to approximate the efficiency benefits of research implies that differences in the timing of research benefits and costs among program alternatives are relatively unimportant. Timing is likely to be unimportant for ranking only if the timing of flows of benefits and costs is similar across research programs or if the discount rate is zero; either of these situations is unlikely. II. Figure 2.7 and associated text show how the distribution of benefits is determined by the relative sizes of the elasticities of supply and demand. That equation 7.3 provides a reasonable approximation can 470 Scoring and Other Shortcut Approaches N =N. Tlj P,I I Tlj + (7.3) Ej Review and Critique of Previous Scoring Studies Scoring, as commonly practiced, violates many of the principles for evaluating research and setting priorities identified in the previous chapters of this book.12 Scoring models have been employed for many years for selecting research projects in private industry. 13 They have also been used in public agricultural research systems for more than 20 years, although most of the examples reported in the literature have occurred in the past 10 years. 14 Early attempts to apply scoring models took place in the mid- to late-1960s in the United States, first in a joint study for the U.S. Department of Agriculture and the state universities and land-grant colleges, and later at the Iowa State and North Carolina State agricultural experiment stations. IS Subsequent studies have been conducted in Peru, the Dominican Republic, Ecuador, Uruguay, Colombia, Venezuela, Argentina, Kenya, West Africa, The Gambia, Bangladesh, Tanzania, and many other countries.16 Most of these studies (including some of our own) have violated, and in some cases grossly violated, several basic principles for research priority setting. While most of the studies established objectives for the research system, some did not. In several cases, weights were elicited from research directors to establish the relative importance of objectives or criteria. Com­ modity and noncommodity research programs, research program compo­ nents, or projects were ranked according to each objective or criterion, and these rankings were multiplied by the elicited weights and summed to arrive at overall research rankings. However, in many cases weights were placed directly on measures that are inappropriate criteria - measures that do not translate usefully into objectives - which made assigning weights effectively meaningless. 17 Sev­ eral criteria related to the efficiency objective, which have been employed in be verified from the algebra in section 4.1.1 (equation 4.1). 12 A number of previous studies have discussed the advantages and disadvantages of scoring and how scoring as commonly practiced relates to economic principles. These include Parton, Anderson and Makeham (1984), Fox (1987), Scobie and Jardine (1988), and Norton (1993). 13. See, for example, Moore and Baker (1969) and the literature review by Havlicek and Norton (1981). 14. See the studies cited in footnotes 2 and 3. 15. These studies included Paulsen and Kaldor (1968), Mahlstede (1971), Williamson (1971), and Shumway and McCracken (1975). 16. See the studies cited in footnotes 2 and 3. 17. Examples include Moscoso et aI. (1986), Venezian and Edwards (1986), and Gryseels et aI. (1992). Scoring and Other Shortcut Approaches 471 previous scoring studies, may partially overlap with one another or with the efficiency indexes defined above. These include, for example, (a) value of production per hectare, (b) number of hectares, (c) yield gap between the country of interest and other countries, (d) foreign exchange earnings, (e) comparative advantage, (f) potential completion of research in a reasonable period of time, (g) likelihood of immediate adoption, and (h) current pro­ gram capacity. Such criteria provide a less precise approximation to eco­ nomic surplus than those calculated by combining multiple "criteria" into an appropriate criterion that corresponds to the efficiency objective. Thus, we argue that these other "efficiency criteria" should not be used. The distributional criteria used in prior studies include (a) the number of people employed in producing a particular commodity, (b) the number of producers, (c) average farm size, (d) the quantity of calories and protein in the diet attributable to consumption of a commodity, (e) contribution to sustainable agriculture, (f) political visibility, and (g) the proportion of a commodity consumed on the farm where it is produced. We argue that these criteria, and others like them, should not be used. Most of them do relate to potentially legitimate distributional objectives; however, as quantitative proxies for the research benefits accruing to particular groups, they are likely to be misleading indicators. For example, if demand for a commodity is inelastic such that the demand for total inputs is reduced when supply increases, then the more people employed in its production, the more will be displaced unless the change in technology is biased in favor of labor. Hence, the number of people employed is a poor proxy for the potential positive benefits of research on employment. As another example, if significant income gains are sacrificed as a result of placing weight on calories or protein as a crude proxy for nutritional benefits, then the malnourished could be harmed more than helped for reasons discussed in chapter 2. The bottom line is that these other proxies for measures of research contributions to objecti ves are just too crude to be of use - and they will often be downright misleading. Previous studies have included overlapping criteria that double-count effects. For example, the Technical Advisory Committee (TAC) for the Consultative Group on International Agricultural Research (CGIAR) con­ ducted a study, reported by Gryseels, et al. (1992), that included both value of production and usable land as separate criteria. Clearly these two criteria are highly correlated and both pertain to efficiency. Medina Castro (1991) included value of production, value of trade, and comparative advantage as purportedly independent measures of research contributions to efficiency. The overlapping nature of these criteria is often subtle, but the fact that they are poor proxies for research contributions to efficiency is not. 472 Scoring and Other Shortcut Approaches A separate and equally serious problem with these criteria is that their units are usually incompatible with one other, even for different criteria related to the same objective. Therefore, if weights are attached to them directly in the scoring model, the choice of units for criteria can dominate the weighting and the resulting ranking in unintended ways, as illustrated in appendix A 7.1. Several studies have attempted to circumvent this problem by weighting the rankings corresponding to the numerical values for the criteria rather than weighting the actual values. However, this procedure introduces a new problem because it eliminates the cardinality of the values within each criterion. And differences across criteria in these cardinal as­ pects of the data can convey useful information to decision makers when programs are compared. Although scoring methods are usually implemented when resource con­ straints preclude a more complete analysis, it is possible to achieve greater consistency with basic principles than has been achieved in the past. Below, we suggest how a scoring model can be constructed and applied so as to achieve as much of this consistency as possible, while economizing on time and other resources. Scoring is not the only shortcut method for informing the priority-setting process when resources for analysis are tight. In many cases it might be preferable to apply some basic priority-setting principles, or guidelines, without calculating economic surplus per se or explicitly weighting even simple surplus measures. Section 7.2 discusses some alter­ native shortcut procedures that might be preferred to scoring in some set­ tings. 7.1.2 Defining a Simple Scoring Model Application of a simple scoring procedure involves steps similar to those defined in the beginning of chapter 5 for implementing an economic surplus analysis. In short, it is necessary to (a) define the objectives of the analysis and of the clients, (b) establish weights on the clients' objectives, and (c) identify the program alternatives to be evaluated and compared. Then the contributions ofthe program alternatives can be assessed and the results used in decision making. A crucial distinction from the approaches discussed in previous chapters, however, is that in scoring models, the objectives are replaced by "criteria" (that purportedly measure the contributions of research to the objectives) and the weights are replaced by "scores" (that are meant to translate a measure of a criterion into a measure of achievement of an objective). Because the criteria and scores are often ordinal rather than cardinal and because the criteria are proxies, explicit optimization is not possible. Indeed, the use of Scoring and Other Shortcut Approaches 473 these approximations means that the result from scoring is, at best, an ordinal ranking. At worst, the ordering will be meaningless. Specifying Objectives One of the two major reasons for using a scoring method is to reconcile multiple objectives. Typically, there will be an efficiency objective and several nonefficiency objectives. In most simple scoring models, as we show later, it is unlikely that measures of security objectives will (or should) be included. Two reasons for this are (a) the difficulty of weighting across different units of measures associated with different criteria (e.g., real dollars for efficiency gains versus percentage changes for variability) and (b) the complexity of the calculations associated with measuring meaningful criteria to represent food safety or food security as components of a security objective. The methods discussed in chapters 5 and 6 may be helpful for incorporating security objectives related to research risk and income risk in a priority-setting analysis. As discussed earlier, means or measures of achieving objectives may be specified, and the analyst then has the task of identifying the implied, and operationally meaningful, corresponding objectives. For example, research directors may specify objectives of increasing agricultural productivity, generating foreign exchange, improving nutrition, increasing production of indigenous crops in upland regions, and increasing self-sufficiency. The implied or measurable objectives that are meaningful in an evaluation frame­ work could be mapped as shown in table 7.1, for instance. This is not always easy to do in a meaningful way and requires a reasonable depth of under­ standing of the economic relationships between research investments and their economic consequences. In other words, it is not appropriate to choose a scoring approach on the grounds that it requires relatively little economic expertise. In many respects, the opposite is true: the requirement for techni­ cal economic skills and economic intuition is often greater when informal alternatives replace formal, structured ones. Table 7.1: Mapping Stated Objectives into Measurable Objectives Stated objectives Measurable objectives • Increased agricultural productivity • Increased economic and physical well­ • Increased foreign exchange being for all producers and consumers • Improved nutrition • Increased economic and physical well­ • Increased production of indigenous being of the poor, many of whom live in upland crops the uplands • Increased self-sufficiency • Increased self-reliance 474 Scoring and Other Shortcut Approaches Eliciting Weights and Dealing with Multiple Objectives Various procedures could be used for eliciting weights for objectives. One method is to collect all of the information required to complete the analysis of the contributions of research to each of the individual objectives, omitting a direct elicitation of weights. Then, the analyst may demonstrate the oppor­ tunity costs of using research for nonefficiency objectives by varying the weights on the objectives (and thereby varying the mix of programs in the portfolio). One practical way to proceed is first to choose a research portfolio considering only the efficiency objective. Then, the implications (cost of efficiency benefits foregone) of placing incremental weight on the noneffici­ ency objectives can be observed by seeing how the decisions change and comparing their efficiency outcomes. However, in the typical scoring con­ text we do not have a proper measure of the research contribution to the efficiency objective. Thus, the efficiency trade-off in a scoring model is strictly qualitative; there is no concrete opportunity cost interpretation. An alternative method, which can provide a starting point for discussion about alternative research priorities, is to elicit initial weights on all objec­ tives at the beginning of a priority-setting analysis. For instance, research policymakers can be asked to assign 100 points to the efficiency objective that reflects a desire to improve the well-being of the average person in society. Then, they can be asked to give additional weight (points) to each distributional objective they have identified. These additional weights place extra emphasis on benefits received by people in the identified group. For example, benefits to low-income producers might receive an additional 20 points, and benefits to people in region X, an additional 5 points. Weight may be given to security objectives as well. A Delphi procedure can be used to arrive at an initial consensus on the weights. To implement this procedure, the objectives that are identified are listed on a form and each participant is asked to place weights on the objectives. An example of a form (with sample objectives) that can be modified for use in eliciting weights on objectives is given in table 7.2. The answers from all the individuals are averaged and they can each be shown the group average and their own responses and asked if they would like to change any of their weights. The process can be repeated until the partici­ pants are "satisfied" with their weights. It may be useful in second or third rounds for the individuals to meet as a group and to endeavor to justify their weights to one another. In practice, it may make little difference whether weights are elicited directly at the beginning of the priority-setting study or indirectly at the end after trade-offs have been demonstrated. Either way, some adjusting of Scoring and Other Shortcut Approaches 475 Table 7.2: Sample Questionnaire for Eliciting Weights on Agricultural Research Objectives Gi ve 100 points to the objecti ve of raising the average level of well-being in society. Then, if desired, assign additional points (weights) to the other distribution or security objectives that you have identified as being important for public agricultural research.a Objective Points Weights 1. Improve the average level of well-being for citizens 100 in the nation 100 150 = 0.67 2. Improve the well-being of producers on small farms 30 in the mountain region. 30 150 = 0.20 3. Reduce the annual fluctation in national agricultural 20 = 0.13 income 20 150 alt is important to recognize that assigning additional weight to distributional and security objectives is likely to reduce the impact of research on economic growth or the average level of well-being. Each individual in society who is a member of the group identified in objective two is weighted 100 points for being a member of group one plus the points for being a member of group two. weights occurs at the end of the study after the decision makers are given information about trade-offs. As we show below, even the choice of units for criteria can have major implications for the results of a scoring analysis with a given set of weights. Thus, it is virtually impossible for decision makers to have a sensible prior view of the appropriate weight to put on each criterion, even when they know the weight for each objective. There is little option but to determine weights empirically in the light of the rankings they imply. Unfortunately, this makes the ranking process somewhat circular. Specifying Commodity and Noncommodity Research Programs Some aggregation of programs is necessary in order to make the analysis cost-effective. The appropriate aggregation of research programs according to their commodity or disciplinary focus will vary among studies. For example Scobie and Jacobsen's (1992) study was concerned solely with Australian wool research and the disaggregation was according to where research applied in the wool production-processing-marketing chain, in accordance with the Australian Wool Corporation's research program. On the other hand, most NARSs have a long list of commodities on which they are currently (or potentially could be) conducting research. Some commod­ ities are clearly of such minor significance that they would never be high on the list of priorities and can be eliminated from more formal analysis. For 476 Scoring and Other Shortcut Approaches example, a 1986 Ecuador study began with a list of 109 commodities that was pared down to 44 for the analysis (Espinosa, Norton and Gross 1988). Some grouping of commodities can also be undertaken. Individual fruits or vegetables or small ruminants are prime candidates for grouping in many countries. But again, it is hard to generalize. It may be inappropriate to aggregate fruits and vegetables when setting research priorities in California, for example, where certain individual horticultural products represent hun­ dreds of millions of dollars in annual output. A possible rationale for grouping commodity programs is whether they are closely related, so that the same researchers could work on them. In some cases, a relatively homogeneous single commodity may be separated into more than one type of commodity for purposes of analysis. For example, rice research might be differentiated into irrigated rice, upland rice, and swamp rice. This kind of disaggregation may be appropriate if different types of the commodity are produced in different regions or if separate researchers (or research programs) are required for each type. In either of these cases a dis aggregated analysis will better reflect the units on which resource allocation decisions are made. Noncommodity research programs must also be defined. Research sys­ tems in developing countries are usually organized in program areas (such as plant protection, plant breeding, animal nutrition, and soil science) in addition to, or as a component of, commodity research programs. While these program areas may correspond to disciplines, usually they do not correspond exactly, and definitions of program areas vary from place to place. It is important to define these research areas in a manner that is meaningful to the scientists who will be responding to requests for informa­ tion during a priority-setting analysis. It may be necessary to tailor the list of research areas to each region or experiment station. Research directors or program leaders are often logical sources of information when the lists of research areas to be evaluated and prioritized are being developed. Efficiency Criteria/or Commodity Research Programs In a scoring model, the criteria are meant to indicate the contributions of research to objectives. The efficiency criteria ought to relate meaningfully to economic surplus measures. Several studies have sought approximate eco­ nomic surplus measures (e.g., Dey and Norton 1993; Palomino and Norton 1992a; Lima and Norton 1993) but others have not (e.g., Venezian and Edwards 1986; Medina Castro 1991; Gryseels et al. 1992). Studies of The Gambia and Tanzania by Cessay et al. (1989) and Teri, Mugogo and Norton (1990) developed simple proxies for economic surplus measures of the contri- Scoring and Other Shortcut Approaches 477 butions of research to the efficiency objective. Studies of Bangladesh, Ecuador, and Venezuela by Dey and Norton (1993), Palomino and Norton (1992a), and Lima and Norton (1993) used explicit economic surplus measures. The efficiency gains from commodity research programs, as measured by the NPV, depend on the size of the research program (research costs), the size of the industry being affected (value of production), and a number of factors that influence the size of the research-induced cost saving, K. These are discussed below. • Size of the research program or research costs (R): Issues associ­ ated with gathering information on research costs are discussed in chapters 3 and 5. We suggest using current costs as a benchmark so that the costs used in interviews with scientists are in the range of their recent experiences. Although it adds some complexity, it is often preferable to ask scientists about the reduction of input costs per unit of production (or yield increase) anticipated under several funding levels - e.g., the current research support and the current funding plus or minus a certain percentage (say, 10% of total costs or 50% of operating costs). • Value of production (PoQo): An average of the pastthree or four years may be used to smooth out annual variations without introducing marked distortions due to trends in prices or output. For certain commodities, estimates of the value of nonmarket products (e.g., manure or straw) are needed. Also, the value of intermediate products (such as forages that are produced and consumed within agriculture) may be subtracted from the value of the final product and included as separate commodities or, in some cases, the intermediate and final products may be considered together as one commodity research program (e.g., as a combined beef cattle and forage research program). • Probability oF-research success (p): Recall that, as discussed in chapter 5, this must be defined carefully (it is jointly determined with the definition of a successful research outcome and depends on the assumed value for research costs). In particular, estimates of the cost reductions or yield increases arising from research are best obtained in conjunction with estimates of the probabilities of research success because there is an element of joint determination in the two values. • Maximum cost reduction per unit of output or yield increase (KMAX): For a simplified scoring analysis, a single summary measure is often used, and to facilitate comparisons across programs, the percent­ age yield or cost changes need to be standardized on a particular period (say, five to 10 years following the initiation of research) and on a common basis in terms ofthe research programs (say, the current level 478 Scoring and Other Shortcut Approaches of research spending). The ability to borrow research results from abroad can be taken into account when establishing the per unit cost reductions or potential yield gains due to research. 18 • Likely extent of adoption (AMAX): Simple scoring models usually incorporate the ceiling rate (anticipated maximum percentage) of adoption, AMAX, and ignore the time dimension. When commodity programs are being ranked, the above criteria can be combined to provide a rough economic surplus measure of the contribution of research to the efficiency objective. As shown in equations 7.1 and 7.2, these individual criteria are multiplicative rather than additive in their impact on research efficiency - and they are also involved in indexes of the distributional impact of research. Thus, they should not be treated as separate criteria, both because of the potential for double-counting and because their effects are not additive. Chapter 5 provides details on sources of data and approaches to measur­ ing these criteria, including some discussion of conceptual and measurement issues. Elicitation forms for obtaining information from scientists and others are included in that chapter in appendix A5.3. Efficiency Criteria for Noncommodity Research Programs or Program Components When noncommodity programs are important, ranking commodities alone is not sufficient for guiding the allocation of resources to research. However, values for many of the criteria identified above (e.g., per unit cost reduction, adoption rates) are especially difficult to estimate for some cross­ commodity or non commodity research areas (such as socioeconomics and soil science). This problem may be handled in three ways. One option is to abandon formal economic surplus measures altogether and to choose a different set of more qualitative proxies for which each program can be ranked high, medium, or low, rather than utilizing a more specific number. 19 Qualitative proxies may provide inferior measures of changes in economic 18. As discussed in some detail in chapter 5, it is important to make sure the information that is gathered on the changes in yields and costs is relative to what costs or yields would have been without the research, rather than relative to current costs or yields. Benchmark information on changes in yields or costs in previous years is particularly valuable (see chapter 5). 19. This alternative has been employed in several studies (e.g., Palomino and Norton I 992a; Dey and Norton 1993). Proxy criteria that have been used are (a) "number and severity of researchable problems" to reflect, in part, per unit cost reductions and probabilities of success, (b) "effects of research on monetary costs" because that factor may influence adoption, (c) research costs, and (d) "complementarity with research in other countries or international centers" on the grounds that in order to maximize efficiency, research in national systems should complement rather than substitute for research abroad. Scoring and Other Shortcut Approaches 479 surplus (or may only roughly correlate with surplus changes) but they can be applied to all research program areas. The drawback of this approach is that it fails to take any advantage of the quantitative information that is readily available for certain research programs. At the opposite extreme, a second alternative is to use explicit economic surplus measures for all research programs. This approach involves trying to estimate, however poorly, the per unit cost reductions, adoption rates, and so on for noncommodity research areas. In many cases this will be totally infeasible. A third alternative is to obtain quantitative estimates on criteria for those research programs that can be reasonably quantified, and then to rank the other programs through structured discussion without any scoring. This last alternative, or some variant of it, is probably the most reasonable approach. Quantitative information on projected yield changes, adoption rates, and the like can be obtained for program components (such as genetic improvement, crop or livestock management, crop protection, animal nutrition, and animal health) when information is gathered to prioritize commodity research pro­ grams. However, it is extremely difficult to quantify relevant measures of yield changes, adoption rates, and other factors for many noncommodity research areas. It could be misleading to attempt to estimate values for the standard criteria used to calculate expected changes in economic surplus in such cases. Hence, the suggestion is to quantify such effects for certain areas and to rely on discussions among and reviews by scientists and research directors to prioritize the others in relation to them. These discussions can focus on criteria such as the number and severity of researchable problems in each area and the cost of research, but not necessarily using a quantitative approach. The less-formal procedures, peer-group discussion, and rules-of­ thumb discussed in section 7.2 may be applicable. Criteria Related to Distributional Objectives The same basic criteria related to efficiency can be included in every priority-setting study (i.e., this is basic information needed to estimate total economic surplus). However, this may not be true for criteria related to distributional objectives because those objectives may differ from situation to situation. As shown above (in equation 7.3), however, the distributional effects that relate to shares of total economic surplus generated by research depend directly on the criteria used to define the efficiency index. If dis­ tributional criteria are defined as fractions of the efficiency criterion, there is a better chance of consistency with economic theory and a smaller chance of problems arising from the use of different units. 480 Scoring and Other Shortcut Approaches Several criteria have been used as indicators of whether research will contribute to distributional objectives. However, most of the criteria have been crude or misleading proxies for the corresponding economic surplus measures. When a simplified economic surplus model is used in a scoring model (i.e., when detailed distributional effects are not calculated), the total economic surplus can still be roughly apportioned to the relevant groups. For example, if the distributional objective were to help small farmers, the measure already calculated for the efficiency objective for each commodity could be multiplied by an estimate of the share of that commodity grown on farms below some specified size. Clearly this share must vary among commodities to be a useful criterion. If the objective were to help producers in a particular region, the aggregate economic surplus might be apportioned among regions according to their (commodity-specific) shares of produc­ tion. If the objective were to help low-income consumers, the aggregate economic surplus for each commodity could be allocated according to the share consumed by low-income consumers. Even though the resulting mea­ sures of distributional benefits are crude and these shares are sometimes difficult to estimate, rough estimates of them are likely to result in more accurate distributional proxies than are the myriad of other criteria used in many previous studies. Few of the prior studies have attempted to approximate economic surplus measures of the distributional effects corresponding to the nonefficiency objectives. Lima and Norton (1993) calculated economic surplus measures of distributional effects. Other studies that have used economic surplus to measure efficiency effects used only crude proxies for distributional effects. Indeed, most scoring studies have failed to develop even a plausible proxy for the contribution of research to the efficiency objective. In many cases, the result has been the use of inaccurate and overlapping criteria and double- count.m g 0 fbe nef iI tS. 20 Criteria Related to Security Objectives Several criteria have been suggested as measures of the contribution of research to security objectives.2l These criteria relate to objectives of reduc­ ing annual income variability, attaining self-sufficiency, and enhancing food 20. When distributional objectives are disaggregated finely, the use of partiCUlar individual charac­ teristics as criteria is liable to involve inaccurate proxies, overlapping criteria, and double-counting. For example, there may be a concern about low-income producers in a particular region, or low-income consumers, rather than simply all low-income people, all people in a particular region, or all consumers. Depending on which single objective or combination of objectives is chosen, a different set of measures may be needed. 21. For example, Venezian and Edwards (1986), Gryseels et aI. (1992), and Dey and Norton (1993). Scoring and Other Shortcut Approaches 481 safety. The implications of incorporating these objectives in an economic analysis of research priorities were discussed briefly in chapters 2 and 5. Yield, production, price, or income stability: Income variability may be reduced by research that leads to increases in the output of commodities with a relatively stable yield, production, price, or income (Venezian and Ed­ wards 1986). Variability in the value of production of a commodity is sometimes measured by calculating its coefficient of variation (Dey and Norton 1990). Coefficients of variation or other variability measures (per­ haps weighted according to the relative importance of the commodity in total income) can be employed to rank commodity research programs according to the potential for research on them to reduce overall income variability by changing the product mix. However, such measures are difficult to incorpo­ rate in scoring models because it is not obvious how to trade off coefficients of variation (as measures of risk) against other criteria (such as those that measure contributions to efficiency), even when the relative importance of efficiency and risk are known. As shown below, when weights are placed on objectives, aggregation problems arise because of differences in units of measurement. If yield, production, price, or income variability is a concern in a particular study, we suggest using formal optimization techniques rather than simple scoring. A separate issue is whether any weight should be attached to research that will reduce yield variability in particular crops by reducing their vulnerabil­ ity to pests and weather. In some cases a stated research objective is to reduce an individual commodity's production variability. Underlying that stated objective may be a desire to reduce the income variability experienced by individual farmers or a goal of reducing the chance of famine. The relation­ ship between achievement of either of these goals and reducing yield vari­ ability is not altogether clear, and the appropriate measure of variability will depend on what fundamental goal is being pursued. Aside from these conceptual and related measurement issues, there is still the difficult (if not intractable) issue of defining the terms of the trade-off between efficiency and variability. Self-sufficiency: Policymakers often mention self-sufficiency as an ob­ jective. A country may prefer to sacrifice some efficiency gains or income to reduce the vulnerability of the food supply to a military conflict or other political trouble elsewhere in the world. Settling on approaches that will achieve this objective is not at all straightforward - even considering more direct interventions than research policy. For instance, if one applies a criterion such as current quantity imported (Venezian and Edwards 1986), so that additional weight is placed on commodities that the country is importing, there is no reason to believe that this emphasis will help the 482 Scoring and Other Shortcut Approaches country become more self-sufficient in the aggregate amount of food, even if it becomes more self-sufficient in a particular commodity. Indeed, the empirical evidence and the weight of the literature on the gains from trade suggest that such attempts to become self-sufficient have been counterpro­ ductive. Food safety: As per capita incomes rise and the proportion of the food supply that is processed increases, concerns often arise about food quality (Pingali and Roger 1995). It is difficult to find quantitative measures of research contributions to food safety, and food safety is seldom included in priority-setting studies in developing countries, the place where simple scoring models are most likely to be used. It is possible, however, to calculate the economic benefits associated with food safety by calculating more complete economic surplus measures. Collecting Information Related to Criteria It is important to note that the types of data required for a well-conceived scoring model are similar to the types of data required for a corresponding mathematical-programming model. Some scoring procedures use measures of NPVs of economic surpluses and their distribution as data. For simple scoring models,less detail is needed and some parameters required for formal measure­ ment of economic surplus are not needed at all. However, there is a correspond­ ing loss of conceptual rigor and the danger of losing any link to measures of efficiency gains and losses. Chapter 5 spells out the procedures for obtaining and constructing measures of the relevant variables. Appendix A5.4 provides examples of elicitation and data-compilation forms that could be used to develop estimates of technical parameters based on information obtained from scientists, extension workers, research administrators, and others. 7.1.3 Implementing a Scoring Model Ranking Commodity Research Programs Scoring models can be developed that incorporate results from the appli­ cation of economic surplus models, as described in chapters 4 and 5. The studies by Palomino and Norton (1992a), Dey and Norton (1993), and Lima and Norton (1993) each incorporated measures of economic surplus changes due to research, as measures of the contributions to the efficiency objective. Scoring was then used to illustrate the efficiency trade-offs involved when alternative weights were applied to distributional objectives. Scoring might also be used to combine these quantitative measures of economic perfor- Scoring and Other Shortcut Approaches 483 mance with qualitative measures or ordinal rankings in relation to other criteria - but this is dangerous to do and may not be informative because it is difficult to meaningfully define the terms of the trade-offs between qualitative and quantitative measures. More commonly, scoring has used some type of efficiency index that is related to economic surplus but perhaps only loosely. Efficiency: Here, we have suggested efficiency indexes that correspond relatively closely to economic surplus measures for ranking commodity research programs when a simplified scoring model is used. Gross and net efficiency indexes, G; and N;, respecti vely, were defined in equations 7.1 and 7.2 as MAX • MAX Gj=Aj pjE(rj )PjQj and The net efficiency indices can be ranked from highest to lowest to provide an ordinal ranking of commodities for the efficiency objective. The Gj and Nj might lead to very different rankings. The Nj at least makes some attempt to factor in the size of the research program and is preferred for that reason. Appendix A 7.2 contains templates for computing gross and net efficiency indexes. Regional distributional objectives: The contributions of each of the commodity research programs to each of the distributional objecti ves chosen for the study may be best measured as transformations of the efficiency indexes. If the objective were to improve the well-being of people in a particular region, then an efficiency index for each commodity research program for that region could be computed using the values of variables for each commodity in the particular region in equation 7.2. This regional efficiency index would measure the contributions of the research programs to the distributional objective. An approximate measure could be obtained by scaling the aggregate efficiency index by the fraction of total production that is produced in the region of interest. Other distributional objectives: Above we suggested how a net effi­ ciency index, Nj' could be scaled to derived an index of producer benefits, Np,; - i.e., Np,j = N/T'I; I(c; + l1J By the same argument, an index of total consumer benefits, Nc,;' is given by Nc.;= N,£; I(c; + l1J A particular group of consumers or producers could be targeted by scaling the total producer and consumer benefits by the group's shares of consumption or production of individual commodities. For example, if the distributional objective of help- 484 Scoring and Other Shortcut Approaches ing small farmers were chosen, the Nc.; could be multiplied by the share of each commodity produced on small farms to obtain a measure of research contributions to that objecti ve by commodity. 22 Of course, the more detailed this disaggregation, the more information required to disaggregate meaning­ fully, and the less clear the justification for not doing a complete economic surplus analysis. Combining criteria: Typically three or four indices of research contribu­ tion may be used: one for efficiency and several for distributional objecti ves. In forming a measure of the weighted contribution of different research programs to the set of objectives, each index could then be multiplied by the average weight for the corresponding objective, using the weights elicited at the start ofthe exercise (see table 7.2). Then the commodity programs could be ranked from highest to lowest, based on this sum, to arrive at the aggregate ordinal ranking. Using elicited weights in this way is dangerous. The problem is that the choice of units for criteria can dominate the weighting in unintended (and unanticipated) ways, as illustrated in appendix A 7.1. It is difficult to con­ ceive of appropriate weights to be attached to criteria that relate quantita­ tively, in an unknown way, to an underlying objective. For this reason, a preferable alternative is to begin by placing all the weight on the efficiency objective and then to demonstrate the implications of placing incremental weights on other objectives. The decision makers would then choose the "final" weights in light of that information. Appendix tables A7.2 and A 7.3 contain templates for computing weighted rankings to use for this purpose. Ranking Research Program Components Analysts conducting strategic priority-setting exercises are often asked to assist with prioritizing research areas only at the regional or experiment-sta­ tion level, typically as components of commodity research programs. It may be difficult to rank research program components - such as plant breeding, crop management, and soil science - meaningfully for a nation as a whole, if problems and resource bases differ regionally. An efficiency index for a particular component of a commodity research program can be calculated analogously to the efficiency index for the commodity program as a whole, using the equivalent of equation 7.2, with the exception that several component-specific parameters replace their ag- 22. This is tricky. For instance, if it were decided not to adjust for elasticities to obtain a producer share of benefits, an entirely inappropriate ranking could result. This can occur if some commodities are traded (producers obtain all benefits) while some are not traded and demand is very inelastic relative to supply so that the producer share is very small. Scoring and Other Shortcut Approaches 485 gregate counterparts. The proportional yield change attributable to a partic­ ular component, c, of the commodity program is equal to the total yield change attributable to the entire commodity research program multiplied by the proportion attributed to that component (see appendix table A 7.3).23 The other parameters - the probability of success, the maximum adoption rate, and research costs - can be specific to the program component or not. Thus, for the cth component of the ith commodity research program (where the subscript, c, denotes the component-specific nature of parameters), MAX Y,tAX Ai,e Pi,c E ( i,e )PiQi Ni,c= R. (7.4) l,e In order to achieve internal consistency in the analysis, where net effi­ ciency indices are computed for all components of a commodity research program, it may be appropriate to compute a net efficiency index for the commodity research program as a whole. Such an index can be formed as a weighted average of the indices for its components rather than using equation 7.2 directly. That is, for a program comprising C components, Ni = Lc. (R~.c .] Ni,c (7.5) c=1 I Consistency is desirable, but the aggregation of components in this way is likely to ignore important interactions among the components.24 On the other hand, the differential information on the disaggregated components has potential value, too, and should not be wasted. Thus, whether it is better to use an aggregate index derived from component indexes rather than one derived directly - or, indeed, to do the converse and derive component indexes by disaggregating an aggregate index - remains moot. This must be left to the judgment of the analyst, based on the problem at hand. Often a ranking of research program components by regions or stations is desired. Once decisions are made on the distribution of the commodity research programs among the regions or experiment stations (often outside the scoring model because simple scoring does not include research spill­ overs), the efficiency and distributional indices can be apportioned by region or station to arrive at noncommodity program priorities at the regional or station level within a country. 23. Recall, as discussed in chapter 5, that when looking at components, we must be especially careful to be clear about what is being held constant in other components, and the substitution or complementarity effects between components must be considered. 24. Of course, if there are no interaction effects, there should be no difference between indexes derived in these alternative ways. 486 Scoring and Other Shortcut Approaches Interpreting and Using Results Rankings of research programs are usually presented by commodity, sometimes also by research program components within commodity pro­ grams or by region and, perhaps, by research program components for each region. Each criterion implies a separate ranking of all programs. Multiple rankings must be reconciled for decisions to be made. As we have seen, the weights on objectives do not carry over as weights on the criteria used to represent those objectives. And choosing appropriate weights is paramount. Using efficiency indexes apportioned across subaggregates as criteria for other (distributional) objectives reduces the potential for totally inappropri­ ate weights. But even when these measures are used, since it is difficult to know what the appropriate weights are (even for sound and consistent measures), we recommend establishing weights or scores empirically, by considering the rankings they imply. The type of guidance offered by scoring analysis is less complete than often imagined. It might be inferred, for instance, that higher-ranking com­ modity programs should receive the highest amount of funding and staffing, other factors being equal. Such inferences cannot be drawn from scoring models because these models can only provide an ordinal ranking of the specific alternatives being considered. Indeed, as discussed in chapter 5, ordinal rankings can never dictate decisions about resource allocation. In chapter 5 we argued that even cardinal rankings of NPVs are of little use for decision making beyond the all-or-nothing choice to close down programs for which the NPVs are negative or for which the NPVs per unit of research mean they cannot be supported. When approximations to economic surplus relative to research costs have been used to generate priorities, one may be tempted to use the relative size of the efficiency or weighted indexes as cardinal measures of research priorities. In other words, if the net efficiency index for rice is twice as large as the index for wheat, one might think that rice research should receive twice as many resources - or, perhaps, all of the available resources. This reasoning is incorrect. Particular alternatives would have to be explicitly scored and ranked if a choice were to be made between them. An example of this would be comparing an existing set of programs with a situation in which there was a 10% increase in crop programs, financed by a proportional reduction in all other programs. Scoring models typically do not include any information about the shape of the research production functions for differ­ ent research programs. Thus, unlike the mathematical-programming ap­ proach to optimization, scoring does not allow for any marginal analysis of program changes or optimization within a portfolio of programs. Scoring and Other Shortcut Approaches 487 In addition, rankings deri ved from using scoring methods can be wrong. The large number of simplifying assumptions incorporated in the analysis, particularly the assumption that the time flows of benefits and costs are the same across programs, means that the results of simple scoring models should not be used as if they represented an accurate cardinal ranking of the performance of research program alternatives. An ordinal ranking of re­ search priorities is provided, nothing more. At best, the ranking derived from a scoring analysis could be used to make the all-or-nothing decision about which programs to support. This could be done by adding up program costs, moving down the ranking, until the total budget for all programs was exhausted - which amounts to treating the results as if they correspond to cardinal rankings, according to NPVs per unit of research resources. Such a procedure might be justifiable but then the programs must be ranked only according to the net efficiency index, N;. 7.2 Other Shortcut Procedures Our analysis suggests that scoring should be used sparingly. The results are unreliable and potentially very misleading. What other approaches are available for evaluating alternatives and guiding decisions about allocating research resources when resources or information constraints preclude an economic surplus analysis? Several options have been developed for cir­ cumventing data or other constraints in a shortcut approach to research evaluation and priority setting. These options include rules ofthumb, ad hoc or informal procedures drawing on the theoretical results of economic surplus models, and peer review.25 7.2.1 Rules of Thumb and Guidelinei6 Parton, Anderson and Makeham (1984) include precedence and congru­ ence as rules of thumb governing the allocation of research resources. Rules of thumb have been widely used because of their simplicity, low data needs, and low cost. Their primary disadvantage is that they are crude methods that permit a low level of scrutiny and, hence, over time may lead to a certain amount of inflexibility in the allocation of funds. 25. Por example, the Office of Teclmology Assessment (1991, appendix D) describes the shortcut approaches used to set priorities for academic and basic research by various agencies in the U.K., Germany, Prance, Japan, The Netherlands, Sweden, and Canada 26. This section dmws heavily, in some parts vematim, on a section with a similar title in Scobie and Jardine (1988, pp. 30-33), which in tum dmws heavily on Shumway (1977). 488 Scoring and Other Shortcut Approaches Precedence The precedence model regards the previous year's funding as the base for allocating funds in the next year for each project or research area. Funds are then either increased or decreased. Typically, however, such changes are small. The approach has the advantage of providing long-term continuity in the funding of research projects or areas. Its disadvantage is that research that has reached the limit of its productivity may continue to be funded because of the inbuilt inertia associated with the reliance on past funding practices. Under a precedence approach, changes in total resources are commonly shared in equal proportion among research activities. Significant changes in the shares of total resources going to programs are likely to occur only when there is a major change in the system and, in effect, the precedence approach is abandoned. Precedence provides no basis for comparing future benefits since decisions are all based on past funding rather than potential perfor­ mance. This makes it difficult to introduce new areas of research and is liable to result in a suboptimal allocation of research resources. The fact that funding tends to be allocated to areas with high historical funding levels is not necessarily irrational. If accumulated research skills and experience represent a greater stock of research capital, then the benefits from additional funding may be higher in traditional areas of emphasis than in a program for which historical funding has been limited. On the other hand, there might be marked diminishing returns to further investment in areas that have traditionally been strongly supported. This would suggest that the marginal return to areas in which the accumulated stock of capital knowledge has been very limited may, in fact, be quite high. This dilemma is symptomatic of the imperfect state of understanding of the processes involved in the generation of new knowledge. Congruence The congruence or parity model allows more flexibility than the prece­ dence model in the allocation process. It involves the allocation of research funds across research areas in proportion to their contribution to the value of agricultural products.21 For example, if the value of com output is twice that of cassava, then com would receive twice as much research funding. Alter- 21. More fonnally, Boyce and Evenson (1975) defined a congruence index, CI, such that n CI = 1-1: (Si - RSi)2 i=I where Sj = share of each connnodity i in total value of output and RSj = share of total research expenditures spent on connnodity i. Perfect congruence between value of production and research expenditure shares would imply CI = I. The greater the mismatch between value of production and research expenditure Scoring and Other Shortcut Approaches 489 natively, a congruence ratio can be defined in which the research budget of an area, program, or discipline as a proportion of the total research budget is computed as a ratio of the value produced (added) to the total value of production of the corresponding area or program. For example, if the value of com output is equal to 20% of the total value of production, congruence would require com research to receive 20% of total research resources (Boyce and Evenson 1975). In effect, congruence equalizes research-inten­ sity ratios (research spending as a fraction of the value of output or value added) across programs. The congruence model can be used to compare resource allocations to research by commodity, by factor, by production stage, by region, and among disciplines. It can be applied to each of these dimensions singly or in combination, but then it becomes increasingly complex. The congruence model is a useful starting point in analyzing resource allocations to research and is one of the simplest techniques for allocating research resources. It provides a relatively gross basis for comparison, but its usefulness lies in identifying areas where the ratio is low. Such a low ratio may well be justified, but it might also indicate areas where a reallocation of research resources could be profitable.28 The role for analysis in a congruence study is in the interpretation of these ratios. The congruence approach is a distinct improvement over precedence in that it considers two of the determinants of the net payoff to research - the size of the industry and the size of the research program. It argues that funds ought to flow towards programs with relatively low research intensities and from programs with relatively high research intensities. This presumes that an additional dollar of research expenditure would have a higher return if spent on areas with a relatively low ratio of research funding to output value - i.e., it means that moving towards equal research intensities increases the overall research benefit. To see the relationship between congruence and the scoring approach (and, in tum, the NPV approach), consider the equation for the net efficiency index: N. = Gi =A MAX p. E (y'fAX ) [Pi Qi] (7.6) I Ri I I I Ri shares, the lower the index. Some versions of congruence go beyond the simple proportionality rules applied between research and the value of production. For instance, Byerlee and Morris (1993) multiplied the value of production by variables representing "expected research progress," "strength of local research effort," and "incidence of poverty." This approach seems to relate more closely to scoring (using a multiplicative objective function with equal weights) or to a modified efficiency index than to congruence as usually understood. 28. For a further discussion of congruence, see Ruttan (1982) and Fox (1987). 490 Scoring and Other Shortcut Approaches The last term, in square brackets, is the inverse of the research intensity ratio, used for congruence. Congruence ignores some key factors that affect the ranking of programs according to Ni (including the probability of re­ search success, likely adoption rates, and likely research-induced productivi­ ty gains) as well as the timing and discounting aspects. At best, congruence would yield the same ranking if the combined effects of all the left-out factors were equal among programs. As discussed above, Nis cannot be used beyond ranking for resource allocation. The same restriction applies to using congruence as a rule of thumb to allocate research resources. Guidelines Implied by the Economic Surplus Model The congruence rule looks at two aspects of the problem of allocating resources to agricultural research: (a) the baseline value of output and (b) the baseline value of research support. As shown above, several other factors can be considered in addition to the rudimentary information requirements of the congruence approach. These additional considerations might be involved in a formal comparison of alternatives or in a subjective decision-making format. Box 7.1 qualitatively summarizes the principal determinants of the ex­ pected net present value of research that can be used to assess research program alternatives in informal or formal processes.29 Peer Review A number of methods require individuals to compare one proposal either to another proposal or to a group of alternative proposals and to indicate their preference (or sometimes the strength of preference) for the chosen alterna­ tive.30 When more than one individual's opinion is sought, a technique must be chosen for eliciting a group opinion. Group techniques can be used for scoring or for cardinal or ordinal ranking of research projects, program areas, criteria, research needs, or research objecti ves or for determining the weights to be placed on criteria in research project evaluation.31 29. A discussion of the principles that underlie these guidelines is presented in chapters 2, 4, and 5. See also Binswanger and Ryan (1977), Ruttan (1982), Norton and Ganoza (1986), and Lloyd, Harris and Tribe (1990). 30. Shumway (1977) discusses several one-dimensional ranking methods, such as Q-sort, ranking, rating, paired comparisons, dollar metric, standard gamble, and successive comparisons. They are simple and easy to implement, especially when there are few items to he compared. In each, a judge compares the overall subjective worth of one item to one or more other items on one criterion. One-dimensional ranking methods are used to group (Q-sort) or rank projects, program areas, criteria, or objectives. 31. Shumway (1 CJ77) and Scobie and Jardine (1988) discuss several of these techniques, including a committee approach, chain of command, the Delphi method, the weighted average method, nominal group technique, and interpretive structural modeling. Scoring and Other Shortcut Approaches 491 Box 7.1: Guidelines for Agricultural Research Priority Setting Market failure: Priorities for public research funding should be in those areas in which there are high social returns and low private returns. Where market failure exists but returns accrue mainly to the private sector, forms of government interven­ tion other than direct funding become appropriate (Lloyd, Harris and Tribe 1990). Efficiency: Domestic net benefits from research are higher • the larger the total preresearch value of production of the commodity • the faster the expected growth of the industry • the greater the proportional reduction in unit costs induced by research • the higher the probability of research success • the higher the ceiling rate of adoption domestically • the faster the adoption of the research results domestically • the lower the adoption of research results in other countries • the sooner the reduction in unit cost is realized • the lower the rate of research depreciation • the lower the research cost • the lower the interest rate • the lower the opportunity cost of government funds • the smaller the domestic production as a share of global production of the commodity • the greater the effect of research on reducing distorting effects of price policies • the greater the effect of research on reducing distorting effects of externalities Net domestic research benefits are not affected by many price-distorting policies, although the distribution of benefits tends to be shifted towards those being assisted by the price policy. Distribution: Research is a relatively blunt tool for meeting distributional objec­ tives, such as income distribution or nutrition, compared with other policy instru­ ments such as taxes and subsidies. Research tends to be both (a) an ineffective and (b) a very costly method for pursuing social policy objectives. Domestic "producer" benefits are increased as a share of total benefits • the higher the domestic price elasticity of demand for the commodity • the lower the price elasticity of supply of the commodity • the smaller the domestic production as a share of global production of the commodity • when the technology applies farther down the marketing chain towards farm­ level production • the lower the adoption of research results in other countries • the faster the adoption of research results domestically relative to other countries. 492 Scoring and Other Shortcut Approaches One of the most widely used approaches to research priority setting is peer review. In this approach. various subjective procedures may be used either by individual judges or in some group decision-making process to rank program or project proposals and to recommend decisions. Peer review is best suited for assessing the scientific merit of proposals because peers are typically best equipped to judge that aspect of a proposal and less well equipped to judge economic merit. For that reason. peer review is most useful for decisions about indi vidual projects rather than broad programs. As such. peer review is a complement to formal economic approaches. which apply at the strategic level and help to define the boundaries within which more detailed project decisions can be made. drawing upon peer reviews. At the same time. an institutionalized peer review program is an important source of insights about technical parameters that feed directly into the measurement of the economic consequences of research. 7.3 Conclusion Scoring is a way of developing shortcut indicators of the consequences of research and. perhaps. for weighting the estimated contributions of research to various stated objectives in order to derive a summary measure of the effects of research. Occasionally a simplified scoring model can be em­ ployed to assist research administrators with priority setting in situations where research resources are being allocated across large numbers of com­ modity research programs or research areas and where resources are not available to allow a more complete analysis. These simple models are less data-intensive than those that include more complete economic surplus models. With careful selection and manipulation of criteria. the models can. in some circumstances. provide results that may be roughly consistent with models that include more complete measures. More confidence can be placed in the results when the measures derived in the analysis correspond more closely to more complete economic surplus measures. Simplified scoring procedures must be used with caution. Even careful applications of these models generally use crude measures of the economic consequences of research. As a result. the ranking according to the efficiency criteria used in scoring models may be very different from the ranking according to efficiency as measured in an economic surplus model. Like­ wise. the distributional criteria used in scoring models may also differ substantially from more complete distributional measures. When multiple objectives are being considered. an overall ranking requires assigning weights. and we have shown that assigning weights is very tricky. Indeed. Scoring and Other Shortcut Approaches 493 we have suggested that the most reasonable way to define weights is to use the rankings to do so, and that means that the process of weighting and ranking is circular. Also, use of these models should be tempered by the fact that research tends to be an inefficient policy mechanism for meeting dis­ tributional objectives, as discussed earlier. Finally, experience with studying and using scoring models over the past few years has taught us that the simple ad hoc weighting of indicators that make no attempt to approximate economic surplus measures provides results that are of little, if any, use in setting research priorities. Unless attempts are made to calculate at least rough efficiency indices along the lines presented in this chapter, a research system is better advised not to score research programs. It is often better to use informal, but still structured judgments based on an economic way of thinking about research. In a similar vein, precedence and congruence approaches are likely to be poor decision rules. There is really no substitute for the economic surplus model. In the worst of all worlds, when quantitative analysis is ruled out, its qualitative results provide a better guide to allocating research resources than do any simple mechanical rules or shortcut evaluation procedures. 494 Scoring and Other Shortcut Approaches Appendix A7.1: The Problem of Units in Eliciting Weights Suppose efficiency is weighted 100 and additional weights of 60 are gi ven to region X and 40 to small farms, and the weights are rescaled by dividing by their sum (200) so that the rescaled weights total 1.0. For research on commodity i, the overall score, Vj , is equal to the sum of (a) the efficiency index for the commodity, Nj' multiplied by O.S (i.e., 1001 200), (b) the efficiency index for the commodity in region X, NX,j , multiplied by 0.3 (i.e., 60/200), and (c) the small farm index for the commodity, SFj, multiplied by 0.2 (i.e., 40/200): Vj =O .SNj + 0.3NX,j + 0.2SFj It might be inferred that this is a scoring rule that puts most weight on efficiency and relatively little weight on the distributional consequences for, say, small farmers. But whether that is so depends entirely on the units chosen for the criteria. To see this, suppose the overall efficiency index (e.g., value of maximum annual research benefits per cost of research over five years, as in equation 7.2) ranges across commodity programs from 10 to 20, and the index of efficiency benefits to region X is a part of overall efficiency benefits that ranges from S to 10. And suppose the small-farm index measures thefraction of output of a commodity produced by small farms, ranging across commod­ ities from 0.1 to 0.5. In such a situation the ranking will be largely unaffected by the small-farm index. As shown in section 1 of table A 7.1, the overall scores in column a indicate that the ranking is program A > B > C. And, in fact this is the ranking implied by efficiency alone, the criterion ostensibly receiving the highest weight. But suppose the small-farm index is instead the number of small farmers producing a commodity, ranging from 200 to 10,000. Clearly, with this latter measure, the ranking based on the overall scores in column b will depend only on the small-farm index. Efficiency and regional location will have no effective weight. Under option a for measuring the small-farm criterion, the ranking of programs is A > B > C, and the overall score for the highest-ranked program, A, is about SO% greater than that for the lowest-ranked, C. Under option b, the ranking is reversed (i.e., C > B > A), and the score for the highest ranking, C, is almost 40 times that for the lowest ranking, A. Two factors have led to this outcome. First, the SFj criteria were initially much smaller numbers than those for the others, but in the second case they were much larger numbers. Thus, in the second case, they had more effective weight in terms of their influence on the overall ranking. The gross inconsis­ tency of rankings and sensitivity to units may be reduced by normalizing all Table A7.1: The Effects of Units in Scoring .j::,. ~ Criteria Overall performance Commodity SFi Overall score Rank (i) N; Nx,i (a) (b) (a) (b) (a) (b) (1) Sensitivity of scoring to units A 20 5 0.1 200 11.52 51.5 3 B 15 10 0.3 1,000 10.56 210.5 2 2 C 10 8 0.5 10,000 7.5 2007.4 3 (2) Effects ofnonnalizing criteria A 20/45 5123 0.110.9 200111,200 0.2097 0.2910 3 3 B 15/45 10123 0.3/0.9 1,000111,200 0.3638 0.3150 2 C 10/45 8/23 0.5/0.9 10,000111,000 0.3266 0.3940 2 (3) Effects of using criteria as ordinal ranks rather thcln cardinal measures A 3 1 2.0 2.0 2 2 B 2 3 2 2 2.3 2.3 1 C 2 3 3 1.7 1.7 3 3 496 Scoring and Other Shortcut Approaches criteria to lie on a similar scale, say from 0 to 1. Second, the variance of the measures matters. The variance of the small-farm numbers is much greater than the variances of the other criteria, which were in tum greater than the variance of the small-farm share. In ordinal rankings, components with larger means tend to matter more, everything else equal. That is obvious. But components with larger variances also get greater effective weight (e.g., if all other criteria were equal among programs, then all of the effecti ve weight would be on the farm size criterion in the ranking, regardless of the elicited weights). Normalizing will reduce (or, at least, conceal) problems arising from gross disparities in units, but it may still lead to inappropriate rankings. To show this, in section 2 of table A 7.1, the criteria from section 1 are normal­ ized before the scores are computed. As it turns out, in this case, yet a third ranking is obtained using the normalized criteria when the fraction of output produced on small farms in column a is used as the small-farm index (i.e., B > C > A). When the small-farm number is used instead, normalizing the units does not affect the ranking (i.e., comparing column b in sections 1 and 2 of table A 7.1, the ranking is the same). The normalization did succeed in reducing the disparity among the criteria and, thus, the variation among the overall scores (i.e., they all were between 0.2 and 0.4). Thus, using normal­ ized scores might be misunderstood to imply that the alternatives are not very different, even when they differ a lot. An alternative normalization is to replace the performance measures in sections 1 and 2 of table A7.1 with corresponding rankings from highest (3) to lowest (1) and then computing an overall, weighted score. Section 3 in table A 7.1 contains the results from doing so. In section 3 it can be seen that the variance of the performance measures (i.e., the rankings) are equal for all of the criteria. Also, the overall rankings are identical between the two alternative measures of the farm size criterion in columns a and b (since, by chance, the rankings of commodities A < B < C were identical for both the fraction of output produced on small farms and the number of small farms producing the commodity). Now, a fourth overall ranking of commodities is revealed when ranks of criteria are used instead of the cardinal values: B > A > C. Which of these four rankings should be preferred remains unclear. Table A7.2: Use of the Scoring Model to Determine Agricultural Research Priorities by Commodity -I::.. IC 'l Net Net Small- Value Yield Prob. of Adoption Efficiency efficiency efficiency Region farm Weighted Weighted Commodity prod. a change b success C rated index' index f rank' index h index i indexi rank k Barley Cotton Maize Millet Potatoes Rice Sorghum Wheat a Pi Qi = value of production without research for commodity i. b E (yflAX) = maximum proportional reduction in per unit cost or proportional yield increase presuming the research is successful. C Pi = probability of research success. d A flAX = maximum proportion of fanners likely to adopt the new technology. • Gi = A flAX p; E (yflAX) P; Q; = gross efficiency index = (adoption rate) x (probability of success) x (proportionate yield change) x (value of production). f N; = G;IR ; = net efficiency index = (gross efficiency index) I (research cost or number of scientists). g Priority rank based on size of N;. h Regional index = N; x fraction of production produced in the priority region. i Small-farm index = N; x share of the commodity produced on small farms. Weighted index = (weight on efficiency objective x net efficiency index) + (weight on regional objective x regional index) + (weight on small-farm objective x small-farm index). k Weighted rank = priority rank based on weighted index. Table A 7.3: Disaggregating the Scoring Model to Include Both Commodities and Research Program Areas ~ ~ Value Total Yield change by Probability of success Adoption rate of Net Small- of yield programC by program d program" Efficiency index f effic. farm Wt'd I Commodity prod. a change b PB CP CM OT PB CP CM OT PB CP CM OT PB CP CM OT Index g indexh index I Barley Cotton Maize Millet Potatoes Rice Sorghum Wheat a Pi Qi = value of production without research for commodity i. b E(lfAX) = maximum proportional yield change due to research presuming the research is successful. C Proportional yield change due to plant breeding (PB), crop protection (cp), crop management (CM), and other (OT), calculated by multiplying the total yield change due to research by the proportion attributable to each program area. d Probability of obtaining the yield change. " Maximum proportion of fanners that will adopt the research results. f Gross efficiency index for each research program area =( value of prod.) x (yield change by program) x (prob. of success by program) x(maximum adoption rate by program). g Net efficiency index = sum of program area gross efficiency indexes I research cost (or number of scientists). h SmaIl-farm index = (Net efficiency index) x (percent small farmers/lOO). A small farm index could also be developed for each program area. I Weighted index = (Net efficiency index x corresponding weight) + (SmaIl-farm index x corresponding weight). A weighted index could also be developed for each program area. Part IV Overview and Assessment 8 Assessment and Conclusion Research resource allocation questions can take several forms. What is the appropriate total amount to spend on agricultural research and, relatedly, how should it be financed (i.e., by whom and through what mechanism)? Given the total resources for research, how much should be allocated to different com­ modity programs and noncommodity programs and to disciplinary (and other) components within programs? Choices about the regional focus and problem orientation of research programs may be involved as well. Another set of questions concern the input mix used in research. How should the budget be allocated between physical capital-investment programs, human resources, and other operating expenses? Some of these questions are rather open-ended, but more specific questions lend themselves to more concrete answers. For instance, what are the implications of a 10% increase (or decrease) in the overall budget? Should all programs grow or be cut in proportion, or should some fare better than others? Or, is it time to cut out some programs, or certain components of programs, altogether and invest the money elsewhere? Some questions concern marginal decisions while others relate to all-or-nothing decisions, requiring different types of information. This book has presented and evaluated various procedures for evaluating research and setting priorities.' These are procedures that can be used to help I. The disaJssion in this book is confined to economic evaluation procedures. There are many other procedures that have been proposed and advocated for research evaluation and priority setting. Some such procedures measure research "inputs" (e.g., scientists' time, resources, organizational structures, and procedures), intermediate "outputs" (e.g., pUblications, new varieties, or experiments conducted or com­ pleted), or components that influence economic effects (e.g., adoption rate). These methods rarely attempt to establish a systematic causal relationship between the costs and benefits of research, and as a result, they are most unlikely to yield any meaningful indications of the economic effects of research. Because they are 501 502 Assessment and Conclusion answer the kinds of questions listed above. Earlier chapters describe the theory underlying these procedures and also provided some guidance on applying them. Our key premise is that any procedure used should draw on a consistent conceptual framework and yet be tailored to the characteristics of the individual research system. The procedures discussed here have been examined for their consistency with economic theory, ease of implementation, and appropriate­ ness for the job at hand. In this chapter we briefly recap the earlier discussions and review the advantages and disadvantages of the various alternatives. 8.1 Conceptual Framework Revisited Agricultural research involves the investment of scarce resources in the production of knowledge to increase future agricultural productivity and, thereby, to contribute to a range of economic and social objectives. The main goal of agricultural research is usually enhanced economic efficiency. There are good reasons for this being the exclusive goal, but equity and security may also be important secondary goals. These three goals may be described, respectively, as increased total income, improved income distribution, and reduced income variability. They are multidimensional, particularly equity and security. They can include such objectives as greater well-being for low-income groups, conservation of natural resources, and increased na­ tional self-reliance. Most "other" objectives are really efficiency or distributional objectives. For instance, concern with "sustainability" or, more concretely, natural resource conservation may reflect either a concern that efficiency requires accounting properly for changes in the stocks of natural resources or a distributional concern about intergenerational equity, or concerns about both efficiency and distribution. The effects of research on some of these objec­ tives are difficult to measure, but most are amenable to economic analysis. Contributions of research to economic efficiency and the distribution of benefits can be measured as the net present value of research-induced changes in economic surplus. The size of these economic gains depends on the size of the research-induced shift in the product supply curve, the nature of the shift, elasticities of supply and demand, the pattern of trade in the commodity, and market distortions. The major determinants of net research benefits are the value of production, probability of research success, size and not systematic, they offer little prospect of establishing the link between changes in the quantity of resources going into research, their deployment, and likely benefits. Therefore, they are not useful for infonning allocation decisions. In this book we focus on methods and ITIea<;ures that are less likely to be vulnerable to this criticism Assessment and Conclusion 503 timing of per unit cost reductions or yield increases if the research is successful, the discount rate, and the cost of the research. Measurement of research benefits is complicated because (a) benefits are spread geographically and vertically in markets for goods and services, (b) research can affect product quality, (c) some research is not commodity oriented (and some commodity-oriented research leads to disembodied tech­ nical changes), (d) some research is aimed at modifying institutions, (e) some research generates externalities, and (f) research may be relatively basic or very applied. Although these factors present measurement difficul­ ties, research evaluation and priority-setting methods should attempt to use measures that approximate changes in economic surplus. The contributions of research to security objectives may involve calculat­ ing how research reduces the variability of agricultural income. Such calcu­ lations, however, are difficult, and agricultural research is a blunt instrument compared with other policy tools for achieving security objectives. Agricul­ tural research is a blunt instrument for achieving distributional objectives as well, even though it does have distributional consequences. Policymakers usually have multiple policy tools at their disposal, and the least-cost solu­ tion for meeting societal objectives might not involve research or might best be achieved using some combination of research and other policies. Hence, an important role for economists involved in research priority-setting analy­ sis is to inform decision makers about the costs of income foregone as a result of biasing the research portfolio in the pursuit of nonefficiency objectives. We recommend incorporating measures of economic surplus changes in any procedure for evaluating agricultural research. The presence of multiple objectives complicates the evaluation and priority-setting process but does not require that we abandon economic principles. Since the comparative advantage of research is primarily in meeting the efficiency objective, great care is needed when an attempt is made to weight alternative objectives in research evaluation or priority setting. 8.2 Deciding on the Method and Degree of Detail Several factors influence the choice of a method and the degree of detail for a research evaluation or priority-setting analysis. The most important of these is the type of question to be answered and the purpose of the analysis as dictated by the problem at hand. Some are operational considerations, such as the data available, the financial and other resources available for the analysis, and the skills of the analyst. Finally, there is the completeness and consistency of the procedures in relation to the conceptual economic frame- 504 Assessment and Conclusion work described in chapter 2. Each procedure has its advantages and disad­ vantages and no one approach is best for every situation. In some cases, alternative procedures can be combined. The degree of detail such as analytical sophistication or commodity coverage can be increased over time as the procedures become more fully integrated into the decision-making processes of a particular research sys­ tem. Indeed, one of the major advantages of institutionalizing an economic approach to allocating research resources is that it supports a progressive accumulation of data, analytical experience, and the ability of policymakers and others to make effective use of the information. Thus, over time, the costs of decision making fall and the decisions get better. Sporadic ad hoc reviews do not generate capital resources in the form of human capital and data and so are less able to take advantage of prior investments. 8.2.1 Methods for Ex Post Evaluation of Research Programs Assessments of previous research are frequently desired by research directors (a) to justify research budgets and (b) as a guide to areas of likely future research payoffs. These assessments can be made in the aggregate or for specific research programs. While we are treating the evaluation of research that has been done as an "ex post" analysis, some of the effects might not yet be realized, calling for the use of methods described under the heading "ex ante" assessment in section 8.2.2 and elsewhere. Aggregate Research Programs Econometric approaches are generally best for ex post evaluations of aggre­ gate agricultural research programs if the quantity and quality of the data allow the use of statistical methods. Such evaluations can reveal the productivity or efficiency benefits of agricultural research and the effects of research on the structure of production. A knowledgeable analyst with good data can use the results of a production-function, cost-function, or profit-function model to statistically test the size and significance of estimated research impacts. The effects of research on economies of size or the input bias of research can also sometimes be examined econometrically. Often a production-function ap­ proach is best for such analysis, if multicollinearity problems are not severe. However, in some cases, other approaches (i.e., productivity, cost, or profit functions) may be preferred, depending on the data available and other factors. Countries with adequate data for conducting an econometric analysis are still primarily those with statistical reporting services that were established long ago. Although the number of countries with this capability continues to Assessment and Conclusion 505 grow, many African nations and several Asian and Latin American countries are excluded from this group. There is little point in proceeding with an econometric analysis unless 25 to 30 years of data are available on quantities (and, perhaps, also prices) of outputs and inputs, along with data on research and extension expenditures going back a further 20 years or so. It is some­ times possible to proceed with shorter time series if they can be combined with cross-sectional (i.e., regional or provincial) data. Research Programsfor Individual Commodities An econometric approach that enables estimation (or derivation) of sup­ ply functions is usually preferred for the ex post evaluation of individual commodity research programs. In this approach, coefficients on research variables can be employed to estimate shifts in supply curves. These shifts (the UK-factor" from the equations in chapters 4 and 5), as well as the supply elasticities generated in the estimation, can be incorporated in economic surplus models. This permits the distributional and efficiency effects to be estimated statistically. Such approaches include the direct estimation of single-equation supply models and the estimation of production, cost, and profit functions from which supply functions can be derived. Which of these approaches is chosen depends on the characteristics of the problem at hand - especially the availability of data and the degree of commodity interdependence. As with aggregate analy­ sis, this set of approaches is recommended only in situations for which adequate historical data are available. Adjustments can be made to the esti­ mated shifts in the supply curve, based on scientists' opinions or experimental data, before they are used in an economic surplus analysis. Still, the econome­ trically estimated relationships between previous research expenditures and supply shifts provide useful standards for comparison as a benchmark. Implementing this combination of methods requires that the supply models be estimated first. Estimated coefficients on the research variables can be used to calculate economic surplus effects and rates of return to research and to gain quantitative insights into the distributional consequences of research (and these may help to justify research budgets). When adequate time-series data are not available, an economic surplus approach can be used that relies on experimen­ tal data and the opinions of scientists and extension workers to estimate the per unit cost changes (or yield improvements) and adoption rates for the key technologies that have been developed over the relevant time period. The degree of detail to include in the economic surplus model depends on the purpose ofthe analysis, available information on factors such as agricul­ tural policies and agroecological zones, the detail of the analysis (e.g., the 506 Assessment and Conclusion number of commodity research programs involved), and the resources avail­ able. The estimated distributional effects of research are particularly sensi­ tive to the degree of detail in the model. As spatially disaggregated market-related and scientific data become increasingly available in many countries, the potential for accounting for the distribution of benefits across regions increases. Likewise, with the development of computerized research evaluation programs such as Dream© (see chapter 5), the effort required to do the analysis has been greatly reduced. The major constraints are the availability of information and the time needed to collect and construct the data sets required and to interpret and present the results in ways that are meaningful for decision makers. Noncommodity Research Programs Noncommodity research programs might focus on a disciplinary area of work (e.g., genetics), on a problem that involves multiple commodities (e.g., pest management), on natural resources and their management (e.g., soil science), or on particular factor markets (e.g., farm labor). The ex post evaluation of disciplinary research program areas relies almost exclusively on direct application of economic surplus methods. If the purpose of the analysis is to evaluate the effects of several disciplinary programs, one approach is to apportion the benefits estimated for each commodity research program to the several disciplines involved, and then to sum the benefits across commodities for each discipline. The benefits of certain program areas that are wholly or partially non­ commodity focused (e.g., parts of natural resource management, agricultural economics, agroecology, and soil science) may be difficult to measure in an economic surplus model. Some noncommodity research areas are difficult to evaluate because they affect multiple commodities, which implies a need for joint estimation of benefits. However, other noncommodity programs that focus on (or have their primary effects in) a market for a particular factor used in the production of several commodities can be analyzed in the context of the single market for that factor, using vertically dis aggregated market models. Certain components of these programs can be evaluated; others are the subject of current research into designing appropriate research-evaluation procedures. 8.2.2 Methods for Ex Ante Research Evaluation Ex ante research evaluation and priority-setting analyses that relate to research yet to be done can use results from econometric analyses to provide a benchmark for the magnitude of supply-curve shifts in economic surplus Assessment and Conclusion 507 models. However, the general purpose of these analyses is to assist with priority setting and resource allocation across a large set of individual commodities and disciplinary research program areas for research yet to be done. In most cases this calls for an economic surplus model to be used without econometric analysis. An economic surplus model can incorporate the geographical spread of research results across regions in the country, the complexity of pricing and other agricultural policies, the division of commodity research programs into their technology types (or research program areas), and the effects of re­ search at various stages in the marketing chain. Many research systems will use a simpler economic surplus method in which market and policy differ­ ences are taken into account and effects are disaggregated among horizon­ tally (but not vertically) related markets. In this type of analysis, certain distributional effects can be calculated (e.g., research benefits accruing to small farms, to producers, to consumers, and to government) but other effects usually cannot (e.g., research benefits to processors). 8.2.3 Setting Priorities As discussed in chapter 5, the economic surplus or net present value measures may be broad indicators of research priorities, but usually some additional work is required before they can be used for allocating research resources. Unless the additional work is done, the typical measures (i.e., the present value of gross or net research benefits or the corresponding internal rates of return) provide at best only an ordinal ranking of research program areas. But they ought not to be used even in this way, because typically they have not taken suitable account of the scales of the programs, which can affect the program rankings. For example, the rice program in an Asian country would be expected to have a much larger total NPV than almost any other program simply because of its size, even though some smaller pro­ grams might be much more productive in terms of benefits per unit of research. Programs could be ranked more meaningfully according to NPV per unit of constraint, such as per scientist or per research dollar. Even still, the ranking does not provide much information about priorities. The proper use of the outcome from an economic surplus analysis of ex ante research benefits is as data for an optimization process that can accommodate the system's objectives, the constraints, and perhaps, the effects of different scales of research programs on their marginal benefits. Chapter 5 discusses some relatively informal approaches for using the results of economic surplus analysis to establish the marginal effects of research program alternatives in order to consider specific allocations of research resources. Chapter 6 shows 508 Assessment and Conclusion how to use the same, or similar, information in more formal optimization approaches using mathematical-programming models. In chapter 7, we illustrate that scoring methods can provide a shortcut approximation to the same problem but that the quality of the approximation rests on how closely the criteria in the scoring model approximate reasonable proxies for measures of performance against program objectives. Moreover, the process involved in applying scoring methods does not correspond closely to the actual constrained optimization problem being faced (i.e., scoring at best ranks current programs; it typically does not compare alter­ native allocations of resources and does not consider the resource con­ straints). Hence, scoring is not as good a complement to economic surplus as mathematical-programming methods are. Indeed, as we showed in chap­ ter 7, the results of scoring can be worthless in decision making. The research program rankings from scoring models are strictly ordinal, with no im­ plications for optimal decisions about resource allocation. Sometimes, be­ cause the units in scoring procedures are incompatible with economic surplus measures, the rankings from scoring models are wrong. In some cases, a simple variant of economic surplus can be used to arrive at a first approximation of research priorities and then a more detailed analysis can be applied to the highest-priority commodities (i.e., those with the highest NPV per unit of constraint). At the other end of the spectrum, a relatively complete variant of the economic surplus model, including an explicit consid­ eration of alternative funding levels for programs, could be applied. The results could be embedded in a mathematical-programming model to explore the opportunity cost of placing alternative weights on objectives and to provide information on research resource allocation, given the constraints on finances, human resources, and facilities facing the research system. In summary, the appropriate research evaluation or priority-setting proce­ dure (or degree of detail) for a particular situation depends on the purpose of the analysis and the resources available. All procedures involve subjective judgments, particularly in ex ante analysis, but by organizing the information in a manner consistent with the conceptual economic framework presented in chapter 2, the odds of providing accurate and transparent assessments of research contributions to objectives are increased. 8.2.4 Selecting Projects or Experiments The research evaluation and priority-setting procedures described in this book are designed for assessing research programs at a strategic level. They are less applicable at the project level, both because of the costs involved relative to the benefits of evaluating a large number of potential projects and Assessment and Conclusion 509 because priorities at the project level are usually influenced primarily by technical questions (as opposed to economic ones).2 Of course, some indi­ vidual projects (particularly large ones) may warrant quantitati ve evaluation. In general, however, although the methods could be applied to individual research projects or experiments, in most cases the information yielded is unlikely to justify the costs. Excessive use of formalized procedures for research evaluation and priority setting could even stifle ingenuity, serendip­ ity, and scientific entrepreneurship. Once decisions have been made on targets for numbers and types of scientists and supporting resources for each program, planning projects and tasks within program areas is typically accomplished through a system of technical committees using peer review and specific criteria. Producer input on these committees can be very useful. The social relevance and technical merit of the projects and tasks proposed by scientists are influenced by the extent to which the scientists are made aware of research system objectives and are provided with incentives and rewards based on performance as measured against appropriate criteria. Micromanagement of scientists can be counterproductive. But delegation of authority for detailed decisions does not mean that those making them ought to be free to ignore their economic implications. On the contrary, it may be especially important for senior management in a scientific agency to ensure that researchers are acquainted with the agency's objectives, and the economic arguments related to how they are achieved, in order to ensure that decentralized decisions are made well. An understanding of simple effi­ ciency indexes (such as equations 7.1 and 7.2) and the guidelines laid out in box 7.1 may be sufficient for these purposes. 8.3 Areas for Future Model Development and Application The methods for research evaluation and priority setting are deficient in some areas, and these are potentially fruitful areas for further model devel­ opment or for refined application in research evaluation analyses. Most research evaluation and priority-setting analyses are undertaken in the pres­ ence of multiple social objectives. Research directors are aware of these multiple objectives, would often like to know the potential contributions of research to each of the objectives, and must decide on the weights to place on the objectives when allocating research resources. We have argued that agricultural research can contribute to the achievement of a wide variety of 2 As Anderson and Hardaker (1992) point out, for a research project manager costs are largely fixed, so the task of increasing efficiency is reduced to that of improving the efficiency of research. 510 Assessment and Conclusion objectives, but that it is a blunt, costly, and often ineffective instrument for achieving nonefficiency objectives compared with other policy instruments. Additional research is needed to demonstrate to policymakers the opportu­ nity costs of achieving their multiple objectives through various combina­ tions of research, tax and subsidy, and other policies. 31f practical procedures for this type of analysis can be further developed, it may reduce the tendency to naively choose among research alternatives as if research were the only (or best) means of achieving the objectives. Further research is also needed on how best to incorporate risk and uncertainty in research evaluation and priority setting. Agricultural produc­ tion is risky. The research process itself is also risky. In addition, the analysis of benefits from research involves a number of uncertain parameters: the impact and adoption of research results is uncertain, the natural, economic, political environments in which agricultural commodities are produced are uncertain, and it is difficult to measure the weights that different policymak­ ers place on stable income from agriculture (the degree of risk aversion) or on food security. In chapters 5 and 7, we suggest ways of incorporating the riskiness of research into the evaluation process, but the approaches so described are rudimentary and in need of further refinement. Anderson (1991) reviews some of the simple approaches that have been employed to incorporate risk in research planning studies, but concludes that the "role of risk in such decisionmaking is a sadly neglected field" (p. 127). In order to measure the benefits from agricultural research in the presence of environmental externalities and to accommodate concerns about natural resource conservation, additional research and application of these methods are needed. We suggested a simple conceptual model in chapter 4, but that model needs considerable elaboration and application in concrete situations. The opportunity cost of using research versus other policy instruments for achieving goals related to natural resource conservation cannot be fully assessed without applying economic models that incorporate research in­ vestments. We have argued that "sustainability" is best understood in rela­ tion to more fundamental concerns with efficiency and equity. Thus, incorporating environmental issues in research-benefit studies is likely to involve further development of the economic surplus approach. A few studies have attempted to evaluate research that influences product quality, but more work is required on that topic as well- particularly as the demand for product quality can be expected to rise with increasing incomes (Senauer, Asp and Kinsey 1991). Changes in product quality can occur at the primary or processing levels and can be aimed at reducing the cost of 3. In this regard, there may be a role for elaboration of the mathematical-programming approach discussed in chapter 6. Assessment and Conclusion 511 producing the product at a specified quality. Per unit cost reductions are usually modeled as downward shifts in supply curves. Previous work has represented changes in quality by shifts in demand curves as if tastes and preferences change (Unnevehr 1986; Voon and Edwards 1991 a, 1992). The issue is complicated because of the potential substitution effects that reduce demand for lower-quality products as higher-quality products are made available through research. Models are needed that represent changes in quality in a way that is conceptually defensible yet implementable. A few studies have attempted to conceptualize and empirically measure the impact of social science research in agriculture (e.g., Norton and Schuh 1981; Lindner 1987); however, little progress has been made. Measuring the effects of research aimed at evaluating institutions (e.g., price policies) is particularly challenging because (a) studies are often directed at a particular institutional change and (b) even if an institutional change occurs, it is often nearly impossible to ascertain the contribution of research compared with political or other factors in influencing the change. Also, social science research is directed at such diverse topics, most of which do not directly shift product supply curves, that aggregate evaluation is extremely difficult. However, it may be possible to evaluate particular categories of social science research using tools that place a value on the new information corning from such research. There is a strong need for additional study of the general-equilibrium effects of research. The implications of focusing only on the partial-equilib­ rium economic surplus measurement of research benefits were briefly men­ tioned in chapters 2 and 4. Elaboration and application of the approach suggested by Martin and Alston (1992, 1994) offers some promise for practical, empirical research evaluations, at least for ex post analysis. Finally, additional work is needed on means for linking the results of research evaluation and priority-setting studies to the development of plans for investing in human resources and facilities, and to decisions on operating budgets. Some research directors have been more successful than others in translating the results of these studies into such plans, decisions, and actions. We need more information on how to maximize the chances that follow-up to research evaluation and priority setting occurs. 8.4 Conclusion The demand for improved methods for evaluating and setting priorities for agricultural research has grown in recent years. Research evaluation and priority-setting procedures must be rigorous yet cost effective. The key to successful research evaluation and priority setting is not simply the method 512 Assessment and Conclusion chosen, but how that chosen method is implemented. We suggest that for many situations, a partial-equilibrium framework based on the concept of economic surplus is the soundest and most practical conceptual framework. The methods discussed in this book can be made consistent with this concept at various levels of approximation. Some general-equilibrium effects can, and in some instances should, be included as well. The application of theoretically consistent measures can yield information that is useful in decision making. Perhaps the major benefit from a process of research program review, evaluation, and priority setting is that the partici pants gain a clearer view of what they are trying to achieve - and how best to get there. Scientists and policymakers will make better decisions as they develop an economic way of thinking about research investment choices. It is especially important that they develop an economic way of thinking that has at its foundations a theoretically consistent and defensible economic structure. This implies institutionalizing a process of evaluation (which does not imply evaluating everything in sight) in order to develop information and incentives that enhance the chance that the invisible hand will do its job in allocating scarce scientific resources. Such an outcome will be less likely with one-off evaluation or priority-setting exercises in which research system personnel are not actively involved. These considerations reinforce the value of a consistent, economically sound approach. This is not a book of rules of thumb for making allocation decisions about research resources. It is not a black box, nor is it a gratuitous complication of the intuitive. Rather, it is an attempt to establish a defensible link between research and its objectives. It is necessary to apply the principles expounded here explicit! y to a particular problem in order to make meaningful progress. The methods might seem demanding, but the problems are inherently diffi­ cult. We believe that the effort is worthwhile in many cases, but usually a real effort is necessary. If the applied work is not done with care, one runs the risk of spurious attribution of the past, actual, and future potential impact of research, which is of questionable value for defending budgets and of no use for economizing on scarce scientific, human, and natural resources. By no means do we countenance the wholesale adoption of highly sophis­ ticated, mechanical approaches to evaluation and priority setting. On the other hand, it is all too easy to let an application of these procedures and processes slip from the simple to the simplistic. While resource constraints, in practice, often necessitate the use of simplifying assumptions and proce­ dures, the fundamental principles developed throughout this book, particu­ larly in chapter 2, should not be forsaken. An overriding consideration is that the benefits from any research evalu­ ation or priority-setting study depend on how the results are used by the Assessment and Conclusion 513 client. Whatever the method chosen, research directors, councils, or boards must find the results of the analysis useful for making decisions on budgets, facilities, and people. Thus, the analyst conducting the research evaluation or priority-setting analysis can view his or her services (and models) as a bridge between the raw data pertaining to the economic context and the technological opportunities, and the information needed by those making decisions on strategic priorities or research budgets. References Adams, J. "Fundamental Stocks of Knowl­ ican Journal ofA gricultural Economics edge and Productivity Growth." Jour­ Vol. 68, No.1 (February 1986): 1-9. nal ofP olitical Economy Vol. 98, No.4 Alston, J.M. "Research Benefits in a Mul­ (August 1990): 673-702. timarket Setting: A Review." Review of Adusei, E. "Evaluation of the Importance Marketing and Agricultural Economics and Magnitude of Agricultural Mainte­ Vol. 59, No.1 (April 1991): 23-52. nance Research in the United States." Alston, J.M. and J.A. Chalfant. "Consumer Unpublished Ph.D. Thesis, Virginia Demand Analysis According to GARP." Polytechnic Institute and State Univer­ North Eastern Journal of Agricultural sity, Blacksburg, Virginia, 1988. and Resource Economics Vol. 21,No. 2 Adusei, E.O. and G.W. Norton. 'The Mag­ (October 1992): 127-39. nitude of Agricultural Maintenance Re­ Alston, J.M. and J.W. Freebairn. "Producer search in the USA." Journal of Price Equalization." Review of Market­ Production Agriculture Vol. 3, No.1 ing and Agricultural Economics Vol. (January-March 1990): 1-{). 56, No.3 (December 1988): 306-39. Akino, M. and Y. Hayami. "Efficiency and Alston, J.M. and B.H. Hurd. "Some Ne­ Equity in Public Research: Rice Breed­ glected Social Costs of Government ing in Japan's Economic Development." Spending in Farms Programs." Ameri­ American Journal of Agricultural Eco­ can Journal ofA gricultural Economics nomics Vol. 57, No.1 (February 1975): Vol. 72, No.1 (February 1990): 149- 1-10. 56. Allen, R.c. and W.E. Diewert. "Direct Ver­ Alston, J.M. and D.M. Larson. Precision vs sus Implicit Superlative Index Number Bias in Choosing Welfare Measures. Formulae." Review of Economics and Department of Agricultural Economics Statistics Vol. 63, No.3 (August 1981): Working Paper No. 92-14. Davis: Uni­ 430--5. versity of California, Davis, 1992. Alston, J.M. "An Analysis of Growth of Alston, J.M. and D.M. Larson. "Hicksian U.S. Farmland Prices, 1963-82." Amer- vs Marshallian Welfare Measures: Why 515 516 References Do We Do What We Do?" American Vol. 65, No.2 (May 1983): 353--6. Journal ofA gricultural Economics Vol. Alston, J.M. and V.H. Smith. "Some Eco­ 75, No.3 (August 1993): 764--9. nomic Implications of Minimum Pric­ Alston, J.M. and W.J. Martin. "Reversal of ing: The Case of Wine Grapes in Fortune: Immiserizing Technological Australia--Comment." Review of Mar­ Change in Agriculture." Invited paper keting and Agricultural Economics Vol. presented at the International Agricul­ 51, No.2 (August 1983): 179-85. tural Trade Research Consortium An­ Alston, J.M. and M.K. Wohlgenant. "Mea­ nual Meeting, St. Petersburg, Florida, suring Research Benefits Using Linear December 1992. Elasticity Equilibrium Displacement Alston, J.M. and W.J. Martin. "A Comment Models." In J.D. Mullen and J.M. Al~ on Border Price Changes and Domestic ston. The Returns to the Australian Wool Welfare in the Presence of Distortions." Industry from Investment in R&D (ap­ Oxford Economic Papers (forthcoming pendix 2). Rural & Resource Economics 1995). Report No. 10. Sydney: New South Alston, J.M. and J.D. Mullen. "Economic Wales Department of Agriculture & Effects of Research into Traded Goods: Fisheries, Division of Rural and The Case of Australian Wool." Journal Resource Economics, 1990. ofA gricultural Economics Vol. 43, No. Alston, J.M., J.R. Anderson and P.G. 2 (May 1992): 268-78. Pardey. "Perceived Productivity, Fore­ Alston, J.M. and P.G. Pardey. "Technical gone Future Farm Fruitfulness, and Change, Constraints and Incentives: The Rural Research Resource Rationaliza­ Damaging Consequences of Getting tion." Invited plenary paper for theme Policies Wrong." Invited paper pre­ VII National and International Re­ sented at the Conference on Agricultural search and Technology Transfer Technology: Current Policy Issues for Twelfth International Conference of Ag­ the International Community and The ricultural Economists, Harare, August World Bank, organized by the Agricul­ 1994. Mimeo. ture and Rural Development Department Alston, J.M., J.A. Chalfant and P.G. Pardey. of The World Bank, October 1991. Structural Adjustment in OEeD Agricul­ Alston, J.M. and P.G. Pardey. "Market Dis­ ture: Government Policies and Techni­ tortions and Technological Progress in cal Change. Center for International Agriculture." Technological Forecast­ Food and Agricultural Policy Working ing and Social Change Vol. 43, No. 3/4 Paper WP93-3. St. Paul: University of (May/June 1993): 301-19. Minnesota, June 1993. Alston, J.M. and P.G. Pardey. "Distortions Alston, J.M., G.W. Edwards and J.W. Free­ in Prices and Agricultural Research In­ bairn. The Effects of Protection Policies vestments." In J.R. Anderson, ed., Agri­ on the Benefitsfrom Research. Econom­ cultural Technology: Current Policy ics Branch Working Paper. Melbourne: Issues for the International Community. Department of Agriculture, Victoria, Wallingford: CAB International, 1994. 1986. Alston, J.M. and G.M. Scobie. "Distribu­ Alston, J.M., G.W. Edwards and J.W. Free­ tion of Research Gains in Multistage bairn. "Market Distortions and Benefits Production Systems: Comment." Amer­ from Research." American Journal of ican Journal ofA gricultural Economics Agricultural Economics Vol. 70, No.2 References 517 (May 1988): 281--8. riculture and Development, 1992. Alston, J.M., P.G. Pardey and H.O. Carter, Anderson, 1.R. and 1.L. Dillon. Risk Analy­ eds. Valuing UC Agricultural Research sis in Dryiand Farming Systems. Farm and Extension. Agricultural Issues Cen­ Systems Management Series No.2. ter Publication No. VR-1. Davis: Uni­ Rome: Food and Agriculture Organiza­ versity of California, March 1994. tion, 1992. Alston, 1.M., J.W. Freebairn and J.J. Anderson, 1.R. and 1.B. Hardaker. "Effi­ Quilkey. "A Model of Supply in the cacy and Efficiency in Agricultural Re­ Australian Orange Growing Industry." search: A Systems View." Agricultural Australian Journal ofA gricultural Eco­ Systems Vol. 40, Nos. 1-3 (1992): 105- nomics Vol. 24, No. 3 (December 23. 1980): 248-67. Anderson, J.R. and P.B.R. Hazell, eds. Vari­ American Agricultural Economics Associ­ ability in Grain Yields: Implications for ation (AAEA). Measurement of u.s. Ag­ Agricultural Research and Policy in De­ ricultural Productivity: A Review of veloping Countries. Baltimore: Johns Current Statistics and Proposals for Hopkins University Press, 1989. Change. ESCS Technical Bulletin No. Anderson, J.R., 1.L. Dillon and 1.B. Harda­ 1614. Washington, OC: U.S. Depart­ ker. Agricultural Decision Analysis. ment of Agriculture, 1980. Ames: Iowa State University Press, Anania, G. and A.F. McCalla. "Assessing 1977. the Impact of Agricultural Technology Anderson, 1.R., P.G. Pardey and 1. Rose­ Improvements in Developing Countries boom. "Sustaining Growth in Agricul­ in the Presence of Policy Distortions." ture: A Quantitative Review of Invited paper presented at the Interna­ Agricultural Research Investments." tional Agricultural Trade Research Con­ Agricultural &onomics Vol. 10, No.2 sortium Annual Meeting, St Petersburg, (April 1994): 107-23. Florida, December 1991. Antle, 1.M. "Infrastructure and Aggregate Anderson, J.E. and J.P. Neary. "Trade Re­ Agricultural Productivity: International form with Quotas, Partial Rent Reten­ Evidence." Economic Development and tion and Tariffs." &onometrica Vol. Cultural Change Vol. 31, No.3 (April 60, No. 1 (January 1992): 57-76. 1983): 609-19. Anderson, 1.R. "Agricultural Research in a Antle, 1.M. "The Structure of U.S. Agricul­ Variable and Unpredictable World." In tural Technology, 1910-78." American P.G. Pardey, J. Roseboom and 1.R. Journal ofA gricultural Economics Vol. Anderson, eds., Agricultural Research 66, No.4 (November 1984): 414-21. Policy: International Quantitative Per­ Antle, J.M. and S.M. Capalbo. "An Intro­ spectives. Cambridge: Cambridge Uni­ duction to Recent Developments in Pro­ versity Press, 1991. duction Theory and Productivity Mea­ Anderson, 1.R. "Research Priority Setting surement." In S.M. Capalbo and 1.M. in Agriculture: Problems in Ex-ante Antle, eds., Agricultural Productivity: Analysis." In D.R. Lee, S. Kearl and N. Measurement and Explanation. Wash­ Uphoff, eds., Assessing the Impact of ington, OC: Resources for the Future, International Agricultural Research for 1988. Sustainable Development. Ithaca: Cor­ Antle, 1.M. and T. McGuckin. 'Technolog­ nell International Institute for Food, Ag- ical Innovation, Agricultural Productiv- 518 References ity, and Environmental Quality." In Ayer, H.W. and G.E. Schuh. "Social Rates G.A. Carlson, D. Zilberman and J.A. of Return and Other Aspects of Agricul­ Miranowski, eds., Agricultural and En­ tural Research: The Case of Cotton Re­ vironmental Resource Economics. New search in Sao Paulo, Brazil." American York: Oxford University Press, 1993. Journal ofA gricultural Economics Vol. Antony, G. "User's Guide to Program CBR: 54, No.4., Pt. I (November 1972): 557- Calculation of the Costs and Benefits of 69. PNG Export-Crop Research: Version Ayer, H.W. and G.E. Schuh. "Social Rates 3.2." Department of Agricultural Eco­ of Return and Other Aspects of Agricul­ nomics and Business Management, Uni­ tural Research: The Case of Cotton Re­ versity of New England, Armidale, search in Sao Paulo, Brazil: Reply." Australia, June 1989. Mimeo. American Journal ofA gricultural Eco­ Apland, J., R.N. Barnes and F. Justus. 'The nomics Vol. 56, No. 4 (November Farm Lease: An Analysis of Owner­ 1974): 842-4. Tenant and Landlord Preferences under Baker, N.R. and J .R. Freeland. "Recent Ad­ Risk." American Journal of Agricul­ vances in R&D Benefit Measurement tural Economics Vol. 66, No.3 (August and Project Selection Methods." Man­ 1984): 376-84. agement Science Vol. 21, No. 10 (June Archibald, S.O. "Incorporating Externali­ 1975): 1164-75. ties into Agricultural Productivity Anal­ Baker, N.R. and W.H. Pound. "R&D Proj­ ysis." In S.M. Capalbo and J.M. Antle, ect Selection: Where We Stand." IEEE eds., Agricultural Productivity: Mea­ Transactions on Engineering Manage­ surement and Explanation Washington, ment Vol. EM-ll (December 1964): DC: Resources for the Future, 1988. 124-34. Arndt, T.M., D.G. Dalrymple and V.W. Baker, N.R., W.E. Souder, C.R. Shumway, Ruttan, eds. Resource Allocation and P.M. Maher and A.H. Rubenstein. "A Productivity in National and Interna­ Budget Allocation Model for Large Hi­ tionalAgricultural Research. Minneap­ erarchical R&D Organizations." Man­ olis: University of Minnesota Press, agement Science Vol. 23, No. 1 1977. (September 1976): 59-70. Arrow, K.J. 'The Economic Implications Ball, V.E. "Output, Input, and Productivity of Learning by Doing." Review ofEco­ Measurement in U.S. Agriculture." nomic Studies Vol. 29, No. 80 (June American Journal of Agricultural Eco­ 1962): 155-73. nomics Vol. 67, No.3 (August 1985): Arrow, K.J. Social Choice and Individual 475-86. Values. New York: John Wiley, 1963. Ball, V.E. "Modeling Supply Response in Arrow, K.J. and R.C. Lind. "Uncertainty a Multiproduct Framework." American and the Evaluation of Public Investment Journal ofA gricultural Economics Vol. Decisions." American Economic Review 70, No.4 (November 1988): 813-25. Vol. 60, No.3 (June 1970): 364-78. Ball, V.E. andR. Chambers. "An Economic Arrow, K.J. and R.C. Lind. "Uncertainty Analysis of Technology in the Meat and the Evaluation of Public Investment Products Industry." American Journal of Decisions: Reply." American Economic Agricultural Economics Vol. 64, No.4 Review Vol. 62, No.1 (March 1972): (November 1982): 699-7W. 171-2. Ballard, C.L., J.8. Shoven and J. Whalley. References 519 "General Equilibrium Computations of Bengsten, D.N. "Price Indexes for Deflat­ the Marginal Welfare Costs of Taxes in ing Public Forestry Research Expendi­ the United States." American Economic tures." Forest Science Vol. 35, No.3 Review Vol. 75, No. 1 (March 1985): (September 1989): 756-74. 1989b. 128-38. Berndt, E.R. Practice of Econometrics: Bantilan, M.C.S. and J.M. Lantican, eds. Classic and Contemporary. New York: Setting Research Priorities for Philip­ Addison Wesley, 1991. pine Agriculture. Quezon City, Philip­ Berndt, E.R. and M. Khaled. "Parametric pines: Bureau of Agricultural Research, Productivity Measurement and Choice 1994. among Flexible Functional Forms." Barnett, W.A. and Y.w. Lee. "The Global Journal of Political Economy Vol. 87, Properties of the Minfex Larent, Gener­ No.6 (December 1979): 1220-45. alized Leontief, and Translog Flexible Bernhart, K. and R.K. Perrin. "On Predict­ Functional Forms." Econometrica Vol. ing Supply Shifts from New Technol­ 53, No.6 (November 1985): 1421-37. ogy: Diethylstilbestrol." Department of Barten, P. "Maximum Likelihood Estima­ Economics and Business, North Caro­ tion of a Complete System of Demand lina State University, 1989. Mimeo. Equations." European Economic Re­ Bernstein, J.l. "The Structure of Canadian view Vol. 1, No.1 (Fall 1969): 7-73. Inter-Industry R&D Spillovers, and the Baumol, W.J. "Williamson's The Eco­ Rates of Return to R&D." Journal of nomic Institutions of Capitalism." Rand Industrial Economics Vol. 37, No.3 Journal of Economics Vol. 17, No. 2 (March 1989): 315-28. (Summer 1986): 279-86. Bernstein, J.I. and M.I. Nadiri. "Interindus­ Beach, D.E. and J.M. Alston. "Market Dis­ try R&D Spillovers, Rates of Return, tortions and the Benefits from Research and Production in High-Tech Indus­ into New Uses for Agricultural Com­ tries." American Economic Review Vol. modities: Ethanol from Corn." Paper 78, No.2 (May 1988): 429-34. presented at CPI session on Policy Eco­ Bernstein, J.I. and M.I. Nadiri. "Research nomics of New Industrial Uses of Agri­ and Development and Intra-Industry cultural Materials, Western Economics Spillovers: An Empirical Application of Association International Annual Meet­ Dynamic Duality." Review ofE conomic ing, Lake Tahoe, June 1993. Studies Vol. 56, No.2 (April 1989): Beattie, B.R. and c.R. Taylor. The Eco­ 249-69. nomics of Production. New York: John Bhagwati, J. "Imrniserizing Growth: A Wiley and Sons, 1985. Geometrical Note." Review of Eco­ Belsley, D.A., E. Kuh and R.E. Welsch. nomic Studies Vol. 25, No. 68 (June Regression Diagnostics: Identifying In­ 1958): 201-5. fluential Data and Sources of Collin­ Bhagwati, J. "Distortions and Imrniserizing earity. New York: John Wiley and Growth: A Generalization." Review of Sons, 1980. Economic Studies Vol. 35, No. 104 (Oc­ Bengsten, D.N. "A Price Index for Deflat­ tober 1968): 481-5. ing State Agricultural Experiment Sta­ Bieri, J., A. de Janvry and A. Schmitz. tion Research Expenditures." Journal of "Agricultural Technology and the Dis­ Agricultural Economics Research Vol, tribution of Welfare Gains." American 41, No.4 (Fall 1989): 12-20. 1989a. Journal ofA gricultural Economics Vol. 520 References 54, No.5 (December 1972): 801-8. Birdsall, N. and A. Steer. "Act Now on Binswanger, H.P. "A Cost Function Ap­ Global Warming-But Don't Cook the proach to the Measurement of Elastici­ Books." Finance and Development Vol. ties of Factor Demand and Elasticities 30, No. I (March 1993): 6-8. of Substitution." American Journal of Blackorby, e. and RR Russell. "Will the Agricultural Economics Vol. 56, No.2 Real Elasticity of Substitution Please (May 1974): 377-86. I 974a. Stand Up? (A Comparison of the Binswanger, H.P. 'The Measurement of AllenlUzawa and Morishima Elastici­ Technical Change Biases with Many ties)." American Economic Review Vol. Factors of Production." American Eco­ 79, No.4 (September 1989): 882-8. nomic Review Vol. 64, No.6 (December Blackorby, e., e.A.K. Lovell and M.e. 1974):964-76. I 974b. Thursby. "Extended Hicks Neutral Binswanger, H.P. The Use of Duality be­ Technological Change." Economic tween Production, Profit, and Cost Journal Vol. 86, No. 344 (December Functions in Applied Econometric Re­ 1976): 845-52. search: A Didactic Note. Economics Bockstael, N.E. and I.E. Strand. 'The Ef­ Program Occasional Paper No. 10. Hy­ fect of Common Sources of Regression derabad, India: International Crops Re­ Error on Benefit Estimates." Land Eco­ search Institute for the Semi-Arid nomics Vol. 63, No. I (February 1987): Tropics, July 1975. 11-20. Binswanger, H.P. "Measuring the Impact of Boisvert, RN. The Trans/og Production Economic Factors on the Direction of Function: Its Properties, Its Severalln­ Technical Change." In T.M. Arndt, D.G. terpretations and Estimation Problems. Dalrymple and V.W. Ruttan, eds., Department of Agricultural Economics Resource Allocation and Productivity in Research Series No. 82-28. Ithaca, National and International Agricultural N.Y.: Cornell University, September Researck Minneapolis: University of 1982. Minnesota, 1977. Bosch, DJ. and L.A. Shabman. "Simulation Binswanger, H.P. "Income Distribution Modeling to Set Priorities for Research Effects of Technical Change: Some An­ on Oyster Production." American Jour­ alytical Issues." South East Asian Eco­ nal of Agricultural Economics Vol. 72, nomic Review Vol. I, No.3 (December No.2 (May 1990): 371-81. 1980): 179-218. Bottomley, lA. and RB. Contant. Methods Binswanger, H.P. and lG. Ryan. "Effi­ for Setting Priorities among Different ciency and Equity Issues in Ex-ante Al­ Lines of Agricultural Researck ISNAR location of Research Resources." Indian Staff Notes No. 88-38. The Hague: In­ Journal ofA gricultural Economics Vol. ternational Service for National Agricul­ 32, No.3 (July/September 1977): 217- tural Research, December 1988. 31. Bouchet, F.C. "An Analysis of the Sources Binswanger, H.P., M.C. Yang, A. Bowers of Growth in French Agriculture, 1960- and Y. Mundlak. "On the Determinants 1987." Unpublished Ph.D. Thesis, Vir­ of Cross-Country Aggregate Agricul­ ginia Polytechnic Institute and State tural Supply." Journal of Econometrics University, Blacksburg, 1987. Vol. 36, No. 1 (September/October Bouchet, F.e., D. Orden and G.W. Norton. 1987): 111-31. "Sources of Growth in French Agricul- References 521 ture." American Journal ofA gricultural Use of Ridge Regression with Produc­ Economics Vol. 71, No.2 (May 1989): tion Function Applications." American 280--93. Journal ofA gricultural Economics Vol. Boulding, K.E. Economics as a Science. 57, No.1 (February 1975): 21-32. New York: McGraw Hill, 1970. Browning, E.K. 'The Marginal Cost of Boussard, J.-M. "Risk and Uncertainty in Public Funds." Journal of Political Programming Models: A Review." In Economy Vol. 84, No.2 (April 1976): J.A. Roumasset, J.-M. Boussard and I. 283-98. Singh, eds., Risk. Uncertainty, andAgri­ Browning, E.K. 'The Marginal Cost of cultural Development. New York: Agri­ Raising Tax Revenue." In P. Cagan, ed., cultural Development Council, 1979. Essays in Contemporary Economic Boyce, J.K. and R.E. Evenson. Agricultural Problems, 1986; The Impact oft he Rea­ Research and Extension Programs. New gan Program Washington, DC: Ameri­ York: Agricultural Development Coun­ can Enterprise Institute, 1986. cil,1975. Browning, E.K. "On the Marginal Welfare Bredahl, M.E. and W.L. Peterson. 'The Pro­ Cost of Taxation." American Economic ductivity and Allocation of Research at Review Vol. 77, No.1 (March 1987): U.S. Agricultural Experiment Stations." 11-23. American Journal of Agricultural eco­ Bruno, M. "Domestic Resource Costs and nomics Vol. 58, No.4 (November 1976): Effective Protection: Clarification and 684-92. Synthesis." Journal of Political Econ­ Bredahl, M.E., W. Myers and KJ. Collins. omy Vol. 80, No. 1 (January/February 'The Elasticity of Foreign Demand for 1972): 16-33. U.S. Agricultural Products: The Import­ Bumiaux, J. and D. van der Mensbrugghe. ance of the Price Transmission Elastic­ Trade Policies in a Global Context; ity." American Journal of Agricultural Technical Specification of the Rura/J Economics Vol. 61, No. 1 (February Urban-North/South (RUNS) Applied 1979): 58-63. General Equilibrium Model. Technical Brennan, J.P. "Measuring the Contribution Paper No. 48. Paris: Organization for of New Varieties to Increasing Wheat Economic Cooperation and Develop­ Yields." Review ofM arketing and Agri­ ment, Development Centre, 1991. cultural Economics Vol. 52, No.3 (De­ Burt, O.R. and V.E. Worthington. "Wheat cember 1984): 175-95. Acreage Supply Response in the United Brennan, J.P. Impact of Wheat Varieties States." Western Journal of Agricul­ from CIMMIT on Australian Wheat Pro­ tural Economics Vol. 13, No.1 (July duction. Agricultural Economics Bul­ 1988): 100--11. letin No.5. Sydney: Department of Buse, R.c. "Total Elasticities-A Predic­ Agriculture New South Wales, Septem­ tive Device." Journal ofF arm Econom­ ber 1986. ics Vol. 15, No.4 (November 1958): Brinegar, G.K. "Discussion-A Frame­ 881-91. work for Establishing Research Priori­ Byerlee, D. Comparative Advantage and ties." Journal of Farm Economics Vol. Policy Incentives for Wheat Production 48, No.5 (December 1966): 1638-40. in Ecuador. CIMMYT Economics Pro­ Brown, W.G. and B.R. Beattie. "Improving gram WOIking Paper No. 01/85. Mex­ Estimates of Economic Parameters by ico: Centro Intemacional de Mejora- 522 References miento de Mafz y Trigo, 1985. cultural Economics Vol. 40, No.4 (De­ Byedee, D. and E.H. de Polanco. "Farmers' cember 1992): 547-60. Stepwise Adoption of Technological Carter, C.A. and W.H. Gardiner, eds. Elas­ Packages: Evidence from the Mexican ticities in International Agricultural Altiplano." American Journal of Agri­ Trade. Boulder, Colorado: Westview cultural Economics Vol. 68, No.3 (Au­ Press, 1988. gust 1986): 519-27. Cassells, I.M. ''The Nature of Statistical Byerlee, D. and S. Franzel. Institutionaliz­ Supply Curves." Journal of Fann Eco­ ing the Role of the Economist in Na­ nomics Vol. 15, No.2 (April 1933): tional Agricultural Research Institutes. 378-87. CIMMYT Economics Program Working de Castro, J .P.R. and G.E. Schuh. "An Em­ Paper 93-02. Mexico: Centro Inter­ pirical Test of an Economic Model for nadonal de Mejoramiento de Malz y Establishing Research Priorities: A Bra­ Trigo, 1993. zil Case Study." In T.M. Arndt, D.G. Byerlee, D. and M. Morris. "Research for Dalrymple and V.W. Ruttan, eds., Marginal Environments. Are We Un­ Resource Allocation and Productivity in derinvestedT Food Policy Vol. 18, No. National and International Agricultural 5 (October 1993): 381-93. Research. Minneapolis: University of Byedee, D., M. Igbal and K.S. Fisher. Minnesota Press, 1977. "Quantifying and Valuing the Joint Pro­ Caves, D. and L. Christensen. "Global Prop­ duction of Grain and Fodder from Maize erties of Flexible Functional Forms." Fields: Evidence from Northern Paki­ American Economic Review Vol. 70, stan." ExperimentalAgriculture Vol. 25, No.3 (June 1980): 422-32. No.4 (October 1989): 435-45. Cessay, S., E. Gilbert, B. Mills, J. Rowe Candler, W. and M. Boehlje. "Use of Lin­ and G.W. Norton. Analysis of Agricul­ ear Programming in Capital Budgeting tural Research Priorities in the Gam­ with Multiple Goals." American Jour­ bia. ACIARIISNAR Project Paper No. nal ofA gricultural Economics Vol. 53, 16. Canberra and The Hague: Austra­ No.2 (May 1971): 325-30. lian Centre for International Agricul­ Capalbo, S.M. and I.M. Antle, eds. Agri­ tural Research and International Service cultural Productivity: Measurement for National Agricultural Research, No­ and Explanation. Washington, DC: Re­ vember 1989. sources for the Future, 1988. Cetron, MJ., 1. Martino and L.A. Roepeke. Capalbo, S., M. Denny, A. Hoque and C.E. ''The Selection of R&D Program Con­ Overton. Methodologies for Compari­ tent-Survey of Quantitative Methods." sons ofA gricultural Output, Input, and IEEE Transactions on Engineering Man­ Productivity: A Review and Synthesis. agement Vol. EM-14 (March 1967): 1- Agriculture and Trade Analysis Divi­ 13. sion Staff Report No. AGES 9122. CGIARffAC. TAC Review ofCGIAR Priori­ Washington, DC: U.S. Department of ties and Future Strategies. Rome: T AC Agriculture, April 1991. Secretariat, Food and Agriculture Orga­ Carew, R., P. Chen and V. Stevens. "Eval­ nization, December 1986. uating Publicly Funded Research in Ca­ Chalfant, J.A. "Choosing Among Flexible nadian Agriculture: A Profit Function Functional Forms: An Application of Approach." Canadian Journal of Agri- the Generalized Box-Cox and the Fou- References 523 rier Flexible Forms to U.S. Agricul­ New York: John Wiley and Sons,1961. ture." Unpublished Ph.D. Thesis. Ra­ Chavas, J.-P. and T.L. Cox. "A Nonpara­ leigh: North Carolina State University, metric Analysis of Agricultural Technol­ 1983. ogy." American Journal ofA gricultural Chalfant, J.A. "Comparisons of Alternative Economics Vol. 70, No.2 (May 1988): Functional Forms with Application to 303-10. Agricultural Input Data." American Chavas, J.-P. and T.L. Cox. "A Nonpara­ Journal ofA gricultural Economics Vol. metric Analysis of the Influence of Re­ 66, No.2 (May 1984): 216--20. search on Agricultural Productivity." Chalfant, J .A. and N .E. Wallace. "Bayesian American Journal of Agricultural Eco­ Analysis and Regularity Conditions on nomics Vol. 74, No.3 (August 1992): Flexible Functional Forms: Application 583-91. to the U.S. Motor Carrier Industry." In Chavas, J.-P. and S.R Johnson. "Supply W. Griffiths, H. Liietkepohl and M.E. Dynamics: The Case of U.S. Broilers Bock, eds., Readings in Econometric and Turkeys." American Journal ofA g­ Theory and Practice. Amsterdam: Else­ ricultural Economics Vol. 64, No.3 vier,1992. (August 1982): 558-64. Chalfant, J.A. and B. Zhang. "Variations Chipman, J.S. and J.C. Moore. ''The New on Invariance or Some Unpleasant Non­ Welfare Economics, 1939-1974." In­ parametric Arithmetic." Department of ternational Economic Review Vol. 19, Agricultural Economics, University of No.3 (October 1978): 581-84. California, Davis, 1994. Mimeo. Chipman, J.S. and J.C. Moore. "Compen­ Chambers, RG. Applied Production Anal­ sating Variation, Consumer's Surplus, ysis: A Dual Approach. Cambridge: and Welfare." American Economic Re­ Cambridge University Press, 1988. view Vol. 70, No.5 (December 1980): Chambers, RG. and R. Lopez. "Public In­ 933-49. vestment and Real-Price Supports." Christensen, L.R., D.W. Jorgenson and LJ. Journal of Public Economics Vol. 52, Lau. "Conjugate Duality and the Tran­ No. 1 (August 1993): 73-82. scendental Logarithmic Production Chang, C.C., B.A. McCarl, J. Mjelde and J. Function." Econometrica Vol. 39, No.4 Richardson. "Sectoral Implications of (July 1971): 255-56. Farm Program Modifications." Ameri­ Christensen, L.R., D.W. Jorgenson and LJ. can Journal ofA gricultural Economics Lau. "Transcendental Logarithmic Pro­ Vol. 74, No. I (February 1992): 38-49. duction Frontiers." Review of Econom­ Chaparro, F., G. Montes, R. Torres, A. ics and Statistics Vol. 55, No. 1 Balzaca and H. Jaramillo. "Research (February 1973): 28-45. Priorities and Resource Allocation in CIMMYT. The Adoption of Agricultural Agriculture: The Case of Colombia." In Technology: A Guide for Survey Design. D. Daniels and B. Nestel, eds., Resource Mexico: Economics Program, Centro Allocation to Agricultural Research. Internacional de Mejoramiento de Maiz Ottawa: International Development Re­ y Trigo, 1993. search Centre, 1981. Cline, P.L. "Sources of Productivity Chames, A. and W.W. Cooper. Manage­ Change in United States Agriculture." ment Models and Industrial Application Unpublished Ph.D. Thesis, Oklahoma of Linear Programming. Volume J. State University, Stillwater, 1975. 524 References Cline, W.R. The Economics of Global Welfare. Oxford: Oxford University Warming. Washington, DC: Institute Press, 1974. for International Economics, 1992. Comes, R. Duality and Modem Econom­ Coble, K.H., C.C. Chang, B.A. McCarl and ics. Cambridge: Cambridge University B.R. Eddelmann. "Assessing Economic Press, 1992. Implications of New Technology: The Cox, T.L. and J.-P. Chavas. "A Nonpara­ Case of Cornstarch-Based Biodegrada­ metric Analysis of Productivity: The ble Plastics." Review of Agricultural Case ofU .S. Agriculture." European Re­ Economics Vol. 14, No. 1 (January view ofA gricultural Economics Vol. 17, 1992): 3~3. No.4 (1990): 449-64. Cochrane, W.W. "Some Nonconformist Craig, B.J. and P.O. Pardey. Patterns of Thoughts on Welfare Economics and Agricultural Development in the United Commodity Stabilization Policy." States. Department of Agricultural and American Journal of Agricultural Eco­ Applied Economics Staff Paper P9O- nomics Vol. 62, No.3 (August 1980): 72. St. Paul: University of Minnesota, 508-11. December 1990. Cohon, J.L. "A Review and Evaluation of Craig, B.J., P.O. Pardey and K.W. Deinin­ Multiobjective Programming Tech­ ger. "Capital Services in U.S. Agricul­ niques." Water Resources Research Vol. ture." Department of Agricultural and II, No.2 (April 1975): 208-20. Applied Economics, University of Min­ Cohon, J.L. Multiobjective Progamming nesota, St. Paul, 1993. Mimeo. and Planning. New York: Academic Craig, B.J., P.O. Pardey and J. Roseboom. Press, 1978. International Agricultural Productivity Cohon, J.L. and D.H. Marks. "Multi-Objec­ Patterns. Center for International Food tive Screening Models and Water and Agricultural Policy Working Paper Resource Investment." Water Resources WP94-1. St Paul: University of Minne­ Research Vol. 9, No.4 (August 1973): sota, February 1994. 826-36. Crosson, P. and J.R. Anderson. Concerns Colman, D. "A Review of the Arts of Sup­ for Sustainability: Integration ofN atural ply Response Analysis." Review of Resource and Environmental Issues in Marketing and Agricultural Economics the Research Agendas ofN ARS. ISNAR Vol. 51, No.3 (December 1983): 201- Research Report No.4. The Hague: In­ 30. ternational Service for National Agricul­ Constantine, J.H., J.M. Alston and V.H. tural Research, October 1993. Smith. "Economic Impacts of the Cali­ Currie, J.M., J.A. Murphy and A. Schmitz. fornia One-Variety Cotton Law." Jour­ "The Concept of Economic Surplus and nal of Political Economy (forthcoming Its Use in Economic Analysis." eco­ 1995). nomic Journal Vol. 81, No. 324 (De­ Cooke, S.C. and W.B. Sundquist. 'The In­ cember 1971): 741-99. cidence of Benefits from U.S. Soybean Dalrymple, D.O. 'The Excess Burden of Productivity Oains in a Context of Taxation and Public Agricultural Re­ World Trade." American Journal ofA g­ search." In R.O. Echeverria, ed., Meth­ ricultural Economics Vol. 75, No.1 ods of Diagnosing Research System (February 1993): 169-80. Constraints and Assessing the Impact of Corden, W.M. Trade Policy and Economic Agricultural Research Volume II, As- References 525 sesssing the Impact of Agricultural Re­ 22. Canberra and The Hague: Austra­ search The Hague: International Ser­ lian Centre for International Agricul­ vice for National Agricultural Research, tural Research and International Service 1990. for National Agricultural Research, Oc­ David, C.C. and K. Otsuka, eds. Modem tober 1990. Rice Technology and Income Distribu­ Davis, 1.S., P. Oram and 1.G. Ryan. Assess­ tion in Asia. Boulder: Lynne Rienner ment of Agricultural Research Priori­ Publishers, 1994. ties: An International Perspective. Davidson, B.R. and B.R Martin. "The Re­ Canberra and Washington, DC: Austra­ lationship between Yields on Farms and lian Centre for International Agricul­ in Experiments." Australian Journal of tural Research and International Food Agricultural Economics Vol. 9, No.2 Policy Research Institute, 1987. (December 1965): 129-40. Deaton, A.S. and 1. Muellbauer. "An Al­ Davidson, B.R, B.R Martin and RG. most Ideal Demand System." American Mauldon. 'The Application of Experi­ Economic Review Vol. 70, No.3 (June mental Research to Farm Production." 1980): 312-26. 1980a. Journal of Farm Economics Vol. 49, Deaton, A.S. and 1. Muellbauer. Economics No.4 (November 1967): 900-07. and Consumer Behavior. Cambridge: Davis, H.T. The Theory of Econometrics. Cambridge University Press. 1980b. Bloomington: Principia Press, 1941. Denison, E.F. "Measurement of Labor Davis, 1.S. "Stability of the Research Pro­ Input: Some Questions of Defmition and duction Coefficient for U.S. Agricul­ the Adequacy of Data." In Output, Input ture." Unpublished Ph.D. Thesis, Uni­ and Productivity Measurement. Prince­ versity of Minnesota, St. Paul, 1979. ton: The National Bureau of Economic Davis, 1.S. '''The Relationship between the Research, Princeton University, 1961. Economic Surplus and Production Func­ Denison, E.F. "Some Major Issues in Pro­ tion Approaches for Estimating Ex Post ductivity Analysis: An Examination of Returns to Agricultural Research." Re­ Estimates by Jorgenson and Griliches." view ofM arketing andAgricultural Eco­ Survey ofC urrent Business Vol. 49, No. nomics Vol. 49, No.2 (August 1981): 5 (May 1969): 1-27. 95-105. Denny, M. and M. Fuss. "The Use of Dis­ Davis, J.S. Spillover Effects of Agricultural crete Variables in Superlative Index Research: Importance for Research Pol­ Number Comparisons." International icy and Incorporation in Research Eval­ Economic Review Vol. 24, No.2 (June uation Models. ACIARIISNAR Project 1983): 419-21. Paper No. 32. Canberra and The Hague: Dey, M. and G.W. Norton. Analysis ofA g­ Australian Centre for International Agri­ ricultural Research Priorities in cultural Research and International Ser­ Bangladesh. ISNAR Discussion Paper vice for National Agricultural Research, No. 93-07. The Hague: International February 1991. Service for National Agricultural Re­ Davis, J.S. and M.C.S. Bantilan. Import search, March 1993. Replacement and Export Enhancement Diewert, W.E. "An Application of the as Objectives of Agricultural Research Shephard Duality Theorem: Generalized Policy: Interpretation and Measure­ Leontief Production Function." Journal ment. ACIARIISNAR Project Paper No. of Political Economy Vol. 79, No.3 526 References (May-June 1971): 481-507. In D. Daniels and B. Nestel, eds., Diewert, W.E. "Functional Forms for Profit Resource Allocation to Agricultural Re­ and Transformation Functions." Jour­ search. Ottawa: International Develop­ nal of Economic Theory Vol. 6, No.3 ment Research Center, 1981. (June 1973): 284--316. Duncan, R.C. "Evaluating Returns to Re­ Diewert, W.E. "Applications of Duality search in Pasture Improvement." Austra­ Theory." In M. Intrilligator and D.A. lian Journal of Agricultural Economics Kendrick, eds., Frontiers of Quantita­ Vol. 16, No.3 (December 1972): 153- tive Economics. Volume II. Amsterdam: 68. North-Holland, 1974. Duncan, R.C. and C. Tisdell. "Research Diewert, W.E. "Exact and Superlative and Technical Progress: The Returns to Index Numbers." Journal ofE conomet­ the Producers." Economic Record Vol. rics Vol. 4, No.2 (May 1976): 115-45. 47,No.1l7(March 1971): 124--9. Diewert, W.E. "The Comparative Statistics Dyer, P.T., G.M. Scobie and S.R. Davis. of Industry Long-Run Equilibrium." The Payoff to Investment in a Recombi­ Canadian Journal of Economics Vol. nant DNA Research Facility at Ruakura: 14, No.1 (February 1981): 78-92. A Monte Carlo Simulation Study. Eco­ Diewert, W.E. The Early History of Price nomics Division Discussion Paper 1184. Index Research. Working Paper No. Hamilton, New Zealand: Ministry of 2713. Cambridge: National Bureau of Agriculture and Fisheries, February Economic Research, 1988a. 1984. Diewert, W.E. "Test Approaches to Inter­ Echeverria, R.G. "Assessing the Impact of national Comparisons." In W. Eichorn, Agricultural Research." In R.G. Eche­ ed., Measurement in Economics. Hei­ verria, ed., Methods for Diagnosing Re­ delberg: Physica Verlag, 1988b. search System Constraints and Dillon, lL. and lR. Anderson. The Analy­ Assessing the Impact ofA gricultural Re­ sis of Response in Crop and livestock search. Volume II, Assessing the Impact Production. Third edition. Oxford: Per­ of Agricultural Research. The Hague: gamon Press, 1990. International Service for National Agri­ Dixon, P., B. Parmenter, 1 Sutton and D. cultural Research, 1990. Vincent. ORANI: A Multisectoral Model Echeverria, R.o., G. Ferreira and M. of the Australian Economy. Amster­ Dabezies. Returns to Investment in the dam: North-Holland, 1982. Generation and Transfer of Rice Tech­ Dorfman, J.H. and D. Heien. 'The Effects nology in Uruguay. ISNAR Working of Uncertainty and Adjustment Costs on Paper No. 30. The Hague: International Investment in the Almond Industry." Service for National Agricultural Re­ Review ofE conomics and Statistics Vol. search, 1989. 71, No.2 (May 1989): 263-74. Edwards, G.W. and lW. Freebairn. Mea­ Drechsler, L. "Weighting ofIndex Numbers suring a Country's Gains From Re­ in Multilateral International Compari­ search: Theory and Application to Rural sons." Review of Income and Wealth Research in Australia. A Report to the Vol. 19, No.1 (March 1973): 17-35. Commonwealth Council for Rural Re­ Drilon, lD. and A.R. Librero. "Defining search and Extension. Canberra: Austra­ Research Priorities for Agriculture and lian Government Publishing Service, Natural Resources in the Philippines." 1981. References 527 Edwards, G.W. and J.W. Freebairn. "The Research, Invention, Extension and Pro­ Social Benefits from an Increase in Pro­ ductivity Change in U.S. Agriculture: An ductivity in a Part of an Industry." Re­ Historical Decomposition Analysis." In view of Marketing and Agricultural AA Araji, ed., Research and Extension Economics Vol. 50, No.2 (August Productivity in Agriculture. Moscow: 1982): 193-210. Department of Agricultural Economics, Edwards, G.W. and lW. Freebairn. 'The University ofidaho, 1980. Gains from Research into Tradeable Evenson, R.E. "Research, Extension, and Commodities." American Journal ofA g­ U.S Agricultural Productivity: A Statis­ ricultural Economics Vol. 66, No. 1 tical Decomposition Analysis." In S.M. (February 1984): 41-49. Capalbo and lM. Antle, eds., Agricul­ Ehui, S.K. and D.S.C. Spencer. "Measur­ tural Productivity: Measurement and ing the Sustainability and Economic Vi­ Explanotion. Washington, DC: Re­ ability of Tropical Farming Systems: A sources for the Future, 1988. Model for sub-Saharan Africa." Agri­ Evenson, R.E. "Spillover Benefits of Agri­ cultural &onomics Vol. 19, No.4 (De­ cultural Research: Evidence from U.S. cember 1993): 279-96. Experience." American Journal ofA gri­ Eichorn, W. "Fisher's Tests Revisited." cultural Economics Vol. 71, No.2 (May Econometrica Vol. 44, No. 2 (March 1989): 447-52. 1976): 247-56. Evenson, R.E. "Priority Setting: Ex Ante Eichorn, W. and J. Voeller. Theory of the Project Evaluation and Ex Post Evalua­ Price Index. Berlin: Springer Verlag, tion: A Comparative Perspective." Paper 1976. prepared for the Rice Research Espinosa, P., G.W. Norton and H.D. Gross. Prioritization Workshop, International Metodologia para Determinar Priori­ Rice Research Institute, Los Baiios, Phil­ dades de InvestigaciOn Agropecuaria. ippines, 13-15 August 1991. Serie Technica, Doc. No. 1. Quito, Ecu­ Evenson, R.E. "Notes on the Measurement ador: Instituto Nacional de Investigaci6n of the Economic Consequences of Agri­ Agropecharia, 1988. cultural Research Investment." In D.R. Evenson, R.E. 'The Contribution of Agri­ Lee, S. Kearl and N. Uphoff, eds., As­ cultural Research to Production." Jour­ sessing the Impact ofI nternational Agri­ nal of Farm Economics Vol. 49, No.5 cultural Research for Sustainable (December 1967): 1415-25. Development. Ithaca: Cornell Interna­ Evenson, R.E. 'The Contribution of Agri­ tional Institute for Food, Agriculture and cultural Research and Extension to Agri­ Development, 1992. cultural Production." Unpublished Ph.D. Evenson, R.E. and H.P. Binswanger. Thesis, University of Chigaco, Chicago, 'Technology Transfer and Research 1968. Resource Allocation." In H.P. Binswan­ Evenson, R.E. A Century of Productivity ger, V.W. Ruttan et aI., Inducedlnnova­ Change in U.S. Agriculture: An Analy­ tion: Technology, Institutions, and sis of the Role of Invention, Research, Development. Baltimore: Johns Hop­ and Extension. Economic Growth Cen­ kins University Press, 1978. ter Discussion Paper No. 276. Yale: Evenson, R.E. and Y. Kislev. Agricultural Yale University, August 1978. Research and Productivity. New Haven: Evenson, R.E. "A Century of Agricultural Yale University Press, 1975. 528 References Evenson, RE. and J. Putnam. "Intellectual nal of Agricultural Economics Vol. 73, Property Management." In GJ. Persley, No.3 (August 1991): 557....fJ7. ed., Agricultural Biotechnology: Op­ Finkelshtain, I. and lA. Chalfant. "Portfo­ portunities for International Develop­ lio Choices in the Presence of Other ment. Wallingford: CAB International, Risks." Management Science Vol. 39, 1990. No.8 (August 1993): 925-36. Hire, R, S. Grosskopf, M. Noms and Z. Fishel, W.L. "Uncertainty in Public Re­ Zhang. "Productivity Growth, Technical search Administration and Scientists' Progress, and Efficiency Change in In­ Subjective Probability &timates about dustrialized Countries." American Eco­ Changing the State of Knowledge." Un­ nomic Review Vol. 84, No.1 (March published Ph.D. Thesis, North Carolina 1994): 66-S3. State University, Raleigh, 1970. Fawson, C. and C.R. Shumway. "A Non­ Fishel, W.L., ed. Resource Allocation in parametric Investigation of Agricultural Agricultural Research. Minneapolis: Production Behavior for U.S. Subre­ University of Minnesota Press, 1971. gions." American Journal of Agricul­ Fisher, B. "Rational Expectations in Agri­ tural Economics Vol. 70, No.2 (May cultural Research and Policy Analysis." 1988): 311-17. American Journal of Agricultural Eco­ Feder, G. andD.L. Umali. ''The Adoption of nomics Vol. 64, No.2 (May 1982): Agricultural Innovations: A Review." 260-65. Technological Forecasting and Social Fisher, I. The Making of Index Numbers. Change Vol. 43, Nos. 3/4 (May/June Boston: Houghton Mifflin, 1922. 1993): 215-39. FIacco, P.R. and D.M. Larson. "Nonpara­ Feder, G., RE. Just and D. Zilberman. metric Measures of Scale and Technical "Adoption of Agricultural Innovations Change for Competitive Firms under in Developing Countries: A Survey." Uncertainty." American Journal ofA g­ Economic Develoment and Cultural ricultural Economics Vol. 74, No. 1 Change Vol. 33, No.2 (January 1985): (February 1992): 173-76. 255-98. Hores-Moya, P., RE. Evenson and Y. Ferguson, C.E. The Neoclassical Theory of Hayami. "Social Returns to Rice Re­ Production and Distribution. Cam­ search in the Philippines: Domestic Ben­ bridge: Cambridge University Press, efits and Foreign Spillover." Economic 1975. Development and Cultural Change Vol. Ferreira, G., G.W. Norton and M. Dabezies. 26, No.3 (April 1978): 591....fJ07. Definicion de Prioriclades de Investiga­ Food and Agriculture Organization (FAD). cion Agropecuaria. Montevideo, Uru­ Agricultural Protection: Domestic Pol­ guay: Centro de Investigaciones Agri­ icy in International Trade. Rome: Food colas Alberto Boerger, 1987. and Agricultural Organization, 1973. Findlay, C.C. and R.L. Jones. ''The Mar­ Fox, G. "Is the United States Really Un­ ginal Cost of Australian Income Taxa­ derinvesting in Agricultural Research?" tion." Economic Record Vol. 58, No. American Journal of Agricultural Eco­ 162 (September 1982): 25~6. nomics Vol. 67, No. 4 (November 1985): Finkelshtain, I. and J.A. Chalfant. "Mar­ 806-12. keted Surplus under Risk: Do Peasants Fox, G. "Models of Resource Allocation in Agree with Sandmo?" American Jour- Public Agricultural Research: A Sur- References 529 vey." Journal ofA gricultural Econom­ puting All Direct and Cross Demand ics Vol. 38, No.3 (September 1987): Elasticities in a Model with Many Sec­ 449-62. tors." Econometrica Vol. 27, No. 2 Fox, G., G. Brinkman, and N. Brown-An­ (April 1959): 177-96. dison. "An Economic Analysis of the Fulginiti, L.E. and R.K. Perrin. Prices and Returns to the Animal Productivity Re­ Productivity in Agriculture. Depart­ search Program of Agriculture Canada ment of Agricultural Economics Journal from 1968 to 1984". Intercambia Lim­ Paper No. J-I4462. Ames: Iowa State ited, Guelph, Ontario, March 1987. Un­ University, 1992. published report. FUlginiti, L.E. and R.K. Perrin. ''The Theory Fox, G., B. Roberts and G.L. Brinkman. and Measurement of Production Re­ ''The Returns to Canadian Federal Dairy sponse under Quotas." Review of Eco­ Cattle Research: 1968-83." Department nomics and Statistics Vol. 75, No. 1 of Agricultural Economics and Business (February 1993): 97-106. Working Paper No. 98-20. Guelph, On­ Fullerton, D. "Reconciling Recent Esti­ tario: University of Guelph, October mates of the Marginal Welfare Cost of 1989. Taxation." American Economic Review Freebaim, J.W. "Evaluating the Level and Vol. 81, No.1 (March 1991): 302-8. Distribution of Benefits from Dairy In­ Fuss, M., D. McFadden and Y. Mundlak. dustry Research." Australian Journal of "A Survey of Functional Forms in the Agricultural Economics Vol. 36, No.2 Economic Analysis of Production." In (August 1992): 141-65. M. Fuss and D. McFadden, eds., Pro­ Freebaim, J.W., lS. Davis and G.W. Ed­ duction Economics: A Dual Approach wards. "Distribution of Research Gains to Theory and Applications. Amster­ in Multistage Production Systems." dam: North-Holland, 1978. American Journal ofA gricultural Eco­ Gallant, A.R. "Unbiased Determination of nomics Vol. 64, No.1 (February 1982): Production Technologies" Journal of 39--46. Econometrics Vol. 20, No.2 (Novem­ Freebaim, J.W., lS. Davis and G.W. Ed­ ber 1982): 285-323. wards. "Distribution of Research Gains Gardner, B.L. ''The Farm-Retail Price in Multistage Production Systems: Re­ Spread in a Competitive Food Indus­ ply." American Journal of Agricultural try." American Journal of Agricultural Economics Vol. 65, No.2 (May 1983): Economics Vol. 57, No. 3 (August 357-59. 1975): 399-409. French, B.C. and lL. Matthews. "A Supply Gardner, B.L. "Futures Prices in Supply Response Model for Perennial Crops." Analysis." American Journal of Agri­ American Journal ofA gricultural Eco­ cultural Economics Vol. 58, No.1 (Feb­ nomics Vol. 53, No.3 (August 1974): ruary 1976): 81-9. 478-90. Gardner, B.L. "Efficient Redistribution Freund, R.I. ''The Introduction of Risk into through Commodity Markets." Ameri­ a Programming Model." Econometrica can Journal ofA gricultural Economics Vol. 24, No.3 (July 1956): 253-63. Vol. 65, No.2 (May 1983): 225-34. Friedman, M. Price Theory. Chicago: Ald­ Gardner, B.L. The Economics of Agricul­ ine Publishing Company, 1976. tural Policies. New York: MacMillan, Frisch, R. "A Complete Scheme for Com- 1987. 530 References Gardner, B.L. Price Supports and Optimal Graham-Tomasi, T. "Sustainability: Con­ Agricultural Research Spending. De­ cepts and Implications for Agricultural partment of Agricultural and Resource Research Policy." In P.G. Pardey, l Economics Working Paper No. 88-1. Roseboom and lR. Anderson, eds., College Park: University of Maryland, Agricultural Research Policy: Interna­ January (revised June) 1988. tional Quantitative Perspectives. Cam­ Gardner, B.L. "How the Data We Make bridge: Cambridge University Press, Can Unmake Us: Annals of Factology." 1991. American Journal of Agricultural Eco­ Greig, I.D. "Agricultural Research Man­ nomics Vol. 74, No. 5 (December agement and the Ex Ante Evaluation of 1992): 1066-75. Research Proposals: A Review." Re­ Geigel, land W.B. Sundquist. A Review view of Marketing and Agricultural and Evaluation of Weather-Crop Yield Economics Vol. 49, No.2 (August Models. Department of Agricultural and 1981): 73-94. Applied Economics Staff Paper P84-5. Griliches, Z. "Specification Bias in Esti­ St. Paul: University of Minnesota, Feb­ mates of Production Functions." Jour­ ruary 1984. nal of Farm &onomics Vol. 39, No.1 Gijsbers, G. Methods and Procedures for (February 1957): 8-20. 1957a. the Development of INFORM. The Griliches, Z. "Hybrid Corn: An Explora­ Hague: International Service for Na­ tion in the Economics of Technological tional Agricultural Research, March Change." Econometrica Vol. 25, No.4 1991. (October 1957): 501-22. 1957b. Godyn, D.L., lP. Brennan and B.G. John­ Griliches, Z. "Research Costs and Social ston. 'The Gains from Research into Returns: Hybrid Corn and Related Inno­ Tradeable Commodities: A Comment." vations." Journal of Political Economy Department of Agriculture, New South Vol. 66, No.5 (October 1958): 419-31. Wales, Sydney, February 1987. Mimeo. Griliches, Z. "Measuring Inputs in Agricul­ Goodwin, B.K. and D.A. Sumner. "Market ture: A Critical Survey." Journal of Supply Parameters under Mandatory Farm &onomics Vol. 42, No.5 (De­ Production Quotas." Department ofEco­ cember 1960): 1411-27. nomics and Business, North Carolina Griliches, Z. "Estimates of the Aggregate State University, October 1990. Mimeo. Agricultural Production Function from de Gorter, H. andG.W. Norton. "A Critical Cross-Sectional Data." Journal ofF arm Appraisal of Analyzing the Gains from &onomics. Vol. 45, No.2 (May 1963): Research with Market Distortions." De­ 419-28. 1963a. partment of Agricultural Economics, Griliches, Z. 'The Sources of Measured Cornell University, Ithaca, April 1990. Productivity Growth: United States Ag­ Mimeo. riculture, 194~." Journal of Politi­ de Gorter, H., DJ. Nielson and G.C. cal &onomy Vol. 71, No.4 (August Rausser. "Productive and Predatory 1963):331-46. 1963b. Public Policies: Research Expenditures Griliches, Z. "Capital Stock in Investment and Producer Subsidies in Agriculture." Functions: Some Problems of Concept American Journal ofA gricultural &0- and Measurement." In C. Christ, et al., nomics Vol. 74, No. 1 (February 1992): eds., Measurement in &onomics: Stud­ 27-37. ies in Mathematical Economics and References 531 Econometrics in Memory of Yehuda Gryseels, G., C.T. de Wit, A McCalla, J. Grunfeld. Stanford: Stanford University Monyo, A Kassam, E. Craswell and M. Press, 1963c. Collinson. "Setting Agricultural Re­ Griliches, Z. "Research Expenditures, Edu­ search Priorities for the CGIAR." Agri­ cation and the Aggregate Agricultural cultural Systems Vol. 40, No. I (1992): Production Function." American Eco­ 59-103. nomic Review Vol. 54, No.6 (December Hanoch, G. and M. Rothschild. ''Testing 1964): 961-74. the Assumptions of Production Theory: Griliches, Z. "Distributed Lags: A Survey." A Nonparametric Approach." Journal Econometrica Vol. 35, No.1 (January of Political Economy Vol. 80, No.2 1967): 16-47. (March-April 1972): 256-75. Griliches, Z. "Issues in Assessing the Con­ Haque, E.A.K., G. Fox and G.L. Brinkman. tribution of Research and Development "Product Market Distortions and the to Productivity Growth." Bell Journal Returns to Federal Laying-Hen Re­ of Economics Vol. 10, No. 1 (Spring search in Canada." Canadian Journal of 1979): 92-116. Agricultural Economics Vol. 37, No. I Griliches, Z., ed. R&D, Patents, and Pro­ (March 1989): 29-46. ductivity. Chicago: University of Chi­ Harberger, AC. ''Three Basic Postulates cago Press, 1984. for Applied Welfare Economics: An In­ Griliches, Z. "Productivity, R&D, and terpretive Essay." Journal of Economic Basic Research at the Firm Level in the Literature Vol. 9, No.3 (September 1970s." American Economic Review 1971): 785-97. Vol. 76, No.1 (March 1986): 141-54. Harberger, AC. Project Appraisal: Col­ 1986a. lected Papers. Chicago: University of Griliches, Z. "Economic Data Issues."In Z. Chicago Press, 1976. Griliches and M.D. Intriligator, eds., Harberger, A.C. "On the Use of Dis­ Handbook of Econometrics. Volume 3. tributional Weights in Social Cost-Ben­ Amsterdam: North-Holland, 1986. efit Analysis." Journal of Political 1986b. Economy Vol. 86, No.2, Pt. 2 (April Griliches, Z. "Patent Statistics as Economic 1978): S87-S 119. Indicators." Journal ofE conomic Liter­ Hart, O. "On the Optimality of Equilibrium ature Vol. 28, No.4 (December 1990): When the Market Structure Is Incom­ 1661-707. plete." Journal of Economic Theory Griliches, Z. "The Search for R&D Spillo­ Vol. 11, No.3 (December 1975): 418- vers." Scandinavian Journal of Eco­ 43. nomics Vol. 94 (Supplement 1992): Hatanaka, M. and T.D. Wallace. "Multicol­ 29-47. linearity and the Estimation of Low­ Griliches, Z. "Productivity, R&D, and the Order Moments in Stable Lag Func­ Data Constraint." American Economic tions." In J. Kmenta and J. Ramsay, eds., Review Vol. 84, No. I (March 1994): Evaluation ofE conometric Models. San 1-23. Diego: Academic Press, 1980. Gruen, F.H. "Agriculture and Technical Hausman, J. A. "Exact Consumer's Surplus Change." Journal of Farm Economics and Dead-Weight Loss." American Vol. 43, No.4 (November 1961): 838- Economic Review Vol. 71, No.4 (Sep­ 58. tember 1981): 662-76. 532 References Havlicek Jr., lG. and G.W. Norton. A Re­ em Rice Varieties in Asia. Los Banos: view of Research Project Evaluation International Rice Research Institute, and Selection Methods: A Report to 1983. BARD. Department of Agricultural Eco­ Herendeen, lB. "A Framework for Estab­ nomics Staff Paper SP-I3-81. Blacks­ lishing Research Priorities: Comment." burg: Virginia Polytechnic Institute and Journal of Farm Economics Vol. 49, State University, November 1981. No.3 (August 1967): 741-3. Hayami, Y. and R.W. Herdt. "Market Price Hertel, T.W. "Factor Market Incidence of Effects of Technological Change on In­ Agricultural Trade Liberalization: Some come Distribution in Semisubsistence Additional Results." Australian Journal Agriculture." American Journal of Ag­ ofA gricultural Economics Vol. 35, No. ricultural Economics Vol. 59, No.2 1 (April 1991): 77-107. (May 1977): 245-56. Hertford, R and A. Schmitz. "Measuring Hayes, K. and S. Porter-Hudak. "Dead­ Economic Returns to Agricultural Re­ weight Loss: Theoretical Size Relation­ search." In T.M. Arndt, D.G. Dalrymple ships and the Precision of Measure­ and V.W. Ruttan, eds., Resource Allo­ ment." Journal of Business Economics cation and Productivity in National and and Statistics Vol. 5, No. I (January International Agricultural Research 1987): 47-52. Minneapolis: University of Minnesota Hazell, P.B.R. "A Linear Alternative to Press, 1977. Quadratic and Semi-Variance Program­ Hicks, lR Value and Capital: An Inquiry ming for Farm Planning under Uncer­ into Some Fundamental Principles of tainty." American Journal of Agricul­ Economic Theory. Second ed. Oxford: tural Economics Vol. 53, No. 1 Clarendon Press, 1946. (February 1971): 53-62. Hicks, J.R. The Theory of Wages. New Hazell, P.B.R. and RD. Norton. Mathe­ York: P. Smith, 1948. matical Programming for Economic Higgs, PJ. Adaptation and Survival in Aus­ Analysis in Agriculture. New York: tralianAgriculture: A Computable Gen­ Macmillan, 1986. eral Equilibrium Analysis of the Impact Hazell, P.B.R. and C. Rarnasamy. Green of Economic Shocks Originating Out­ Revolution Reconsidered: The Impact side the Agricultural Sector. Melbourne: of the High-Yielding Rice Varieties in Oxford University Press, 1986. South-India. Baltimore: Johns Hopkins Holloway, GJ. "Distribution of Research University Press, 1991. Gains in Multistage Production Sys­ Heady, E.O. and lL. Dillon. Agricultural tems: Further Results." American Jour­ Production Functions. Ames: Iowa Sate nal of Agricultural Economics Vol. 71, University Press, 1961. No.2 (May 1989): 338-43. Heim, M.N. and L.L. Blakeslee. "Biologi­ Homer, F.B. "Elasticity of the Demand for cal Adaptation and Research Impacts on Exports of a Single Country." Review of Wheat Yields in Washington." Paper Economics and Statistics Vol. 34, No.4 presented at annual meeting of the (November 1952): 326-42. American Agricultural Economics As­ Horton, D., P. Ballantyne, W. Peterson, B. sociation, Reno, 1986. Uribe, D. Gapasin and K. Sheridan, eds. Herdt, RW. and C. Capule. Adoption Monitoring and Evaluating Agricul­ Spread, and Production Impact ofM od- turalResearch. Wallingford, U.K.: CAB References 533 International, 1993. 1993. Hotelling, H. "Edgeworth's Taxation Para­ Huffman, W.E. and J.A. Miranowski. "An dox and the Nature of Demand and Sup­ Economic Analysis of Expenditures on ply Functions." Journal of Political Agricultural Experiment Station Re­ Economy Vol. 40, No.5 (October 1932): search." American Journal of Agricul­ 577-616. tural Economics Vol. 63, No. 1 Hout, M.-F., G. Fox and G.L. Brinkman. (February 1981): 104-18. 'The Returns to Swine Research in Hulten, c.R. "Divisia Index Numbers." Canada." North Central Journal ofA g­ Econometrica Vol. 41, No.6 (Novem­ ricultural Economics Vol. 11, No. 2 ber 1973): 1017-25. (July 1989): 189-201. Hulten, C.R. "Growth Accounting with In­ Hu, F. and 1M. Antle. "Agricultural Policy termediate Inputs." Review ofE conomic and Productivity: International Evi­ Studies Vol. 45, No.3 (October 1978): dence." Review ofA gricultural Econom­ 511-8. ics Vol. 15, No.3 (September 1993): Hulten, C.R. "Growth Accounting When 495-505. Technical Change Is Embodied in Cap­ Hueth, D.L. and R.E. Just. "Applied Gen­ ital." American £Conomic Review Vol. eral Equilibrium Welfare Analysis: Dis­ 82, No.2 (September 1992): 964-80. cussion." American Journal of Idachaba, F.S. "Agricultural Research Al­ Agricultural Economics Vol. 73, No.5 location Priorities: The Nigerian Expe­ (December 1991): 1517-19. rience." In D. Daniels and B. Nestel, Huffman, W.E. "Decision-Making: The eds. Resource Allocation to Agricul­ Role of Education." American Journal tural Research Ottawa: International ofA gricultural Economics Vol. 56, No. Development Research Centre, 1981. 1 (February 1974): 85-97. Industries Assistance Commission (lAC). Huffman, W.E. "Assessing Returns to Ag­ Financing Agricultural Research Can­ ricultural Extension." American Jour­ berra: Australian Government Publish­ nal ofA gricultural Economics Vol. 60, ing Service, June 1976. No.5 (December 1978): 969-75. Jaffe, A.B. 'Technological Opportunity Huffman, W.E. and R.E. Evenson. "Supply and Spillovers of R&D: Evidence from and Demand Functions for Multi­ Firms' Patents, Profits, and Market product U.S. Cash Grain Farms: Biases Value." American Economic Review Caused by Research and Other Poli­ Vol. 76, No.5 (December 1986): 984- cies." American Journal ofA gricultural 99. Economics Vol. 71, No.3 (August Jaffe, A.B. "Characterizing the 'Techno­ 1989): 761-73. logical Position' of Firms, with Appli­ Huffman, W.E. and R.E. Evenson. "Contri­ cation to Quantifying Technological butions of Public and Private Science Opportunity and Research Spillovers." and Technology to us Agricultural Pro­ Research Policy Vol. 18, No.2 (April ductivity." American Journal of Agri­ 1989): 87-97. cultural Economics Vol. 74, No. 3 Jahnke, H.E. and D. Kirschke. "Quantitative (August 1992): 752-6. Indicators for Priorities in International Huffman, W.E. and R.E. Evenson. Science Agricultural Research." Background for Agriculture: A Long-Tenn Perspec­ Working Paper commissioned by FAO tive. Ames: Iowa State University Press, for TAClCGIAR. Food and Agriculture 534 References Organization, Rome, September 1984. Replacement and Depreciation." In W. Mimeo. Sellekaerts, ed., Economics and Econo­ Jain, H.K. Organization and Structure in metric Theory, Essays in Honor of Jan National Agricultural Research Sys­ Tinbergen. London: Macmillan, 1974. tems. ISNAR Working Paper No. 21. Jorgenson, D.W. and Z. Griliches. 'The The Hague: International Service for Explanation of Productivity Change." National Agricultural Research, 1989. Review of Economic Studies Vol. 34, Jarrett, F.G. and R.K. Lindner. "Research No. 99 (July 1967): 249-83. Benefits Revisited." Review of Market­ Josling, T. and S. Tangerman. "Measuring ing and Agricultural Economics Vol. Levels of Protection in Agriculture: A 45, No.4 (December 1977): 167-78. Survey of Approaches and Results." Jarvis, L.S. "Cattle as Capital Goods and Paper presented at the Twentieth Confer­ Ranchers as Portfolio Managers: An ence of the International Association of Application to the Argentine Cattle Sec­ Agricultural Economists, Buenos Aires, tor." Journal ofP olitical Economy. Vol. Argentina, 24-31 August 1988. 82, No.3 (May/June 1974): 489-520. Just, R.E. "American Agricultural Supply." Johnson, G.L. "Discussion: Economic Im­ Paper presented at the International As­ plications of Agricultural Experiments." sociation of Agricultural Economics Journal ofF arm Economics Vol. 39, No. Meetings, Tokyo, Japan, 1991. 1 (February 1957): 390-7. Just, R.E. "Discovering Production and Johnson, H.G. International Trade and Supply Relationships: Present Status and Economic Growth. London: George Future Opportunities." Review of Mar­ Allen and Unwin, 1958. keting and Agricultural Economics Vol. Johnson, H.G. "The Possibility of Income 61, No.1 (April 1993): 11-40. Losses from Increased Efficiency or Just, R.E. and J.M. Antle. "Interactions be­ Factor Accumulation in the Presence of tween Agricultural and Environmental Tariffs." Economic Journal Vol. 77, Policies." American Economic Review No. 305 (March 1967): 151-4. Vol. 80, No.2 (May 1990): 197-202. Johnson, P.R. 'The Social Cost of the To­ Just, R.E. and D.L. Hueth. "Welfare Mea­ bacco Program." Journal of Farm Eco­ sures in a Multimarket Framework." nomics Vol. 47, No.2 (May 1965): American Economic Review Vol. 69, 242-55. No.5 (December 1979): 947-54. Johnson, P.R. ''The Elasticity of Foreign Just, R.E., D.L. Hueth and A. Schmitz. Ap­ Demand for U.S. Agricultural Prod­ plied Welfare Economics and Public ucts." American Journal ofA gricultural Policy. Englewood Cliffs, New Jersey: Economics Vol. 59, No.4 (November Prentice-Hall, Inc., 1982. 1977): 735--6. Kaldor, D. "A Framework for Establishing Johnson, S.R., Z.A. Hassan and R.D. Green. Research Priorities: Reply." Journal of Demand Systems Estimation: Methods Farm Economics Vol. 49, No.3 (Au­ and Practice. Ames: Iowa State Univer­ gust 1967): 743-45. sity Press, 1984. Kamien, M.1. and N.L. Schwartz. Market Jorgenson, D.W. "The Embodiment Hy­ Structure and Innovation. Cambridge: pothesis." Journal of Political Economy Cambridge University Press, 1982. Vol. 74 (February 1966): 1-17. Karagiannis, G. and W.H. Furtan. "Produc­ Jorgenson, D.W. ''The Economic Theory of tion Structure and Decomposition ofBi- References 535 ased Technical Change: An Example 68, No.3 (August 1992): 318--28. from Canadian Agriculture." Review of Kmenta, J. and R. Gilbert. "Small Sample Agricultural Economics Vol. 15, No.1 Properties of Alternative Estimators of (January 1993): 21-37. Seemingly Unrelated Regressions." KARl. Kenya's Agricultural Research Pri­ Journal ofA merican Statistical Associa­ orities to the Year 2000. Nairobi: Kenya tion Vol. 63, No. 324 (December 1986): Agricultural Research Institute, May 1180-2000. 1991. Kohli, U. Technology, Duality and Foreign Karp, L.S. "Dynamic Games in Interna­ Trade: The GNP Function Approach to tional Trade." Unpublished Ph.D. The­ Modeling Imports and Exports. Ann sis. University of California, Davis, Arbor: University of Michigan Press, 1982. 1991. Kennedy, C. and A.P. Thirwa11. "Surveys in Kornbluth, 1. "A Survey of Goal Program­ Applied Economics-Technical Prog­ ming." Omega Vol. I, No.2 (April ress." Economic Journal Vol. 82, No. 1973): 193-205. 325 (March 1972): 11-72. Krueger, A.O. "Government Failures in De­ Khaldi, N. "Education and Allocative Effi­ velopment." Journal of Economic Per­ ciency in U.S. Agriculture." American spectives Vol. 4, No.3 (Summer 1990): Journal ofA gricultural Economics Vol. 9-23. 57, No.4 (November 1975): 650-57. Krueger, A.O., M. Schiff and A. Valdes. Kim, C.S., G. Schaible, 1. Hamilton and K. "Agricultural Incentives in Developing Barney. "Economic Impacts on Con­ Countries: Measuring the Effects of sumers, Growers, and Processors Re­ Sectoral and Economywide Policies." sulting from Mechanical Tomato World Bank Economic Review Vol. 2, Harvesting in California-Revisited." No.3 (September 1988): 255-71. Journal of Agricultural Economics Re­ Krueger, A.O., M. Schiff and A. Valdes, search Vol. 39, No.2 (Spring 1987): eds. The Political Economy of Agricul­ 39-45. tural Pricing Policy. Volume I, Latin Klein, K.K. and W.H. Furtan, eds. eco­ America. Baltimore: Johns Hopkins nomics of Agricultural Research in University Press, 1991. Canada. Alberta, Canada: The Univer­ Lau, LJ. "A Characterization of the Nor­ sity of Calgary Press, 1985. malized Restricted Profit Function." Kling, C.L. 'The Reliability of Estimates of Journal of Economic Theory Vol. 12, Environmental Benefits from Recreation No.1 (February 1976): 131-63. Demand Models." American Journal of Lau, LJ. "On Exact Index Numbers." Re­ Agricultural Economics Vol. 70, No.4 view of Economics and Statistics Vol. (November 1988): 892-901. 61, No.1 (February 1979): 73-82. Kling, c.L. "Estimating the Precision of Lau, LJ. and P.A. Yotopoulos. "Profit, Sup­ Welfare Measures." Journal of Envi­ ply, and Factor Demand Functions." ronmental Economics and Management American Journal of Agricultural Eco­ Vol. 21, No.3 (November 1991): 244- nomics Vol. 54, No.1 (February 1971): 59. 11-18. Kling, c.L. "Some Results on the Variance Lau, LJ. and P.A. Yotopoulos. 'The Meta­ of Welfare Estimates from Recreation Production Function Approach to Tech­ Demand Models." Land Economics Vol. nological Change in World Agricul- 536 References ture." Journal ofD evelopment Econom­ Thesis, University of Minnesota, St. ics Vol. 31, No.2 (October 1989): 241- Paul, 1981. 69. Lindner, RK. "Toward a Framework for Lee, S.M. Goal Programmingfor Decision Evaluating Agricultural Economics Re­ Analysis. Philadelphia: Auerbach, 1972. search." Australian Journal ofA gricul­ Leiby, J.D. and G.D. Adams. ''The Returns tural Economics Vol. 31, No.2 (August to Agricultural Research in Maine: The 1987): 95-111. Case of a Small Northeastern Experi­ Lindner, R.K. and F.G. Jarrett. "Supply ment Station." Northeastern Journal of Shifts and the Size of Research Bene­ Agricultural and Resource &onomics fits." American Journal ofA gricultural Vol. 20, No.1 (April 1991): 1-14. &onomics Vol. 60, No.1 (February Lekvall, P. and C. Wahlbin. "A Study of 1978): 48-58. Some Assumptions Underlying Innova­ Lindner, RK. and F.G. Jarrett. "Supply tion Diffusion Functions." Swedish Shifts and the Size of Research Bene­ Journal of Economics Vol. 75, No.4 fits: Reply." American Journal ofA gri­ (December 1973): 362-77. cultural Economics Vol. 62, No. 4 Lemieux, C.M. and M.K. Wohlgenant. "Ex (November 1980): 841-44. Ante Evaluation of the Economic Im­ Lindner, RK., P.G. Pardey and F.G. Jarrett. pact of Agricultural Biotechnology: The "Distance to Information Source and the Case of Porcine Somatotropin." Amer­ Time Lag to Early Adoption of Trace ican Journal ofA gricultural Economics Element Fertilizers." Australian Jour­ Vol. 71, No.4 (November 1989): 903- nal ofA gricultural Economics Vol. 26, 14. No.2 (August 1982): 98-113. Lichtenberg, E., D.P. Parker and D. Zilber­ Lipton, M.L. "Accelerated Resource Deg­ man. "Marginal Analysis of Welfare radation by Third World Agriculture: Costs of Environmental Policies: The Created in the Commons, in the West, Case of Pesticide Regulation." Ameri­ or in Bed." In S.A. Vosti, T. Reardon can Journal ofA gricultural &onomics and W. von Urff, eds., Agricultural Vol. 70, No.4 (November 1988): 868- Sustainability, Growth and Poverty Al­ 74. leviation: Issues and Policies. Washing­ Lim, H. and C.R Shumway. "Separability ton, DC: International Food Policy in State-Level Agricultural Technol­ Research Institute, 1991. ogy." American Journal ofA gricultural Lipton, M.L. with R Longhurst. New Seeds &onomics Vol. 74, No.1 (February and Poor People. Baltimore: Johns 1992): 120-31. Hopkins University Press, 1989. Lima, M. and G.W. Norton. "Determina­ Little, I.M.D. A Critique of Welfare &0- ci6n de Prioridades de Investigaci6n nomics. Oxford: Oxford University Agropecuaria en Venezuela." Fondo Press, 1960. Nacional de Investigaciones Agro­ Little, I.M.D. and J.A. Mirrlees. Project pecuarias and International Service for Appraisal and Planning for the Devel­ National Agricultural Research, Mara­ oping Countries. London: Heineman cay, Venezuela and The Hague, Septem­ Educational Books, 1974. ber, 1993. Mimeo. Lloyd, A.G., M. Harris and D.E. Tribe. Lindner, RK. "Adoption as a Decision Australian Agricultural Research: Theoretic Process." Unpublished Ph.D. Some Policy Issues. Melbourne: Craw- References 537 ford Fund for International Agricultural tion of Gains from Quality Improving Research,1990. Research in a Multimarket Framework: Lloyd, P. and A. Schweinberger. 'Trade The Case of Barley." Unpublished Expenditure Functions and the Gains Ph.D. Thesis, University of Minnesota, from Trade." Journal of International St. Paul, September 1990. Economics Vol. 24, No.3 (May 1988): Maddala, G.S. Econometrics. New York: 257-97. McGraw Hill, 1977. Lopez, R.E. 'The Structure of Production Mahlstede, J.P. "Long-Range Planning at and the Derived Demand for Inputs in the Iowa Agricultural and Home Eco­ Canadian Agriculture." American Jour­ nomics Experiment Stations." In W.L. nal ofA gricultural Economics Vol. 62, Fishel, ed., Resource Allocation in Agri­ No.1 (February 1980): 38-45. cultural Research. Minneapolis: Uni­ Lopez, R.E. "Estimating Substitution and versity of Minnesota Press, 1971. Expansion Effects Using a Profit Func­ Mansfield, E. "Basic Research and Produc­ tion Approach." American Journal of tivity Increase in Manufacturing." Agricultural Economics Vol. 66, No.3 American Economic Review Vol. 70, (August 1984): 358-67. No.5 (December 1980): 863-73. Lopez, R.E. "Supply Response and Invest­ Mansfield, E. "Price Indexes for R and D ment in the Canadian Food Processing Inputs, 1969-1983." Management Sci­ Industry." American Journal of Agri­ ence Vol. 33, No. 1 (January 1987): cultural Economics Vol. 67, No.1 (Feb­ 124-9. ruary 1985): 40-8. 1985a. Maredia, M.K. The Economics oft he Inter­ Lopez, R.E. "Structural Implications of a national Transfer of Wheat Varieties. Class of Flexible Functional Forms for Unpublished Ph.D. Thesis, Michigan Profit Functions." International Eco­ State University, East Lansing, 1993. nomic Review Vol. 26, No.3 (October Markowitz, H.M. "Portfolio Selection." 1985):593-601. 1985b. Journal of Finance Vol. 7 (March Lucas, R.E. "Adjustment Cost and the The­ 1952): 77-91. ory of Supply." Journal of Political Marschak, J. and W.H. Andrews. "Random Economy Vol. 75, No.4 (August 1967): Simultaneous Equations and the Theory 321-34. of Production." Econometrica Vol. 12, Lynam, J.K. and P.G. Jones. "Benefits of No.3 (July-October 1944): 143-206. Technical Change as Measured by Sup­ Marshall, A. Principles of Economics. ply Shifts: An Integration of Theory and London: Macmillan, 1890. Practice." Centro Intemacional de Martin, M.A. and J. Havlicek, Jr. "Some Agricultura Tropical, Colombia, Febru­ Welfare Implications of the Adoption of ary 1984. Mimeo. Mechanical Cotton Harvesters in the Lyu, S.-J. L., F.e. White and Y.-C. Lu. United States." American Journal of "Estimating Effects of Agricultural Re­ Agricultural Economics Vol. 59, No.4 search and Extension Expenditures on (November 1977): 737-44. Productivity: A Translog Production Martin, W.J. and J.M. Alston. An Exact Function Approach." Southern Journal Approach for Evaluating the Benefits ofA gricultural Economics Vol. 16, No. from Technological Change. Policy Re­ 2 (December 1984): 1-8. search Working Paper Series No. WPS Macagno, L.F. 'The Nature and Distribu- 1024. Washington, DC: International 538 References Economics Department, World Bank, Firms: The Case of Australian Agricul­ November 1992. ture." International Economic Review Martin, W.J. and 1.M. Alston. ''Trade Dis­ Vol. 24, No.2 (June 1983): 323-39. tortions and the Global and Domestic McKenzie, G.W. Measuring Economic Impacts of Technological Change in Welfare: New Methods. Cambridge: Agriculture." Invited paper presented at Cambridge University Press, 1983. the International Trade Research Con­ McKenzie, G.W. and I.F. Pearce. "Welfare sortium Meeting, Calabria, Italy, June Measurement-A Synthesis." Ameri­ 1993. can Economic Review Vol. 72, No.4 Martin, W.J. and J .M. Alston. "A Dual Ap­ (September 1982): 669-82. proach to Evaluating Research Benefits McNally, M.M. "Strategic Trade Interac­ in the Presence of Trade Distortions." tion in the International Wheat Market." American Journal ofA gricultural Eco­ Unpublished Ph.D. Thesis, University nomics Vol. 76, No.1 (February 1994): of California, Davis, 1993. 26-35. Medina Castro, H. "Prioridades de Investi­ Mayshar,1. "On Measures of Excess Bur­ gaci6n Agropecuaria en Costa Rica Bajo den and Their Application." Journal of Differentes Escenarios." San Jose, Costa Public Economics Vol. 43, No.3 (De­ Rica: Instituto Interarnericano de Coope­ cember 1990): 263-89. raci6n para la Agricultura, November McCalla, A.F. and 1.G. Ryan. "Setting Ag­ 1991. ricultural Research Priorities: Lessons Medina Castro, H. "Prioridades de Investi­ from the CGIAR Study." American gaci6n Agropecuaria en Los Paises de Journal ofA gricultural Economics Vol. America Central." San Jose, Costa Rica: 74, No.5 (December 1992): 1095-100. Instituto Interamericano de McCloskey, D.N. The Rhetoric ofEconom­ Cooperaci6n para la Agricultura, June ics. Madison: University of Wisconsin 1993. Press, 1985. Mellor, 1.W. and B.F. Johnston. ''The McElroy, M.B. "Additive General Error World Food Equation: Interrelations Models for Production, Cost, and De­ among Development, Employment and rived Demand or Share Systems." Jour­ Food Consumption." Journal of Eco­ nal ofP olitical Economy Vol. 95, No.4 nomic literature Vol. 22, No.2 (June (August 1987): 737-57. 1984): 531-74. McFadden, D. "Factor Substitutability in de Melo, 1. and S. Robinson. "Trade Policy the Economic Analysis of Production." and Resource Allocation in the Presence Unpublished Ph.D. Thesis, University of Product Differentiation." Review of of Minnesota, Minneapolis, 1962. Economics and Statistics Vol. 63, No.2 McKay, L., D. Lawrence and C. Vlastuin. (May 1981): 169-77. "Production Flexibility and Technical Miedema, A.K. ''The Retail-Farm Price Change in Australia's Wheat-Sheep Ratio, the Farmer's Share and Technical Zone." Review of Marketing and Agri­ Change." American Journal ofA gricul­ cultural Economics Vol. 50, No. 1 tural Economics Vol. 58, No.4 (No­ (April 1982): 9-26. vember 1976): 750-56. McKay, L., D. Lawrence and C. Vlastuin. Mikesell, R.F. "Project Evaluation and "Profit, Output Supply, and Input De­ Sustainable Development." In R. Good­ mand Functions for Multi-Product land, H. Daly and S. EI Serafy, eds., References 539 Environmentally Sustainable Economic Allocating Resources for Developing Development: Building on BruntlaruJ. Country Agricultural Research, World Bank Environmental Working Bellagio, Italy, 6-10 July 1987. Paper 46. Washington, DC: World Moscoso, W., A. Coutu, D. Bandy and Bank, July 1991. G.W. Norton. Contenido Tecnico de la Miller, G.Y., J.M. Rosenblatt and LJ. Investigaci6n del Instituto Dominicana Hushak. ''The Effects of Supply Shifts de Investigaciones Agropecuarias. Re­ on Producers' Surplus." American port prepared for ISNAR. Dominican Re­ Journal ofA gricultural Economics Vol. public: Instituto Superior de Agricultura, 70, No.4 (November 1988): 886-91. June 1986. Miller, W.L. and D.M. Byers. "Develop­ Mullen, J.D. and J.M. Alston. The Returns ment and Display of Multiple Objective to the Australian Wool lruJustry from Project Impacts." Water Resources Re­ Investment in R&D. Division of Rural & search Vol. 9, No.1 (February 1973): Resource Economics Report No. 10. 11-20. Sydney: New South Wales Department Minasian, J.R. "Research and Develop­ of Agriculture & Fisheries, 1990. ment, Production Functions, and Rates Mullen, J.D. and J.M. Alston. ''The Impact of Retum." American Economic Review on the Australian Lamb Industry of Pro­ Vol. 59, No.2 (May 1969):~0-5. ducing Larger Leaner Lamb." Austra­ Mishan, EJ. Elements of Cost-Benefit lian Journal ofA gricultural Economics Analysis. London: George Allen and Vol. 62, No.1 (April 1994): 43--61. Unwin Ltd, 1972. Mullen, J.D., J.M. Alston and M.K. Mishan, EJ. Cost-Benefit Analysis: An In­ Wohlgenant. ''The Impact of Farm and troduction. New York: Praeger Publish­ Processing Research in the Australian ers,1976. Wool Industry." Australian Journal of Mishan, EJ. Introduction to Normative Agricultural Economics Vol. 33, No.1 Economics. New York: Oxford Univer­ (April 1989): 32-47. sity Press, 1981. Mullen, J.D., T.L. Cox and W.E. Foster. Mishan, EJ. What Political Economy Is All "Measuring Productivity Growth in About-An Exposition and Critique. Australian Broadacre Agriculture." Cambridge: Cambridge University Paper presented to the Annual Confer­ Press, 1982. ence of the Australian Agricultural Eco­ Monke, E.A. and S.R. Pearson. The Policy nomics Society. Canberra, February Analysis Matrixfor Agricultural Devel­ 1992. Mimeo. opment. Ithaca: Cornell University Mullen, J.D., M.K. Wohlgenant and D.E. Press, 1989. Farris. "Input Substitution and the Dis­ Moore, J.R. and N.R. Baker. "Computa­ tribution of Surplus Gains from Lower tional Analysis of Scoring Models for U.S. Beef Processing Costs." American R&D Project Selection." Management Journal ofA gricultural Economics Vol. Science Vol. 16, No. 4 (December 70, No.2 (May 1988): 245-54. 1%9): B212-32. Mundlak, Y. "Elasticities of Substitution Moscardi, E.R. "Allocating Resources for and The Theory of Derived Demand." Agricultural Research and Extension in ReviwofE conomic Studies Vol. 35, No. Argentina: Experiences of I.N.T.A." 2 (April 1968): 225-36. Paper presented at the conference on Murphy, JA, W.H. Furtan and A. Schmitz. 540 References 'The Gains from Agricultural Research tural Economics Vol. 59, No.2 (May under Distorted Trade." Journal ofP ub­ 1977): 307-74. lic Economics Vol. 51, No. 2 (June North, D.C. "Government and the Cost of 1993): 161-72. Exchange in History." Journal of Eco­ Musalem, A.R. "Social Rates of Return and nomic History Vol. 44, No.2 (June Other Aspects of Agricultural Research: 1984): 255-64. The Case of Cotton Research in Sao North, D.C. "Institutions, Transactions Paulo, Brazil: Comment." American Costs, and Economic Growth." Eco­ Journal ofA gricultural Economics Vol. nomic Inquiry Vol. 25, No. 3 (July 56, No.4 (November 1974): 837-39. 1987): 419-28. Muth, R.F. 'The Derived Demand Curve Norton, G.W. "Scoring Methods." In D. for a Productive Factor and the Industry Horton, P.G. Ballantyne, W. Peterson, Supply Curve." Oxford Economic Pa­ B. Uribe, D. Gapasin and K. Sheridan, persVol.16, No. 2 (July 1964): 221-34. eds., Monitoring and Evaluating Agri­ Nagy, J. "The Pakistan Agricultural Devel­ cultural Research: A Sourcebook. Wal­ opment Model: An Economic Evalua­ lingford, U.K.: CAB International, 1993. tion of Agricultural Research and Exten­ Norton, G.W. and J.S. Davis. "Evaluating sion Expenditures." Unpublished Ph.D. Returns to Agricultural Research: A Re­ Thesis, University of Minnesota, St. view." American Journal of Agricul­ Paul,1984. tural Economics Vol. 63, No. 4 Neely, W.P., R.M. North and J.E. Fortson. (November 1981): 683-99. "An Operational Approach to Multiple Norton, G.W. and V.G. Ganoza. "Guide­ Objective Decision-Making for Public lines for Allocations of Resources to Water Resources Projects Using Integer Agricultural Research and Extension in Goal Progamming." American Journal Peru." Paper prepared for INIPA and ofA gricultural Economics Vol. 59, No. USAID. Lima, Peru, February 1986. 1 (February 1977): 198-203. Mimeo. Nelson, R.R. and S.G. Winter. An Evolu­ Norton, G.W. and P.G. Pardey. Priority tionary Theory of Economic Change. Setting Mechanisms for National Agri­ Cambridge, MA: Harvard University cultural Research Systems: Present Ex­ Press, 1982. perience and Future Needs. ISNAR Nerlove, M. "Estimates of the Elasticity of Working Paper No.7. The Hague: Inter­ Supply of Selected Agricultural Com­ national Service for National Agricul­ modities." Journal of Farm Economics tural Research, November 1987. Vol. 38, No.2 (May 1956): 496-509. Norton, G.W. and G.E. Schuh. "Evaluating Nerlove, M. "Returns to Scale in Electricity Returns to Social Science Research: Is­ Supply." In c.F. Christ, et aI., eds., sues and Possible Methods." In G.W. Measurement in Economics: Studies in Norton, W.L. Fishel, A.A. Paulsen and Mathematical Economics and Econo­ W.B. Sundquist, eds. Evaluation of Ag­ metrics. Stanford: Stanford University ricultural Research. Minnesota Agricul­ Press, 1963. tural Experiment Station Miscellaneous Nguyen, D. "Intersectoral Distributional Publication 8-1981, St. Paul: University Implications of Agricultural Technical of Minnesota, April 1981. Progress in an Open Economy: An Ex­ Norton, G.W., J.D. Coffey and E.B. Frye. tension." American Journal ofA gricul- "Estimating Returns to Agricultural Re- References 541 search, Extension, and Teaching at the to Research in Distorted Markets: Re­ State Level." Southern Journal ofA gri­ ply." Agricultural Economics Vol. 5, cultural Economics Vol. 16, No. 1 (July No.1 (January 1991): 83-8. 1984): 121-28. Oehmke, J.F. and X. Yao. "A Policy Pref­ Norton, G.W., V.G. Ganoza and C. erence Function for Government Inter­ Pomareda. "Potential Benefits of Agri­ vention in the U.S. Wheat Market." cultural Research and Extension in American Journal ofA gricultural Eco­ Peru." American Journal ofA gricultural nomics Vol. 72, No.3 (August 1990): Economics Vol. 69, No.2 (May 1987): 631-40. 247-57. Office of Technology Assessment (OTA). Norton, G.W., P.G. Pardey, and J.M. Al­ Federally Funded Research: Decisions ston. "Economic Issues in Agricultural for a Decade. Washington, DC: Con­ Research Priority Setting." American gress of the United States, May 1991. Journal ofA gricultural Economics Vol. Ohta, M. "A Note on the Duality Between 74, No.5 (December 1992): 1089-94. Production and Cost Functions: Rate of Norton, G.W., W.L. Fishel, AA Paulsen Returns to Scale and Rate of Technical and W.o. Sundquist, eds. Evaluation of Progress." Economic Studies Quarterly Agricultural Research. Minnesota Ag­ Vol. 25, No.3 (December 1994): 63-5. ricultural Experiment Station Miscella­ Oskam, A "Productivity Measurement, In­ neous Publication 8-1981. St. Paul: corporating Environmental Effects of University of Minnesota, April 1981. Agricultural Production." In K. Burger OECD. The Measurement of Scientific and et al., eds., Agricultural Economics and Technical Activities-"Frascati Man­ Policy: International Challengesfor the ual" 1980. Paris: Organization for Eco­ Nineties. Amsterdam: Elsevier, 1991. nomic Cooperation and Development, Otto, D.M. "An Economic Assessment of 1981. Research and Extension Investments in OECD. National Policies and Agricultural Corn, Wheat, Soybean, and Sorghum." Trade. Paris: Organization for Eco­ Unpublished Ph.D. Thesis, Virginia nomic Cooperation and Development, Polytechnic Institute and State Univer­ 1987. sity, Blacksburg, Virginia, 1981. OECD. Environmental Policy Benefits: Pachico, D., J.K. Lynam and P.G. Jones. Monetary Valuation. Paris: Organiza­ 'The Distribution of Benefits from tion for Economic Cooperation and De­ Technical Change among Classes of velopment, 1989. Consumers and Producers: An Ex-Ante Oehmke, J.F. A Multiple Objective Expla­ Analysis of Beans in Brazil." Research nation of Government Research and Policy Vol. 16, No.5 (October 1987): Target Price Policy. Staff Paper 88-5. 279-85. East Lansing: Department of Agricul­ Pakes, A. and Z. Griliches. "Patents and tural Economics, Michigan State Uni­ R&D at the Firm Level: A First Report." versity, January 1988a. Economics Letters Vol. 5, No.4 (April Oehmke, IF. 'The Calculation of Returns 1980): 377-81. to Research in Distorted Markets." Ag­ Palomino, J. and G.W. Norton. "Determi­ ricultural Economics Vol. 2, No.4 (De­ naci6n de Prioridades de Investigaci6n cember 1988): 291-302. 1988b. en Ecuador." Unpublished Report. Oehmke, J.F. 'The Calculation of Returns Quito, Ecuador: INIAPIFUNDAGRO/ 542 References ISNAR, July 1992a. sity Press, 1991. Palomino, J. and G.W. Norton. "Evalua­ Pardey, P.G., J. Roseboom and B.J. Craig. ic6n de Investigaci6n Agropecuarias en "A Yardstick for International Compari­ Ecuador: EI Case de Marz." Quito, Ecu­ sons: An Application to National Agri­ ador: INIAPIFUNDAGROIISNAR, No­ cultural Research Expenditures." Eco­ vember 1992. Unpublished Report. nomic Development and Cultural (1 992b). Change Vol. 40, No.2 (January 1992): Pardey, P.G. 'The Diffusion of Trace Ele­ 333-49. ment Technology: An Economic Analy­ Pardey, P.G., RK. Lindner, E. Abdurach­ sis." Unpublished M.Sc. Thesis, Univer­ man, S. Wood, S. Fan, W.M. Eveleens, sity of Adelaide, Adelaide, 1978. B. Zhang and J.M. Alston. The Eco­ Pardey, P.G. "Public Sector Production of nomic Returns to Indonesian Rice and Agricultural Knowledge." Unpublished Soybean Research Unpublished AARDI Ph.D. Thesis, University of Minnesota, ISNAR Report. Jakarta and The Hague: St. Paul, 1986. Agency for Agricultural Research and Pardey, P.G. 'The Agricultural Knowledge Development and International Service Production Function: An Empirical for National Agricultural Research, No­ Look." Review ofE conomics and Statis­ vember 1992. tics Vol. 71, No. 33 (August 1989): Paris, Q. An Economic Interpretation of 453-61. Linear Programming. Ames: Iowa Pardey, P.G. and B. Craig. "Causal Rela­ State University Press, 1991. tionships between Public Sector Agri­ Parton, K.A., J.R. Anderson and J.P. cultural Research Expenditures and Makeham. Evaluation Methodsfor Aus­ Output." American Journal ofA gricul­ tralian Wool Production Research Ag­ tural Economics Vol. 71, No.1 (Febru­ ricultural Economics Bulletin No. 29. ary 1989): 9-19. Armidale, Australia: University of New Pardey, P.G. and J. Roseboom.ISNAR Ag­ England, 1984. ricultural Research Indicator Series: A Paulsen, A. and D.R Kaldor. "Evaluation Global Database on National Agricul­ and Planning of Research in the Exper­ tural Research Systems. Cambridge: iment Station." American Journal of Cambridge University Press, 1989. Agricultural Economics Vol. 50, No.5 Pardey, P.G. and S. Wood. ''Targeting Re­ (December 1968): 1149-62. search by Agricultural Environments." Paz, L.J. "A Methodology for Establishing In J.R. Anderson, ed., Agricultural Priorities for Research on Agricultural Technology: Current Policy Issues for Products." in D. Daniels and B. Nestel, the International Community. Walling­ eds. Resource Allocation to Agricul­ ford, U.K.: CAB International, 1994. tural Research Ottawa, Canada: Inter­ Pardey, P.G., B.J. Craig and M.L. Halla­ national Development Research Center, way. "U.S. Agricultural Research De­ 1981. flators: 1890-1985." Research Policy Pearce, D., E. Barbier and A. Markandya. Vol. 18, No.5 (October 1989): 289-96. Sustainable Development: Economics Pardey, P.G.,J. RoseboomandJ.R. Ander­ and Environment in the Third World. son, eds. Agricultural Research Policy: Aldershot: Edward Elgar, 1990. International Quantitative Perspec­ Pearl, R Studies in Human Biology. Balti­ tives. Cambridge: Cambridge Univer- more: Williams and Wilkins, 1924. References 543 Perrin, R.K. "Asset Replacement Princi­ Phlips, L. Applied Consumption Analysis. ples." American Journal ofA gricultural Amsterdam: North-Holland, 1974. Economics Vo. 54, No. 1 (February Pingali, P. and P. Roger, eds. Impact of 1972): 60-7. Pesticides on the Rice Environment and Perrin, R.K. 'The Impact of Component Human Health. Boston: Kluwer Aca­ Pricing of Soybeans and Milk." Ameri­ demic Publishers, 1995. can Journal ofA gricultural Economics Pinstrup-Andersen, P. "Decision-Making Vol. 62, No.3 (August 1980): 445-55. on Food and Agricultural Research Pol­ Perrin, R.K. "Equilibrium Displacement icy: The Distribution of Benefits New Models." EB641 Class Notes. Depart­ Agricultural Technology among Con­ ment of Economics and Business, North sumer Income Strata." Agricultural Ad­ Carolina State University, 1981. Mimeo. ministration Vol. 4, No. 1 (January Perrin, R.K. "Ex-Ante Evaluation of Ex­ 1977): 13-28. perimental Technology." Department Pinstrup-Andersen, P. "A Methodological of Agricultural and Resource Econom­ Note on the Measurement of the Nutri­ ics, North Carolina State University, tion Impact of Agricultural Research." 1992. Mimeo. Paper prepared for workshop on Meth­ Perrin, R.K. and G.M. Scobie. "Market In­ odological Problems in Measuring Re­ tervention Policies for Increasing the search Impact. CGIAR Secretariat, Consumption of Nutrients by Low In­ World Bank, Washington, DC, April come Households." American Journal 1984. ofA gricultural Economics Vol. 63, No. Pinstrup-Andersen, P. and D. Franklin. "A 1 (February 1981): 73-82. Systems Approach to Agricultural Re­ Persley, G.J., ed. Agricultural Biotechnol­ search Resource Allocation in Develop­ ogy: Opportunities for International ing Countries." In T.M. Arndt, D.G. Development. Wallingford, U.K.: CAB Dalrymple and V.W. Ruttan, eds., International, 1990. Resource Allocation and Productivity in Peterson, W.L. "Return to Poultry Re­ National and International Research. search in the United States." Journal of Minneapolis: University of Minnesota Farm Economics Vol. 49, No.3 (Au­ Press, 1977. gust 1967): 656-70. Pinstrup-Andersen, P., N. Ruiz de Londono Peterson, W.L. "International Farm Prices and E. Hoover. ''The Impact of Increas­ and the Social Cost of Cheap Food Pol­ ing Food Supply on Human Nutrition: icies." American Journal of Agricul­ Implications for Commodity Priorities in tural Economics Vol. 61, No. 1 AgriculturaI Research." American Jour­ (February 1979): 12-21. nal of Agricultural Economics Vol. 58, Peterson, W.L. "Land Quality and Prices." No.2 (May 1976): 131-42. American Journal ofA gricultural Eco­ Pomareda, C. "A Spatial Equilibrum Model nomics Vol. 68, No.4 (November for the Central American Agricultural 1986): 812-9. Sector." Guatemala: ECID, 1978 Peterson, W.L. and Y. Hayarni. ''Technical Pope. R.D. ''To Dual or Not to Dual?" Change in Agriculture." In L.R. Martin, Western Journal of Agricultural Eco­ ed.,A Survey ofA gricultural Economics nomics Vol. 7, No.2 (December 1982): Literature. Volume 1. Minneapolis: 337-52. University of Minnesota Press, 1977. Pray, C.E. and C.F. Neumeyer. "Problems 544 References of Omitting Private Investments in Re­ search Policy: International Quantita­ search When Measuring the Impact of tive Perspectives. Cambridge: Cam­ Public Research." In R.G. Echeverria, bridge University Press, 1991. ed., Methods for Diagnosing Research Romano, L. "Economic Evaluation of the System Constraints and Assessing the Colombian Agricultural Research Sys­ Impact of Agricultural Researck Vol­ tem." UnpUblished Ph.D. Thesis, Okla­ ume II, Assessing the Impact ofA gricul­ homa State University, Stillwater, 1987. tural Researck The Hague: Interna­ Romero, C., F. Amador and A. Barco. tional Service for National Agricultural "Multiple Objectives in Agricultural Research,1990. Planning: A Compromise Programming Prescott, E.C. and M. Visscher. "Organiza­ Application." American Journal ofA g­ tion Capital." Journal ofP olitical Econ­ ricultural Economics Vol. 69, No. 1 omy Vol. 88, No.3 (July 1980): 446-61. (February 1987): 78~6. Rao, D.S.P. Intercountry Comparison of Rose, R.N. "Supply Shifts and Research Agricultural Output and Productivity. Benefits: A Comment." American Jour­ FAO Economic and Social Develop­ nal ofA gricultural Economics Vol. 62, ment Paper No. 112. Rome: Food and No.4 (November 1980): 834-44. Agriculture Organization, 1993. Roseboom, 1. and P.G. Pardey. Statistical Ravenscraft, D. and F.M. Scherer. 'The Brief on the National Agricultural Re­ Lag Structure of Returns to Research search System of Kenya. Statistical and Development." Applied Economics Brief No.5. The Hague: International Vol. 14, No.6 (December 1982): 603- Service for National Agricultural Re­ 20. search, November 1993. Ray, A. Cost-Benefit Analysis: Issues and Rosenberg, N. Perspective on Technology. Methodologies. Baltimore: Johns Hop­ Cambridge: Cambridge University kins University Press, 1986. Press, 1976. Ray, S. "A Translog Cost Function Analy­ Ruble. W. "Improving the Computation of sis of U.S. Agriculture, 1939-77." Simultaneous Stochastic Linear Equa­ American Journal ofA gricultural Eco­ tion Estimates." Unpublished Ph.D. nomics Vol. 64, No.3 (August 1982): Thesis, Michigan State University, East 490-98. Lansing, 1968. Richter, M.K. "Invariance Axioms and Runge, C.F. and R.1. Myers. "Shifting Economic Indexes." Econometrica Vol. Foundations of Agricultural Policy 34, No.4 (December 1966): 739-55. Analysis: Welfare Economics When Robinson, S. Multisectoral Models of De­ Risk Markets Are Incomplete." Ameri­ veloping Countries: A Survey. Depart­ can Journal of Agricultural Economics ment of AgriculturaI and Resource Eco­ Vol. 67, No.5 (December 1985): 1010. .. nomics Worldng Paper No. 401. 16. Berkeley: University of California at Russell, D.G. "Resource Allocation in Ag­ Berkeley, 1986. ricultural Research Using Socio-Eco­ Roe, T.L. and P.G. Pardey. "Economic Pol­ nomic Evaluation and Mathematical icy and Investment in Rural Public Models." Canadian Journal ofA gricul­ Goods: A Political Economy Perspec­ tural Economics Vol. 23, No. 2 (July tive." In P.G. Pardey, 1. Roseboom and 1975): 29-52. J.R. Anderson, eds., Agricultural Re- Ruttan, V.W. "The Green Revolution: References 545 Seven Generalizations." International System Constraints and Assessing the Development Review Vol. 19, No.4 Impact of Agricultural Research. Vol­ (August 1977): 13-16. ume I, Diagnosing Research System Ruttan, V.W. Agricultural Research Pol­ Constraints. The Hague: International icy. Minneapolis: University of Minne­ Service for National Agricultural Re~ sota Press, 1982. search, 1990. Ruttan, V.W. "Why Foreign Economic As­ Schimmelpfennig, D. and C. Thirtle. "Co­ sistanceT Economic Development and integration, and Causality: Exploring Cultural Clumge Vol. 37, No.2 (Janu­ the Relationship Between Agricultural ary 1989): 411-24. R&D and Productivity." Journal ofA g­ Ruttan, V. W. "Constraints on the Design of ricultural Economics Vol. 45, No.2 Sustainable Systems of Agricultural (May 1994): 220-31. Production." Ecological Economics Schmitz, A. and D. Seckler. "Mechanized Vol. 10(1994): 209-19. Agriculture and Social Welfare: The Ruttan, V.W. and Y. Hayami. '"Technology Case of the Tomato Harvester." Ameri­ Transfer and Development." Technol­ can Journal ofA gricultural Economics ogy and Culture Vol. 14, No.2 (April Vol. 52, No.4 (November 1970): 569- 1973): 119-51. 77. Ryan, J.G. and 1.S. Davis. A Decision Sup­ Schuh, G.E. and H. Tollini. Costs and Ben­ port System to Assist Agricultural Re­ efits ofA gricultural Research: The State search Priority Setting: Experience at oft he Arts. World Bank Staff Working ACIAR and Possible Adaptations for the Paper No. 360. Washington, DC: The TAC/CGIAR. ACIARIISNAR Project World Bank, October 1979. Paper No. 17. Canberra: Australian Cen­ Schultz, T.W. The Economic Organization tre for International Agricultural Re­ of Agriculture. New York: McGraw­ search, March 1990. Hill, 1953. 1953a. Sakong, Y. and D.J. Hayes. "Testing the Schultz, T.W. "Gauging the Relevant In­ Stability of Preferences: A Nonpara­ come Elasticity." In T.W. Schultz, The metric Approach." American Journal of Economic Organization ofA griculture. Agricultural Economics Vol. 75, No.2 New York: McGraw-Hill, 1953. 1953b. (May 1993): 269-77. Schultz, T.W. "Reflections on Agricultural Samuelson, P.A. "Prices of Factors and Production, Output and Supply." Jour­ Goods in General Equilibrium." Review nal of Fann Economics Vol. 38, No.3 ofE conomic Studies Vol. 21 (1953/54): (August 1956): 748--62. 1-20. Schultz, T.W. 'The Value of the Ability to Sarhangi, R.F., J. Logan, R.C. Duncan and Deal with Disequilibria." Journal of P. Hagan. "Who Benefits from Agricul­ Economic Literature Vol. 13, No. 3 tural Research: Comment." Review of (September 1975): 827-46. Marketing and Agricultural Economics Schultz, T.W. "Uneven Prospects for Gains Vol. 45, No.4 (December 1977): 179- from Agricultural Research Related to 85. Economic Policy." In T.M. Arndt, D.G. Sarles, M. "USAID Experiments with the Dalrymple, and V.W. Ruttan, eds., Private Sector in Agricultural Research Resource Allocation and Productivity in in Latin America." In R.G. Echeverria, National and International Agricultural ed., Methods for Diagnosing Research Research. Minneapolis: University of 546 References Minnesota Press, 1977. Waikato, 1992. Schultz, T.W. "On Economics and Politics Scobie, G.M. and V. Jardine. Investing in of Agriculture." In T.W. Schultz, ed., Wool Research and Development: Allo­ Distortions in Agricultural Incentives. cating Research Budgets. Hamilton, Bloomington: Indiana University Press, New Zealand: Scobie Economic Re­ 1978. search, February 1988. Schultz, T.W. ''The Economics of Agricul­ Scobie, G.M. and R.T. Posada. The Impact tural Research." In C.K. Eicher and J .M. ofH igh-Yielding Rice Varieties in Latin Staatz, eds., Agricultural Development America-with Special Emphasis on in the Third World (Second edition). Colombia. Series JE-01. Cali, Colom­ Baltimore: Johns Hopkins University bia: Centro Internacional de Agricultura Press, 1990. Tropical, April 1977. Scobie, a.M. "Who Benefits from Agricul­ Scobie, a.M. and R.T. Posada. "The Im­ tural Research?" Review of Marketing pact of Technical Change on Income and Agricultural Economics Vol. 44, Distribution: The Case of Rice in Co­ No.4 (December 1976): 197-202. lombia." American Journal ofA gricul­ Scobie, G.M. ''The Demand for Agri­ tural &onomics Vol. 60, No. 1 cultural Research: A Colombian Illustra­ (February 1978): 85-92. tion." American Journal ofA gricultural Scobie, a.M., J.D. Mullen andJ.M. Alston. &onomics Vol. 61, No. 3 (August "The Returns to Investment in Research 1979): 540--45. 1979a. on Australian Wool Production." Aus­ Scobie, a.M. Investment in International tralian Journal of Agricultural &0- Agricultural Research: Some Economic nomics Vol. 35, No.2 (August 1991): Dimensions. World Bank Staff Work­ 179-95. ing Paper No. 361. Washington, DC: Seaton, M.L. "Interactions between Basic The World Bank, October 1979b. and Applied Research: The Example of Scobie, a.M. "On Estimating Money Flex­ Multiple Disease Resistant Potato Culti­ ibility: A Note." American Economist var Development." Unpublished M.S. Vol. 24, No.1 (Spring 1980): 76-8. Thesis, Virginia Polytechnic Institute Scobie, a.M. Investment in Agricultural and State University, Blacksburg, Vir­ Research: Some &onomic Principles. ginia, 1986. CIMMYT Working Paper 08/84. Mex­ Senauer, 8., E. Asp and J. Kinsey. Food ico: Centro Internacional de Mejora­ Trends and the Changing Consumer. St. miento de Maiz y Trigo, August 1984. Paul: Eagan Press, 1991. Scobie, a.M. and W.M. Eveleens. The Re­ Shephard, R. Cost and Production Func­ turn to Investment in Agricultural Re­ tions. Princeton: Princeton University search in New Zealand: 1926-27 to Press, 1953. 1983-84. Economics Division Research Shumway, c.R. "Allocation of Scarce Re­ Report 1/87. Hamilton, New Zealand: sources to Agricultural Research: Re­ Ministry of Agriculture and Fisheries , view of Metbodology." American Jour­ October 1987. nal of Agricultural &onomics Vol. 55, Scobie, a.M. and V. Jacobsen. Allocation No.4 (November 1973): 557-66. of R&D Funds in the Australian Wool Shumway, C.R. "Models and Methods Industry. Hamilton, New Zealand: De­ Used to Allocate Resources in Agricul­ partment of Economics, University of tural Research: A Critical Review." In References 547 T.M. Arndt, D.G. Dalrymple and V.W. Jr. "Impacts of Public Research Expen­ Ruttan, eds., Resource Allocation and ditures on Agricultural Value Added in Productivity in National and Interna­ the U.S. and the Northeast." Journal of tional Research. Minneapolis: Univer­ the Northeastern Agricultural Econom­ sity of Minnesota Press, 1977. ics Council Vol. 12, No.2 (Fall 1983): Shumway, c.R. "Supply, Demand, and 109-14. Technology in Multi-Product Industry: Solow, R.M. "Technical Change and the Texas Field Crops." American Journal Aggregate Production Function." Re­ ofA gricultural Economics Vol. 65, No. view of Economics and Statistics Vol. 4 (November 1983): 748-60. 39, No.3 (August 1957): 312-20. Shumway, C.R. 'The Statistical Base for Solow, R.M. "On a Family of Distributed Agricultural Productivity Research: A Lags." Econometrica Vol. 28, No. 2 Review and Critique." In S.M. Capalbo (April 1960): 393-406. and J.M. Antle, eds., Agricultural Pro­ Souder, W.E. "A Scoring Methodology for ductivity: Measurement and Explana­ Assessing the Suitability of Manage­ tion. Washington, DC: Resources for ment Science Models." Management the Future, 1988. Science Vol. 18, No. 10 (June 1972): Shumway, c.R. and R.J. McCracken. "Use 526-43. of Scoring Models in Evaluating Re­ Sprow, F.B. "Evaluation of Research Ex­ search Programs." American Journal of penditures Using Triangular Distribu­ Agricultural Economics Vol. 57, No.4 tion Functions and Monte Carlo (November 1975): 714-8. Methods." Industrial and Engineering Shumway, c.R., R.S. Saez and P.E. Gottret. Chemistry Vol. 59, No.7 (July 1967): "Multiproduct Supply and Input De­ 35-8. mand in U.S. Agriculture." American Squire, L. and H.G. van der Tak. Economic Journal ofA gricultural Economics Vol. Analysis of Projects. Baltimore: Johns 70, No.2 (May 1988): 330--7. Hopkins University Press, 1975. Sidhu, S. and C. Banante. "Estimating Squires, D. "Long-Run Profit Functions for Farm Land Input Demand and Wheat Multiproduct Firms." American Jour­ Supply in the Indian Punjab Using a nal of Agricultural Economics Vol. 69, Translog Profit Function." American No.3 (August 1987): 558--69. Journal ofA gricultural Economics Vol. Srinivasan, T.N. and J.N. Bhagwati. 63, No.2 (May 1981): 237-46. "Shadow Prices for Project Selection in Silver, J.L. and T.D. Wallace. "The Lag the Presence of Distortions: Effective Relationship between Wholesale and Rates of Protection and Domestic Consumer Prices: An Application of the Resource Costs." Journal of Political Hatanaka-Wallace Procedure." Journal Economy Vol. 86, No. 1 (February of Econometrics Vol. 12, No.3 (April 1978): 97-116. 1980): 375--87. Stallings, J.L. "Weather Indexes." Journal Smith, V.K. "Estimating Recreation De­ ofFarmEconomicsVoI.42,No.l (Feb­ mand Using the Properties of the Im­ ruary 1960): 180--6. plied Consumer Surplus." Land Eco­ Star, S. "Accounting for the Growth of Out­ nomics Vol. 66, No.2 (May 1990): put." American Economic Review Vol. 111-20. 64, No.1 (March 1974): 123-35. Smith, B.L., G.W. Norton and J. Havlicek Steuer, R.E. and F.W. Harris. "Inter-Set 548 References Point Generation and Filtering in Deci­ Economics, Virginia Polytechnic Insti­ sion and Criterion Space." Computing tute and State University, November and Operations Research Vol. 7 (1980): 1985. 41-53. Swanson, E.R. "Problems of Applying Ex­ Stevenson, R. "Measuring Technological perimental Results to Commercial Prac­ Bias." American Economic Review Vol. tice." Journal of Fann Economics Vol. 70, No.1 (March 1980): 162-73. 39, No.1 (February 1957): 382-9. Stiglitz, I.E. ''The Inefficiency of the Stock Teri, J., S.E. Mugogo and G.W. Norton. Market Equilibrium." Review of Eco­ "Analysis of Agricultural Research Pri­ nomic Studies Vol. 49, No.2 (April orities in Tanzania." The Hague: Interna­ 1982): 241-61. tional Service for National Agricultural Stiglitz, lE. "Information and Economic Research, July 1990. Mimeo. Analysis: A Perspective." Economic Thirtle, C.G. and V.W. Ruttan. The Role of Journal Vol. 95 (Supplement 1985): 21- Demand and Supply in the Generation 41. and Diffusion of Technical Change. Stranahan, H.A. and I.S. Shonkweiler. Chur, Switzerland: Harwood Academic "Evaluating Returns to Post Harvest Re­ Publishers, 1987. search in the Florida Citrus-Processing Thompson, G.D. "Choice of Flexible Func­ Subsector." American Journal of Agri­ tional Forms: Review and Appraisal." cultural Economics Vol. 68, No.1 (Feb­ Western Journal of Agricultural Eco­ ruary 1986): 88-94. nomics Vol. 13, No. 2 (December Summers, L.H. "Summers on Sustainable 1988): 169-83. Growth." The Economist 30 May 1992, Thurman, W. "Applied General EqUilib­ p.71. rium Welfare Analysis." American Sumner, D.A. and lM. Alston. Effects of Journal ofA gricultural Economics Vol. the Tobacco Program: An Analysis of 73, No.5 (December 1991): 1508-16. Decontrol. Washington, DC: American 1991a. Enterprise Institute for Public Policy Thurman, W. "On the Welfare Significance Research, 1986. and Nonsignificance of General Equi­ Sundquist, W.B., ed. Evaluating Agricul­ librium Demand and Supply Curves." tural Research and Productivity. Min­ Department of Agricultural and Re­ nesota Agricultural Experiment Station source Economics, North Carolina State Miscellaneous Publication 52-1987. St. University, Raleigh, 1991. Mimeo, Paul: University of Minnesota, 1987. 1991b. Sundquist, W.B., ed. The Economic Im­ Timmer, C.P. "On Measuring Technical Ef­ pacts of Agricultural Research. Pro­ ficiency." Food Research Institute Stud­ ceedings of a workshop sponsored by ies in Agricultural Economics, Trade, the Agricultural Research Institute, and Development Vol. 9, No.2 (1970): Washington, DC, May 1991. 99-171. Swallow, B.M., G.W. Norton, T.B. Tower, E. Effective Protection, Domestic Brumback Jr. and lR. Buss. Agricul­ Resource Costs, and Shadow Prices: A tural Research Depreciation and the General Equilibrium Perspective. Importance of Maintenance Research. World Bank Staff Working Paper No. Agricultural Economics Report No.5. 664. Washington, DC: World Bank, Blacksburg: Department of Agricultural September 1984. References 549 Tower, E. and G. Pursell. On Shadow Pric­ Agriculture." Quarterly Journal ofE co­ ing. World Bank Staff Working Paper nomics Vol. 101, No. 10 (February No. 792. Washington, DC: World Bank, 1986): 101-29. January 1986. Ulrich, A., W.H. Furtan and A. Schmitz. Traxler, G. and D. Byerlee. "A Joint-Prod­ 'The Cost of a Licensing System Regu­ uct Analysis of the Adoption of Modem lation: An Example from Canadian Cereal Varieties in Developing Coun­ Prairie Agriculture." Journal of Politi­ tries." American Journal of Agricul­ cal Economy Vol. 95, No.1 (February tural Economics Vol. 75, No.4. 1987): 160-78. (November 1993): 981-9. UNESCO. Manual for Statistics on Scien­ Traxler, G. and D. Byerlee. "National and tific and Technology Activities. Paris: International Research in the Post­ Division of Statistics on Science and Green Revolution: Complement or Sub­ Technology, UNESCO, June 1984. stitute Investments?" Paper presented at Unnevehr, LJ. "Consumer Demand for the annual meeting of the American Ag­ Rice Grain Quality and Returns to Re­ ricultural Economics Association, San search for Quality Improvement in Diego, CA, August 1994. Southeast Asia." American Journal of Trigo, EJ. Agricultural Research Organi­ Agricultural Economics Vol. 68, No.3 zation in the Developing World: Diver­ (August 1986): 634-41. sity and Evolution. ISNAR Working Unnevehr, LJ. "Assessing the Impact on Paper No.4. The Hague: International Improving the Quality of Food Com­ Service for National Agricultural Re­ modities." In R.G. Echeverria, ed., search, 1986. Methodsfor Diagnosting Research Con­ Tripp, R., ed. Planned Change in Fanning straints and Assessing the Impact ofA g­ Systems: Progress in On{ann Re­ ricultural Research. Volume ll, Assess­ search. Chichester: John Wiley, 1991. ing the Impact ofA gricultural Research. Tsakok, I. Agricultural Price Policy-A The Hague: International Service for Na­ Practioner's Guide to Partial-Equilib­ tional Agricultural Research, 1990. rium Analysis. Ithaca: Cornell Univer­ USDA. Estimates of Producer and Con­ sity Press, 1990. sumer Subsidy Equivalents: Govern­ Tweeten, L. "The Demand for U.S. Farm ment Intervention in Agriculture. ATAD Output." Food Research Institute Stud­ Staff Report No. AGES 880127. Wash­ ies Vol. 7, No.3 (1967): 343~9. ington, DC: U.S. Department of Agri­ Tweeten, L. "The Elasticity of Foreign De­ culture, April 1988. mand for U.S. Agricultural Products: Uzawa, H. "Duality Principles in the The­ Comment." American Journal of Agri­ ory of Cost and Production." Interna­ cultural Economics Vol. 59, No.4 (No­ tional Economic Review Vol. 5, No.1 vember 1977): 737-8. (January 1964): 216-20. Tyers, R. and K. Anderson. Disarray in Vanzetti, D. ''Trade Games: Noncoopera­ World Food Markets: A Quantitative tive Strategies in the International Wheat Assessment. Cambridge: Cambridge Market." Unpublished Ph.D. Thesis, La­ University Press, 1992. Trobe University, 1989. Ulrich, A., W.H. Furtan and A. Schmitz. Varian, H.R. Microeconomic Analysis. "Public and Private Returns from Joint New York: Norton, 1978. Venture Research: An Example from Varian, H.R. 'The Nonparametric Approach 550 References to Production Analysis." Econometrica and Profit Functions." Review of Mar­ Vol. 52, No.3 (May 1984): 579-97. keting and Agricultural Economics Vol. Varian, H.R. Microeconomic Analysis. 3rd 56, No.3 (December 1988): 383-404. Edition. New York: Norton, 1992. Weaver, R.D. ''Multiple Input, Multiple Out­ Venezian, E. and G. Edwards. "A Model for put Production Choices and Technology Setting Agricultural Research Priorities in the U.S. Wheat Region." American in the West Africa Region." Unpublished Journal ofA gricultural Economics Vol. report submitted to the World Bank, 65, No.1 (February 1983): 45-56. Washington, OC, June 1986. Mimeo. Weber, A. and M. Sievers. Instability in Von Oppen, M. and J.G. Ryan. "Research World Food Production: Statistical Resource Allocation-Determining Re­ Analysis, Graphical Presentation, and gional Priorities." Food Policy Vol. 10, Interpretation. Kiel, Germany: Wissen­ No.3 (August 1985): 253-64. schaftsverlag Vauk, 1985. Vo on, J.P. "Measuring Research Benefits Weinberg, G.M. An Introduction to Gen­ from a Reduction of Pale, Soft and Ex­ eral Systems Thinking. New York: udative Pork in Australia." Journal of Wiley, 1975. Agricultural Economics Vol. 42, No.2 Welch, F. "Education in Production." Jour­ (May 1991): 180-84. nal ofP olitical Economy Vol. 78, No.1 Voon, J.P. and G.W. Edwards. "Research (January/February 1970): 35-59. Payoff from Quality Improvement: The White, H. "Using Least Squares to Approx­ Case of Backfat Depth in Pigs." Journal imate Unknown Regression Functions." ofA gricultural Economics Vol. 42, No. International Economic Review Vol. 1 (January 1991): 66-76. 1991a. 21, No.1 (February 1980): 149-70. Voon, J.P. and G.W. Edwards. 'The Calcu­ Williamson, J.e. 'The Joint Department of lation of Returns to Research in Dis­ Agriculture and State Experiment Sta­ torted Markets: Comment." Agricultural tions Study of Research Needs." In W.L. EconomicsVoI.5,No.l (January 1991): Fishel, ed., Resource Allocation in Agri­ 75~2. 1991b. cultural Research. Minneapolis: Univer" Voon, J.P. and G.W. Edwards. 'The Calcu­ sity of Minnesota Press, 1971. lation of Research Benefits with Linear Williamson, O.E. The Economic Institu­ and Nonlinear Specifications of Demand tions of Capitalism New York: Free and Supply Functions." American Jour­ Press, 1985. nal of Agricultural Economics Vol. 73, Willig, R.O. "Consumer's Surplus without No.2 (May 1991): 415-20. 1991c. Apology." American Economic Review Voon, J.P. and G.W. Edwards. "Research Vol. 66, No.4 (September 1976): 589- Payoff from Quality Improvement: The 97. Case of Protein in Australian Wheat." Willis, C.E. and R.D. Perlack. "A Compari­ American Journal ofA gricultural Eco­ son of Generating Techniques and Goal nomics Vol. 74, No.3 (August 1992): Programming for Public Investment, 565-72. Multiple Objective Decision Making." Vousden, N. The Economics of Trade Pro­ American Journal of Agricultural Eco­ tection. Cambridge: Cambridge Univer­ nomics Vol. 62, No.1 (February 1980): sity Press, 1990. 66-74. Wall, C.A. and B.S. Fisher. "Supply Re­ Wingate-Hill, R. and G.R. Davis. "Assess­ sponse and the Theory of Production ment of Research and Development References 551 Projects." JOUrTUJl of the Australian In­ vestigations. New York: Harper & stitute of Agricultural Science Vol. 42, Row, 1976. No.4 (December 1976): 231-37. Young, D.L., R.R. Mittelhammer, A. Ros­ Wise, W.S. and E. Fell. "Supply Shifts and tamizadeh and D.W. Holland. Duality the Size of Research Benefits: Com­ Theory and Applied Production Eco­ ment." American JOUrTUJl of Agricul­ nomics Research: A Pedagogical Trea­ tural Economics Vol. 62, No. 4 tise. College of Agriculture and Home (November 1980): 838-40. Economics Research Bulletin No. EB de Wit, C.T. 'The Use of Interactive Multi­ 1473. Pullman: Washington State Uni­ ple Goal Programming in the Resource versity, (revised) September 1987. Allocation Process of the CGIAR Sys­ Zachariah, O.E.R., G. Fox and G.L. Brink­ tem." Paper prepared for TAClCGIAR, man. "Product Market Distortions and Rome, January 1988. Mimeo. the Returns to Broiler Chicken Research Wohlgenant, M.K. The Retail-Fann Price in Canada." Journal ofA gricultural eco­ Ratio in a Competitive Food Industry nomics Vol. 40, No.1 (January 1989): with Several Marketing Inputs. Depart­ 40-51. ment of Economics and Business Work­ Zeleny, M. "Compromise Progamming." ing Paper No. 12. Raleigh: North Caro­ In lL. Cochrane and M. Zeleni, eds., lina State University, 1982. Multiple Criteria Decision Making. Co­ Wohlgenant, M.K "Demand for Farm Out­ lombia: University of South Carolina put in a Complete System of Demand Press, 1973. Functions." American JOUrTUJl of Agri­ Zentner, R.P. "Analysis of Public Wheat cultural Economics Vol. 71, No.2 (May Research Expenditures in Canada." Un­ 1989): 241-52. published Ph.D. Thesis, University of Wohlgenant, M.K "Distribution of Gains Minnesota, St. Paul, 1982. from Research and Promotion in Multi­ Zentner, R.P. "Returns to Public Invest­ Stage Production Systems: The Case of ment in Canadian Wheat and Rapeseed the U.S. Beef and Pork Industries." Research." In KK. Klein and W.H. American Journal ofA gricultural Eco­ Furtan, eds., Economics ofA gricultural nomics Vol. 75, No.3 (August 1993): Research in Canada. Alberta: The Uni­ 642-51. versity of Calgary Press, 1985. Wood, S. and P.G. Pardey. Agroecological Zentner, R.P. and W.L. Peterson. "An Eco­ Dimensions of Evaluating and Prioritiz­ nomic Evaluation of Public Wheat Re­ ing Research from a Regional Perspec­ search and Extension Expenditures in tive: Latin America and Caribbean. Canada." Canadian JOUrTUJl ofA gricul­ ISNAR Discussion Paper No. 93-15. The tural Economics Vol. 32, No.2 (July Hague: International Service for National 1984): 327-53. Agricultural Research, June 1993. Zhang, B., P.G. Pardey, S. Fan and lM. Yotopoulos, P.A. "From Stock to Flow Alston. "Economics of Technical Capital Inputs for Agricultural Produc­ Change and Productivity Growth in In­ tion Functions: A Microanalytical Ap­ donesian Rice and Soybean Produc­ proach." JOUrTUJl of Farm Economics tion." Department of Agricultural Vol. 49, No.2 (May 1967): 476-91. Economics, University of California, Yotopoulos, P.A. and lB. Nugent. Eco­ Davis, June 1993. Mimeo. nomics of Development-Empirical In- Author Index A B AAEA 153 Baker, N.R. 470 Abdurachman, E. 146,168,186,338 Ball, V.E. 148,153,158,336 Adams,G.D. 189 Ballard, C.L. 77,234 Adams, 1. 26 Balzaca, A. 464 Adusei, E.O. 11,32 Banante, C. 146 Akino,M. 62,212,266 Bandy,D. 470 Allen, R.C. 129 Bantilan, M.C.S. 87,420 Alston, J.M. 7,12,17,44,47,4861-64, Barbier, E. 33 69,71,72,75,77,80,88,90,98,115, Barco, A. 447,449 119,133,146,148,153,162,168,186, Barney, K. 60,299 200,208,213,221,222,226,234,235, Barton, P. 149 237,244,253,254,256,262-264,266, Baumol, WJ. 52 268-270,273,282-286,289,297,314, Beach,D.E. 77 319,322,331,333,335,338,340,364, Beattie, B.R. 143, 189, 191 398,400,411,511 Belsley, D.A. 189 Amador, F. 447,449 Bengsten, D.N. 173 Anania,G. 266 Berndt, E.R. 136, 149 Anderson, J.E. 234,379 Bernhart, K. 331 Anderson, J.R. 3, 10,34,36,38,98, 107, Bernstein, 1.1. 185 133,175,351,365,366,367,368,453, Bhagwati, J. 56,286 487,509,510 Bieri, 1. 42 Anderson, K. 80,319 Binswanger, H.P. 10,67,78,83,100, Andrews, W.H. 100, 189 106,114,137,141,146,166,229,321 Anthony, G. 362 464,490 Antle,1.M. 77, 106, 133, 136, 137, Birdsall, N. 33,325 139,141,166 Blackorby, C. 137, 139, 150 Archibald, S.O. 133 Blakeslee, L.L. 11 Arrow, K.J. 14,36-38,88 Bocksteal, N.E. 47 Asp, E. 510 Boehlje, M. 449 Ayer, H.W. 61,78 Boisvert, R.N. 145 Bosch, DJ. 442,448 Bottomley, J.A. 54 553 554 Author Index Bouchet, F.e. 114 Collins, KJ. 324 Boulding, K.E. 43 Collinson, M. 471,476,480 Boussard, J.-M. 453 Colman,D. 102,104,152,321 Bowers, A. 106,166,321 Constantine, J.H. 340 Boyce, J.K. 489 Contant, R.B. 54 Bredahl, M.E. 100,324,335 Cooke, S.C. 65,337,338 Brennan, J.P. 10,60,186,243,342 Cooper, W.W. 445 Brinkman,O.L. 26,75,104, 114, 151 Corden, W.M. 82 266,268,334 Comes, R. III Brown, W.O. 189 Cox, T.L. 98,101,108,116,118, 119, Brown-Andison, N. 114,151 167, 174 Browning, E.K. 77 Craig, B.J. 106, 107, 131, 153, 155, 157, Brumback, T.B. Jr. 183 159,164,166,167, 172, 173, 190, Bruno, M. 56 202,349 Bumiaux, J. 80 Craswell, E. 471,476,480 Burt,O.R. 153,321 Coutu, A. 464,470 Buse, R.C. 232 Crosson, P. 379 Buss, J.R. 181 Currie, J.M. 44,50 Byeriee, D. 17,56,178,313,341, 357,489 D Byers, D.M. 455 Dabezies, M. 464 Dalrymple, D.O. 77,269,333 C David, C.C. 82 Candler, W. 449 Davidson, B.R. 339 Capalbo, S.M. 77,137,139,141 Davis, J.S. 10,30,40,54,63,67,68,69, Capule, C. 178 71,87,100,110,114,208,213,221, Carew, R. 146 222,225,253,254,256,262,263,313, Carter, C.A. 320,324 333,346,358,362,422 Carter, H.O. 200,297 Davis, S.R. 34 Cassells, J.M. 113, 152,205, 321 Deaton, A.S. 320 de Castro, J.P.R. 212 Deininger, K.W. 131, 157, 159 Cessay, S. 312,464,476 Denison, E.F. 188 Chalfant, J.A. 100, 116, 119, 131, 143, Denny,M. 131 144,268,455 Dey,M. 314,420,464,476,478,481, Chambers, R.O. 100,111,191,266, 482,480 286,336 Diewert, W.E. 100,121,127,129,131, Chang,C.C. 74,80,230,314 146,254,336 Chaparro, F. 464 Dillon, J.L. 36,107,143,366,367, Charnes, A. 445 453,464 Chavas, J.-P. 98,101,116,118,119, Dixon, P. 80,114 152, 167 Dorfman, J.H. 153 Chen, P. 146 Drechsler, L. 128 Chipman, J.S. 43,44 Duncan, R.C. 71, 114 Christensen, L.R. 100, 145 Dyer,P.T. 34 CIMMYT 356 Cline, P.L. 30 E Cline, W.R. 33, 174 Echeverria, R.O. 81,338,411 Coble, K.H. 230,314 Eddelmann, B.R. 230, 314 Cochrane, W.W. 52 Edwards,O.W. 10,17,30,49,60,63, Coffey, J.D. 189 67-69,71,74,75,208,213,222,225, Cohon, J.L. 444,447,458 243,253,254,256,262,263,266,268, Author Index 555 270,273,282,283,285,313,319,337, G 398,400,464,470,476,480,481,511 Gallant, A.R. 100, 109, 144 Ehui, S.K. 133 Ganoza, V.G. 40,54,213,229,245,266, Eichorn, W. 121 314,321,464,490 Espinosa, P. 464,476 Gardiner, W.H. 320,324 Eveleens, W.M. 146,168,186,198,335, 338,340 Gardner, B.L. 16,17,43,75,88,152, 153,253,264,266,268,269 Evenson, R.E. 8,10,12,24,26,31,57, 67, 100, 106, 107, 146, 169, 174, Geigel, J. 165 176, 183,185,186,201,202,213, Gijsbers, G. 426 328,335,489 Gilbert, E. 312,464,476 Gilbert, R. 149 Godyn,D.L. 60 F Goodwin, B.K. 322 Fan, S. 146,148,168,186,336,338,340 de Gorter, H. 16,75,88,268-270,318 Gottret, P.E. 148 FAD 319 Fare, R. 120 Graham-Tomasi, T. 297 Ferreira, G. 411,464 Green, R.D. 320 Farris, D.E. 71,241,242,251,264 Griliches, z. 10,21,26,30,40,54,100, Fawson,C. 116 106,132,153,157,158,172,175,183 Feder, G. 178,356 188,190,200,321,335,339,357 Fell, E. 63 Gross, H.D. 464,476 Ferguson, C.E. 191 Grosskopf, S. 120 Findlay, C.C. 77 Gruen, F.H. 286 Finkelshtain, I. 455 Gryseels, G. 464,470,471,476,480 Fisher, B.S. 102,152 Fisher, I. 121 H Fisher, K.S. 341 Hallaway, M.L. 172,173 Fishel, W.L. 34,351 Hamilton, J. 60,299 Flacco, P.R. 116 Hanoch,G. 116 Flores-Moya, P. 213 Haque, E.A.K. 104,266,268 Fortson, J.E. 445 Harberger, A.C. 40,41,43,75,88,91, Foster, W.E. 108,116,118,174 208,373 Fox,G. 26,75,77,104,114,151,266, Hardaker, J.B. 37,366,367,453,509 268,269,333,334,489 Harris, F.W. 449 Franklin, D. 442,448 Harris, M. 490,491 Franzel, S. 17 Hart, O. 52 Freebairn, J.W. 10,17,30,67-69,71,75, Hassan, Z.A. 320 153,213,222,225,253,254,256,262, Hatanaka, M. 184 263,266,268,270,273,282,283,285, Hausman, J.A. 44,47 313,319,337,398,400 Havelick, J.G. Jr. 183,213,470 French, B.C. 153 Hayami, Y. 62,67,83,212,229, Freund, R.J. 453 230,266 Friedman, M. 246 Hayes, OJ. 119, 120 Frisch, R. 320 Hayes, K. 47 Frye, E.B. 189 Hazell, P.B.R. 38,83,175,365,442, Fulginiti, L.E. 27, 102, 106, 107, 453,455 114,322 Heady, E.O. 143 Fullerton, D. 77 Heien, D. 153 Furtan, W.H. 17,75,139,243,244, Heim,M.N. 11 266,286 Herdt, R.W. 83,178,229,230 Fuss, M. 131,144,191 Hertford, R. 52,61,212 556 Author Index Hicks, J.R. 44,45,47,80,136,138, KARl 464 139, 146 Kassam, A. 471,476,480 Higgs, PJ. 80 Kennedy,C. 153 Holland, D.W. 111,112,146 Khaled,M. 136 Holloway, G.J. 71,251,253,264 Kim,C.S. 60,299 Hoover, E. 85,230,321 Kinsey, 510 Horner, F.B. 324 Kirschke, D. 464 Horton, D. xxxi Kislev, Y. 24,26,176,185 Hotelling, H. 146 Kling, c.L. 47 HU,F. 166 Kmenta, J. 149 Hueth, D.L. 44,74,210,232,233, Kohli, U. 114 235,238 Kornbluth, J. 445 Huffman, W.E. 8,26,31,106,107,146, Krueger, A.O. 76,82,293,318 169,174,178,191,202,336 Kuh,E. 189 Hulten, c.R. 125,137,157 Hurd, B.H. 77,88 L Lantican, J.M. 420 I Larson, D.M. 44,47,48, 116 lAC 312 Lau, LJ. 100,106,127,145,166 Idachaba, F.S. 464 Lawrence, D. 146 Igbal, M. 341 Lee, S.M. 445 Leiby, J.D. 189 J Lekvall,P. 357 Jacobsen, V. 34,39,314,334,355,358, Lemieux,C.M. 244,245,253,331 364,367,368,420,426,443,450,453, Librero, A.R. 464 457,461,462,475 Lichtenberg, E. 77 Jaffe, A.B. 185 Lim, H. 116 Jahnke, H.E. 464 Lima, M. 314,464,476,480,482 Jain, H.K. 5 Lind, R.c. 14,36-38 de Janvry, A. 42 Lindner, R.K. 49,64,114,146,168,178, Jaramillo, H. 464 186,208,331,337,338,340,356,511 Jardine, V. 21,470,487,490 Lipton, M.L. 33, 83 Jarrett, F.G. 49,64,114,186,208, Little, I.M.D. 50 331,337 Lloyd, A.G. 490,491 Jarvis, L.S. 153 Lloyd, P. 234 Johnson, H.G. 286 Longhurst, R. 83 Johnson, P.R. 289,324, 339 Lopez,R.E. 143,146,266,286,336 Johnson, S.R. 151,320 Lovell, C.A.K. 137, 139 Johnston, B.F. 266 Lucas, R.E. 39 Johnston, B.G. 60 Lynam, J.K. 62, 362 Jones, P.G. 62, 362 Jones, R.L. 77 M Jorgenson, D.W. 100, 145, 157, 158, 188 Macagno, L.F. 243 Josling, T. 319 Maddala, G.S. 179 Just, R.E. 44,74,77,102,143,152,210, Mahlstede, J.P. 464,470 232,233,235,238,321,356 Makeham, J.P. 36,487 Mansfield, E. 26,173 K Maredia, M.K. 347 Kaldor, D.R. 470 Markandya, A. 33 Kamien, M.I. 175 Markowitz, H.M. 453 Karagiannis, G. 139 Marks, D.H. 444,447,458 Author Index 557 Marschak, J. 100, 189 Myers, W. 324 Marshall, A. 40,44,45,47,79 Martin, B.R 339 N Martin, MA 213 Nadiri, M.I. 185 Martin, W.1. 75,80, 115,234,235,237, Nagy, 1. 229 266,268,282,284,286,314,331, Neary,1.P. 234 398,511 Neely, W.P. 445 Matthews,1.L. 153 Nelson, RR 177 Mauldon, RG. 339 Nerlove, M. 146,152,320 Mayshar, J. 235 Neumeyer, C.F. 26 McCalla, A.F. 266,313,471,476,480 Nguyen, D. 212,266 McCarl, BA 74,80,230,314 Nielson, D.J. 16,75,88,266,26, McCloskey, D.N. 379 269 McCracken, R.J. 464,470 Norris, M. 120 McElroy, M.B. 149 North, D.e. 51 McFadden, D. 143,146,191 North, RM. 445 McGuckin, T. 133, 136, 141 Norton, G.W. 11, 32,40, 54, 63, 75, 90, McKay, L. 146 114,183,189,208,213,229,245,266, McKenzie, G.W. 44 268,270,312,314,318,321,338,411, McNally, M.M. 285 420,442,448,453,464,470,476,478, Medina Castro, H. 464,471,476 480,481,482,490,511 Mellor, lW. 266 Nugent, J.B. 153 de Melo, J. 80 Miedema, A.K. 253 Mikesell, RF. 33 o Miller, W.L. 455 OECD 22,168,319 Mills, B. 312,464,476 Oehmke, J.F. 75, 266, 268, 270, 286 Minasian, J.R. 175 Office of Technology Assessment 487 Miranowski, JA 191 Ohta, M. 136 Mishan, E.J. 43,44,48,49,51, 337, 364 Oram, P. 10,40,54,67,68,213,221, Mittelhammer, RR 111, 112, 146 222,225,313,333,346,362,422 Mjelde,1. 74,80 Orden, D. 114 Monke, EA 56 Oskam, A. 133 Montes, G. 464 Otsuka, K. 82 Monyo,1. 471,476,480 Otto, D.M. 104,150,186,334 Moore, 1.e. 43, 44 Moore, R.J. 470 p Morris, M. 489 Pachico, D. 62 Moscardi, E.R 464 Pakes, A. 106,175 Moscoso, W. 464, 470 Palomino, 1. 314,338,411,464,476, Muellbauer, 1. 320 478,482 Mugogo, S.E. 464 Pardey, P.G. 3,10,17,25,26,75,90, Mullen, lD. 7, 12,71,72, 108, 116, 118, 98, 106, 107, 130, 133, 146, 148, 153, 174,213,226,234,241,242,244,251, 156,157,159,164,166,167,168,172, 253,254,262,264,269,314,333,364, 173,175,186,187,190,200,202,266, Mundlak, Y. 106,143,150,165, 268,269,297,335,338,339,343,348, 191,321 349,358,411,420,428 Murphy, JA 44,50,75,266,286 Parker, D.P. 77 Musalem, A.R 78 Paris, Q. 442,448 Muth, RF. 253,254,256,267 Parmenter, B. 80 Myers, RJ. 52 Parton, K.A. 36,487 Paulsen, A. 470 558 Author Index Paz, L. 464 Russell, RR 150 Pearce,I.F. 33,44 Ruttan, V.W. 7,10,13,33,67,83,176, Pearl, R 358 266,489,490 Pearson, S.R 56 Ryan, J.G. 10,40,54,67,68,213,221, Perlack, RD. 444,445,447 222,225,313,333,346,362,422, Penin,R.K. 20,27,85,102,106,107, 464,490 114,222,230,240,264,322,331 Persley, GJ. 8,176 Peterson, W.L. 40, 100, 163,321,335 s Philips, L. 320 Saez, RS. 148 Pingali, P. 76, 293, 482 Sakong, Y. 119,120 Pinstrup-Anderson, P. 85, 230, 321, Samuelson, P.A. 146 442,448 Sarles, M. 5 de Polanco, E.H. 178, 357 Schaible, G. 60,299 Pomareda, C. 40,54,213,229,24,266, Scherer, F.M. 182,183 314,321 Schiff, M. 76, 293, 317 Pope, RD. 146 Schimmelpfennig, D. 190 Porter-Hudak, S. 47 Schmitz, A. 17, 40, 42, 44, 50, 52, 61, Posada, RT. 61,83,230,339 74,75,78,210,212,232,233,238, Pray, C.E. 26 243,244,266,286 Prescott, E.C. 39 Schuh, G.E. 61,78,212,511 Putman, J. 12 Schultz, T.W. 54,98, 132, 156,266,321 Schwartz, N.L. 175 Q Schweinberger, A. 234 Quilkey, I.I. 153 Scobie, G.M. 21,24,34,39,49,61,69, 71,83,84,85,114,198,222,230,253, 263,314,321,334,335,339,354,358, R 364,366,367,368,420,426,443,450, Ramasamy, C. 83 453,457,461,462,470,475,487,490 Rao, D.S.P. 318 Seaton, M.L. 8 Rausser, G.C. 16,75,88,266,268,269 Seckler, D. 40, 52, 78 Ravenscraft, D. 182, 183 Senauer, B. 510 Ray, A. 324 Shabman, L.A. 442, 448 Ray,S. 58,146 Shephard, R 146 Richardson, J. 74,80 Shoven, J.B. 77,234 Richter, M.K. 125, 126 Shumway, C.R 116,146,148,153,464, Robinson, S. 80 470,487,490 Roe, T.L. 17,75,266,268,269 Sidhu, S. 146 Roger, P. 482 Sievers, M. 38 Romano, L. 127 Silver, J.L. 184, 185 Romero, C. 447,449 Smith, V.H. 340 Rose, RN. 49,61,63,64,114,331,337 Smith, V.K. 47, 183 Roseboom, J. 3,10,106,107,153,155, Solow,RM. 134,183 166,168,428 Spencer, D.S.C. 133 Rosenberg, N. 175 Sprow, F.B. 366,369 Rostamizadeh, A. 111, 112, 146 Srinivasan, T.N. 56 Rothschild, M. 116 Stallings, J.L. 165 Rowe, J. 312,464,476 Star, S. 131,157 Ruble, W. 150 Steer, A. 33,325 Ruis de Londono, N. 85,230,321 Steuer, RE. 449 Runge,C.F. 52 Stevens, V. 146 Russell, D.G. 443,457 Stevenson, R 137, 141 Author Index 559 Stiglitz, lE. 52 Voeller, J. 121 Strand, I.E. 47 von Oppen, M. 464 Summers, L.H. 293 Voon, lP. 49,60,63,74,75,208,243, Sumner, D.A. 289,322 244,511 Sundquist, W.B. 65,165,337,338 Vousden, N. 234 Sutton,J. 80 Swallow, B.M. 183 W Swanson, E.R. 339 Wahlbin, C. 357 T Wall,C.A. 102 Wallace, N.E. 116 Tangerman, S. 319 Wallace, T.D. 184, 185 Taylor, e.R. 143, 191 Weaver, R.D. 146 Teri, J. 464,476 Weber, A 38 Thirlwall, AP. 153 Weinberg, G.M. 356 Thirtle, C. 190 Welch,F. 156 Thompson, G.D. 131 Welsch, R.E. 189 Thurman, W. 233,235 Whalley, J. 77,234 Thursby, M.C. 137,139 White,H. 144 Timmer, e.P. 153 Williamson, lC. 464,470 Tisdell, C. 114 Williamson, O.E. 51 Torres, R. 464 Willig, R.O. 44,48 Tower, E. 56 Willis, C.E. 444,445,447 Traxler, G. 313,341,347 Winter, S.G. 177 Tribe, D.E. 490,491 Wise,W.S. 63 Trigo, EJ. 5 de Wit, e.T. 443,471, 476, 480 Tripp, R. 9 Wohlgenant, M.K. 61,62,63,64,71, Tsakok, I. 320 208,213,222,226,234,241,242,244, Tweeten, L. 324 245,251,253,254,256,264,333 Tyers, R. 80,319 Wood,S. 10,146,168,186,187,338, 340, 343, 348 U Worthington, V.E. 153,321 Ulrich, A 17,243,244 Umali, D.L. 178,356 Y UNESCO 168 Unnevehr, LJ. 74,243,244,511 Yang, M.e. 106,166,321 USDA 319 Yao,X. 75 Uzawa, H. 146, 150 Yotopoulos, P.A. 100,106,153,157,166 Young, D.L. 111,112,146 V Valdes, A 76,293,318 Z van der Mensbrugghe, D. 80 Zachariah, O.E.R. 26,75,266,334 Vanzetti, D. 285 Zeleny, M. 447 Varian, H.R. 98,112,116,117,190 Zentner, R.P. 104, 114, 150 Venezian, E. 464,470,476,480,481 Zhang, B. 119,146,148,168,186,336, VincentD. 80 338,340 Visscher, M. 39 Zhang,Z. 120 Vlastuin, e. 146 Zilberman, D. 77,356 Subject Index A economics 5, 506 Academic research priorities 487 exports 87 ACIAR 362 machinery 175,192 Ad valorem tariff 221 Agricultural research (see also Research) (see also Tax) complementarities with teaching fixed or variable 279 and extension 13 Adaptive research 67 councils 4 Adaptive-expectations model 152 Agroclimatology 461 Adjustment costs 20, 39 Agroecological zones 6,9,221,299, and organizational capital 39 309,342,343,348,505 Adopters vs nonadopters 228-229 Alfred Marshall 207 Adoption· 9, 22, 28, 35, 58, 83,110,349 Allocating research resources 372,442 piecemeal 178 Allocative efficiency Adoption lag 29,177,366 and market distortions 266 (see also Research lag) Analyst skills 503 linear 30 Animals 307 logistic 30,357-358 (see also Livestock and Capital) polynomial 30 breeding stock vs traction animals 157 trapezoidal 31,357 health 479 Adoption parameters 467,468,478, nutrition 461,476,479 501,510 welfare 76, 293 ceiling rate 478,491 Argentina 470 domestic vs other countries 491 Asia 5,82,163,322,344,505,507 elicited 356, 357 Australia 37,72,80, Ill, 119, ex post studies 35 226,264,283,342,364 rate 23,479,485,490 Australian Wool Africa 7,37,344,505 Corporation 443,457,475 Aggregating inputs and outputs 121,123 Research Council 366 Aggregation biases 66, 121-125 A voidable fixed costs 48 Agribusiness 9 Agricultural B chemicals 164,175,294 Balance-of-trade function 80, 234-236, 561 562 Subject Index 238 factor of proportionality 160 B~~~ ~ln generation 78 black sigatoka 176 lightbulb or one-hoss-shay B~gladesh 5, 111,470,477 depreciation 158 Barley 154, 1n market imperfections 14 malting characteristics 243 physical 7, 157 Bayesi~-type approach 328 research expenditures 7,171-173 Beef 71,150,231,233,239 service flow profiles 159 (see also Cattle, livestock, ~d Meat) stock vs service flows 157 Benchmarking 315,326,340-343, Tornqvist-Theil Divisia 162 420,423,427,429,477,478 Cardinality 472 ex post studies 316 C~h crops 82 Benefit tr~sformation curves (BTC) Cassava 488 88,89,90,373-375,445,468 Cattle 157,4n empirically derived 444 (see also Beef, livestock, ~d Meat) Benefit-cost (see Cost-benefit) Ceiling prices 269,278-280 Benefits (see Research benefits) Cereals 168, 177 Biasing research 503 CGIAR 68,313,320,443,471 Biased technical ch~ge 84, 250-252, Chain-of-comm~d approach 490 254,262,265,471,504 Cheap food policy 83,271 Muth models 264 vis-a-vis export taxes 286 Biological technologies 83 Cheese 231 Biotechnology 8, 168 (see also Dairy) Black markets 272 Chemicals 31 Blunt instrument 81, 310, 375, 450, (see also Agricultural chemicals) 468,503,510 inputs 9 Bounded rationality 51 pollution 133 Bovine somatotropin 329 Chickens 157,231,233,239,240 Box-Cox tr~sformation 150 (see also Broiler and Poultry) Brewers 243 China 85 Broilers 150 Choice of technology (see also Chicken, Poultry, ~d production functions 110 ~d livestock) Cigarettes 244 Brown pl~thopper 176 Climate 37,221 Budget cut, distributing Cobb-Douglas among programs 34,371 functional form 108, 116, 179 Buffalo 157 cost function 112,147,148 Buffer stocks 39 growth accounting 191-193 Butter 231 production function 102, 109, 189, (see also Dairy) 193,194-196,336 "self-dual" 147 C Cocoa 163 California 299,476 Coconuts In Cal~~ ~1 Coefficient of variation 481 C~ada 150,487 Coffee 13,86,163 Capital 157-163, 164,207 Colombia 13,83,339,470 aggregate service flows 161 food policy 222 biological 157 Committee approach 490 breeding stock vs traction ~imals 157 Common-property issues budgeting 32,362-364, 393 ~d measuring l~d 164 declining-bal~ce depreciation 159 Communications 104, 106 Subject Index 563 Comparisons, paired 490 deadweight cost of taxation 77, 269 Comparative advantage 56,75,87 deadweight cost of government 471,503 spending 77 and research 16 of decision making 15 Comparative-static model 208 -efficiency index 337 Compensated demand 235 environment 379 Compensating variation 235 least-cost approach 320 Compensation principle 42, 43, 81 minimization 103,118 Competitive structure of markets 13, 14 opportunity (see Opportunity cost) Complementarity reduction induced by research 491 effects 485 reduction, per unit 511 in production 231 reductions, size and timing of 502 Complements in demand 73 sunk 51,307 Compromise programming transactions 14, 16,51-52 filtering techniques 449 transportation 106, 112,221,222, ideal point 447 226, 317, 390 Computable general equilibrium Cost-benefit (COE) models 80 methods 20, 32, 54-55, 311, 378 Congruence approach 57,464,488-489, price effects 54 493 ratio 110, 363, 364 Constant returns to scale 241, 254, 265 Cost function 103,104,111,241,331, Consultative Group on International 334,336,337,412,504,505 Agricultural Research (see CGIAR) Cobb-Douglas 147, 148 Consumer benefits rough methods for estimation issues 149 apportioning 230 error terms 149 Consumer subsidy equivalent 319 joint estimation Consumer surplus 41,48,315,361,385 with input demand 113 compensating vs equivalent long vs short run 112 variation 45,46 multi-output version 112 disaggregation 228-231 output-constrained domestic 214 factor-demand functions 111 externalities 51 regularity conditions 113 Marshallian vs Hicksian share equations 149 measures 40,236 translog 148 money metric 44, 45 vs profit functions 147 trade 66 Costs and benefits of research value judgments 43 causal relationship between 350,501 vs compensating variation 44 Cotton 243 vs equivalent variation 44 Criteria 494 Consumers 6,9,17,41,69,507 as objectives 465 foreign 69 choice of units 475 groups 84 collecting information Contingent valuation 22 related to 482 Com 111,150,322,457,488 combining 484 Costs commodity research adjustment costs 20,39 programs 476-477 anticipated proportional double-counting 471,478,480 reduction in 468 for noncommodity research programs avoidable fixed costs 48,341,377 or program components 478 borrowing 370 inaccurate 480 changes 327,337,340,477,505 incompatible 472 564 Subject Index multiple 463 functional form 35,45,49 multiplicative rather than general-equilibrium 232,233 additive 478 general- vs partial- overlapping 480 equilibrium 210,212 photoperiodicity 356 Hicksian vs Marshallian 45 poor proxies 471 local approximation 45 . vis-a-vis objectives 472 Marshallian 45,231,236 Crops 38,167,481 money income 49 (see also specific crops and Plants) nature of shifts 49 area 468 prices of other goods 49 breeding 154 shifts 232, 245, 325 cash 82 supply-and-demand curve, insurance 459 linear vs nonlinear 49,406 management 175,461,479,484 tastes 49 perennial 153 Demand elasticity programs 486 (see Elasticity of demand) protection 461,479 Demand shift residues 154 population effects 245, 325-326, upland 473 387-388 varieties 17, 175,501 per capita income effects 245, 325, Cross-commodity effects of 326,387,388 research 73,234-243 Desertification 15, 76 Cytology 169 Developed countries 3, 7 Developing countries 3,5 D Discount rate 32, 158, 366, 393, Dairy 150, 157322 469,503 (see also Butter and Milk) conserving natural resources 33 processing industry 231 externalities 33 David Ricardo 207 in less-developed countries 325 Deadweight intergenerational distributive cost of taxation 77,269 weights 33 cost of government spending 77 priority-setting analyses 324-325 loss 91 risk premium 324 Decision makers' preference social vs private 297 personal vs professional uncertainty 33 judgment 459 variable vs fixed 363 in mathematical programming vis-a-vis sustainability 294 models 444 Discounting 490 Decision-making costs 15 vis-a-vis natural resource Decisions depletion 297 strategic 7 Discovery function 57 tactical 6 Diseases 85 Decoupled income transfer 271 Distorted markets 92 Deficiency payments 276-277 Distorted trade-expenditure Deforestation 15,76 function 234 Degree of risk aversion 510 Distribution Delegation of authority 509 income 81 Delphi method 474,490 objectives: security 29 Demand for research 16 Distributional Demand curve criteria 471,479-480 cost-benefit analysis 54 effects of research 372-377 Subject Index 565 effects of research in food criticisms 20,43-47,49-52,54 production, distribution, and domestic resource cost approaches 56 marketing chain 82 externalities 50 effects of research related to farm Uarberger's postulates 40,41,75,269 size, income, and location 82 horizontal aggregation 210 weights 236, 372, 376, 377, 459, incomplete risk markets 51 460,466,474,475 measures, approximations of 468, 469 Diversification 86 measurement error 44 Divisia indexing procedures monopolistic industries 50 (see Index numbers) normativeness 43,44 Dollar-metric method 490 partial welfare analysis 50 Domestic resource cost models 56 policy irrelevance 52 Dominican Republic 470 second-best 50 Double counting welfare measures 232 transaction costs 50, 51 Dream approach 299, 304, 326, 362, vertical aggregation 210 370,386-394,506 vs income received 53 Drought 85,341 vs cost-reducing or -resistant crop variety 455 yield-enhancing effects 53 -tolerant crop variety 86 welfare measure 43,45-54 Duality approaches 97, 102, 111-113, Economies 146-150 of scale 81,102,142,322,338 methods for modeling general­ of scope 10,419 equilibrium trade and of size 10,322,419,504 welfare effects 234 of size in obtaining information 83 models of production 100 Ecuador 470,476,477 problems of simultaneity III Education 4, 14,23,83, 104, 106, theory 111 112,189 Durable factors of production 157 and labor 156 Dynamics 10,20,25,104,113,321, higher education 13 350,351,420,433-437 Effective quantity and price 114, 115 Efficiency E criteria 471 Exogenous shift variables 105, 106, 111, gross efficiency index 469 112,223,383,387,388 index 479,494,509 Econometric approaches to estimating modified index 489 research benefits objectives 80 55,193-206,504,506 Eggs 150,283 Economic (see also Chicken and Poultry) efficiency 28 Elasticity 35,57,262,263,469 profit 115 aggregate excess-demand profit, change in 48 curve 213 profit and producer surplus 48 aggregate (world) supply skills and intuition 473 and demand 224 way of thinking 18,464 Allen 265 Economic surplus 27, 41 compensated derived demands 108 alternatives 20, 54, 56 cross-price 74, 262 adjustment costs 52 excess-demand 322-324 apportioned to relevant groups 480 excess-supply 322-324 basic concepts 40-42, 43 factor demand 262, 265 congruence rule 57 factor substitution 108, 242 cost-benefit analysis 54, 55 factor supply 314 566 Subject Index imposing assumptions 27,54 bias in subjective data 349 local supply and demand 224 maintenance research 342-343 Marshallian 232 research program product transformation 242 components 341-342 output 108 research spillovers 343-348 price transmission 324 sample questionnaires 359,419-440 size vis-a-vis distribution of Employment 22,309 research benefits 59,208 regional 81 Elasticity of demand 20,42,49,66, Energy 192 210,230,313,320-321,469,502 Entomology 11,303 constant 222 Environmental cross-price 244,321 conditions 341 effects on research benefits 58, 59 costs of production 379 for priority-setting analyses 321 costs vis-a-vis sustainability 294 Hicksian vs Marshallian 47 deterioration 76 income 84, 85 externality 294 income effects 45,47 externality effects and natural inelastic 42 resource degradation 133 natural vs absolute values 238 issues 22,379,510 own- and cross-price 238 Environmentally friendly projects 34 own-price 87 Equilibrium relative to supply 484 -displacement models 223 total 232 factor and product markets 247 traded vs internally produced price change 228 and consumed goods 73 quantity change 224 variation across income classes 228 linear vs constant elasticity Elasticity of output and functions 256 stock -of-research 195 Equity objectives 80 Elasticity of substitution 71,73,101, income distribution 82 116,254,314 Equivalent variation 235 Allen vs Morishima 150 Estimates 102, 103 constant 108 efficiency and internal consistency 103 input 148, 149 Ex post evaluation 33,102,511 in vertical market models 253 research programs 504-506 substitution vis-a-vis factor Excess demand 66,214 demand vis-a-vis elasticities 322-324 elasticities of supply 73 ROW 216 Elasticity of supply 20,42,49,64,66, Excess supply 66,214 210,222,469,502,505 elasticities 322-324 constant 222 Excess-supply-excess-demand effects on research benefits 58,59 framework 213 for priority-setting analyses 321-322 Exchange rate 315,318 horizontal vs vertical shifts 322,361, distortions 76,208,268,269 411-418 market 318 length of run 321-322 misaligned 75 Elicitation forms 419-440,478 overvalued or Eliciting K undervalued 291-293,318 aggregation, substitution, and Executing agencies 4 complementary 342 Exogenous allocatable fixed factors 331 changes 326,383 benchmarking 342-343 prices 329,336,397 Subject Index 567 shift effects 380 owners 212 world price 383,392 price (see Price, factor) Exogenous growth returns, effects of research 72, 73 factors 324-326 scarcities 67 in demand and priority-setting share and biased technical change 142 analyses 325-326 substitution 101 in output 326, 380, 383 suppliers 262 in supply and demand 387-388, 395 total productivity 98, 130-132,200 in supply and priority-setting Factor-biased technical analyses 326 change 98,231 Expectations 104, 113 Hicksian vs scale effects 141-142 Experiment stations 172,476,485 vis-a-vis substitution effects 101 Experimental Factor-neutral technical change data 338-340,505 vis-a-vis economies of scale 10 1 yields 329,360,411-416,429,430, Factors of production 6, 234 432,433,468,477 durable 157 Export fixed, quasi-fixed, and variable48, 207 crops 86 gross complements 262 revenues 82 owners 17 Export taxes shared 231, 240 in small-country used in fixed proportions 71 trader models 277,279,280 Famine 365,481 in large-country Farm 321 trader models 286, 288 buildings 157 optimal 285 distribution effects of research on 82 Extension 13,19,25,27,55,97,108, labor 52, 506 114,148,174,175,188,189,356 machinery 9, 157 4-H activities 173 on-farm trials 339 costs 333 organic farming 76 expenditures 334, 336 part-time farming 156 home economics 173 prices 315 lag 26 profit 336 measurement 171, 173 size 6,83,211,228,471,496 production vs allocative small 458, 494, 507 efficiency 174 tenure 83,229 public vs private 174 yield gap 339 reseruch, and 18,173,178 Farmers 9,13,25,69,315,347 services 9,105 (see also Ranchers) systems 10 and adoption 339,351 urban issues 173 education 356 workers 356-358,467,482,505 large 87 Externalities 50,65,75,76,92,208, small 453, 480, 484 211,212,293,491,503 Feed grain 240 consumption 76 costs 231 environmental 294,295,297,510 Fertilizer 192,273,342,360,431,433 production 76,270 (see also Manure) consumption 346 F inorganic 164 Facilities for research 508,511,513 organic 164 Factor recommendations 177,341 markets 18,506 trials 338,329 568 Subject Index Financial inflow from abroad 235 Functional form 98,101,107,112,142, Financing agricultural research 4,7,82, 143,208,223,357 250,501 local vs global flexibility 101, 144 agricultural taxes 7 local linear approximation 64 commodity check-off schemes 7 of supply and demand 60-62, 64 cost-sharing arrangements 13 and supply shift 152 donors 7 Funding export taxes 7,13 level 468 general government revenues 13,14 long-term continuity in 488 import taxes 7 past practices 488 incidence of costs of a levy 263 Future generations 297 levies 13 output taxes 12 G private sources 7 Gambia, The 312,470,476 public sources 7 GAIT 319 Firms, entry and exit 64 General equilibrium Fixed factors 113,114 demand curve 234,239 proportions 251 effects 511,512 in vertical market models 246-252 feedback effects 65,244,299 Flexible functional forms 100,102 feedback and double Fourier 100, 109 counting 231,232 globally 100, 109 implications of research 78, 79 locally 100,108 model 79, 236 as locally quadratic forms 109 multiple sources of feedback 236 local vs global 101,144 supply curve 233 "semi-nonparametric" 100 welfare measures 239 Flooding 76 Generalized Leontief 109 Flour 243 Genetics 506 Food 15,85,321 gene manipulation and control 176 crops 5,317 improvement 334, 479 distribution effects of research material 359 on production, distribution, Geographical information and marketing 82 systems (GIS) 347 luxury 321 Germany 487 prices 78,85 Global warming 293,294 quality 85,482 Goal operator in mathematical residues 294 programming models 444 safety 293,473,482 Goal programming 445 security 15,80,81,455,473 Goats 157 self-sufficiency 22,80,82 (see also Ruminants) vulnerability of supply 481 Government 12,507 Forage 477 bureaucracy 4 (see also Pasture) changes in revenues 211,212,225, Foreign debt problems 87 234,235,269 Foreign exchange 15,56,87,309,473 cost 270 earnings 78,471 failure 82 Fourier 100 full cost of government funds 77 degrees of freedom 109 intervention 81,491 France 487 intervention with small- Free-riding 12 country assumption 227 Fruit 212,317,476 policies 92,266,318-319,365 Subject Index 569 policies and priority-setting Hotelling's lemma 241 studies 318-319 Household policies as tax or subsidy consumption of home-produced equi valents 318 goods 228,229 regulations 17 income 85 Grazing livestock 71 Human capital 7,24,27, 105, 106, 108, Green issues 293 112,114,156,167,170,172,176,192, Green revolution 442,504,508,511 technologies 178 farmers 105,106,147,156,165, varieties 342 178, 190 Greenhouse effect 76,293 Humid tropics 37, 231 Gross annual research benefit (see Research benefits) I Gross complements Ideal input-output formula 132 factors of production 262 Immiserizing technological change 286 Gross efficiency index 469 Imperfect information 51 as a proxy for gross annual Import research benefits 468 quotas for a small Group country in trade 279,281-283 decision-making process 492 subsidies for a small techniques 490 country in trade 277,279,280 Growth accounting 99, 191-193 tariff 285 and multiple-cropping effects 192 tariffs for a small Growth hormones 169,329 country in trade 279,281-283 Growth outside agriculture tariffs for a large effects on demand 245 country in trade 287 Guidelines for priority setting Inappropriability of returns to 472,487,488,491,509 invention 81 Income H decoupled income transfer 271 Harberger demand elasticity, income effect 47 postulates 40,41,75,208,269 distribution 15,81,311,317,365, triangles 91 375,502 Hausman test 111 economic surplus vs Harvesting income received 53 equipment 157 elasticities 320,321,388 inputs 100 elasticity of Health 14 demand 45,48,84,85,319,320 care 106 equity objectives and income Hectares, number of 471 distribution 82 Herbicides 164,297 functional distribution of 82, 230 Hicks-neutral technical change groups 230,320 (see Technical change) growth 325, 350, 387, 388 Hogs 240,253 household income 85 (see also Pork, Swine, and Livestock) increased total 502 Home-produced goods levels 229 household consumption of 229 per capita 85,317,321,325,482 Horizontal market relationship 507 personal vs functional 81 for internationally traded good 212 population and other demand Horses 157 shifters 245 Horticultural products 476 risk 455,456 570 Subject Index risk to producers vs society 456 Informal procedures 487 stability 37 Information transfer 374 costs of transfer 25 variability 15,86,89,304,365,376 imperfect 51 481,502,503 Infrastructure 104,112,114,163,166 wage rates 156 communications 23 Incomplete markets 81 education 23 Index irrigation 23, 164, 171 efficiency index 479, 494, 509 public investment 104 gross efficiency index 469 roads 23, 346 international land-quality index 163 Input purchasing-power-parity indexes 318 -augmenting technical change (see Index numbers 120-142 also Technical change) 101, 115 aggregating inputs and bias (see also Technical change) 72 outputs 121-125 -constrained output-supply axiomatic approach 121 function 241 biased productivity index 123 conventional 106, 108, 112, 143, 164 characteristicity 121, 128 conventional and unconventional 99 cost-minimizing producers 121 -demand functions 25, 103 Divisia 125-128 -demand system 104 Divisia, Fisher ideal distribution of 247 approximation of 126-128 harvesting 100 Divisia, invariance property 125 infrastructural 27 Divisia, Laspeyres "inside" 165 approximation of 126, 128 nonfarm 71 Divisia, Paasche "outside" 157 approximation of 126, 128 pest control 100 Divisia, Tornqvist-Theil prices 98,105,112,147,156-167 approximation of 126, 128 purchased 69 growth in agricultural output 98, 120 quality and double counting 106 growth in agricultural quality change 100,106,112 productivity 98, 120 quality of conventional 23, 105 linear aggregation inputs 123 quantities 156-167 measured productivity 123 saved 54 fixed vs chained weights 128 -saving technical advance 115 problems, definition of 121 subsidies 273, 277 shift in isoquants 123 substitution 71,263 shift in production-possibility substitution, implications for frontier 123 substitution possibilities 251 simple accounting devices 98 the distribution of research superlative 127 benefits 263 total vs multi factor productivity 98 supplied by farmers 69 India 3 suppliers 10,246 Indifference curves 89,468 -supplying sectors 235 Indonesia 148,176 -using sectors 235 Induced-innovation hypothesis 106 Input-output models 79 Industry Insects 85,358 data 340 (see also Pest) expected growth 491 Institutional change 511 size of 489 Institutionalization 18,349,419,512 Inferior goods 47,84 Instrumental variables approach 112, 190 Subject Index 571 Integrability conditions 232 perrect 103 Integrating supply-and production 19,173 -demand equations 237 production function 20, 24, 25, 57, Intensity of land use 297,332 105,106,326,349,441,450, Interactive programming 452,486 procedures 449,450 productivity 23 Interest groups 88,361,364,393 rate of decay of stock 176 Interest rate 158,324,363,491 service flow 24 (see also Discount rate) stock 23,24,25,29,31,57,98, Intergenerational 104,105,143,148,167,175, equity 502 195,488 equity and agricultural stock of knowledge research 297,298 and functionalform 175, 177 transfers 34 stock of knowledge Intermediate inputs and research expenditures 167,175 net revenue functions 235 truncated research lag 107 Internal rate of return (IRR) 32, 54, 78, utilization 23, 105, 176, 186 110,198,363-364,469,507 vis-a-vis omitted-variable (see also Rates of return) problem 107 IRR method and ranking 33 IRR method and scale of L investment 33 Labor 6,128,148,156-157,207,471 Interpretive structural modeling 490 (see also Human capital) Invisible hand 512 capital-labor ratios 372 Iowa State 470 demand 85 Irrigation 23,39, 104, 112,338,459 education 156 infrastructure 164 factor of production 52,431,506 family 156 J farm 52,506 Japan 487 hired 156 Jute 111 intensity 73 number of people employed 471 K part-time farming 156 K-factor 327,397,505 quality 156,338 (see also Supply shift) use 78 Kaldor-Hicks criterion 42 wage rates 157 Kenya 470 Lags (see Adoption lags and Knives 246 Research lags) Knowledge Lamb 244 agricultural productivity 21 (see also Sheep and Livestock) capital stock 23 Land 6,129,163-164,229,378 constant-elasticity functional common-property issues form 177 and measuring 164 current increments to 105 degradation 22, 133 depreciation 24, 106, 107 "fixed" factor 330, 331,360,431 finite lags of research and intensity of land use 297, 332 extension 107 intercropping systems 163 in the agricultural production international land-quality index 163 function 23,106 multicropping 163 increments to useful 25 number of hectares 471 infinite lag structure 105 opportunity cost 332 572 Subject Index property rights 163 spending 77 quality 163,338 Marginal-social-cost curves quasi-fixed factor 207 vis-a-vis externalities 296 rental rates 163 Marginal utility of income 43 -saving technical Market change 414-415,416 failure 11,12,14,52,80,88,491 supply 322 inputs, increase in sustainability 22 supply of 247,249,250 tenure 356 multiple-market model, -using technical change 415,416 horizontal disaggregation 65 Land-grant colleges 470 multiple-market model, vertical Landlords 83 disaggregation 65 Latin America 5,505 power in trade 285 Le Chlitelier principle 328 power, optimal exploitation of 285 Length of run vis-a-vis supply curve 207 power vis-a-vis research benefit 299 Linear programming shares 224, 311 models 372 structure 312 problem 442 Market-clearing conditions Livestock 124,129,152,153,154,164, in a two-market model 214 165,168,187,326,336 Market distortions 65,75,92,207, (see also Beef, Hogs, etc.) 237,502 grazing 71 Marketing 69 industries 240,322 chain 69-72,82,491,507 management practices 17 inputs in efficiency units 251 weight 317 margins 318 Logistic curve, Pearl-Reed method 358 research 69,70 Logarithmic differential Markets approximation 222 geographically spread 503 Low-income incomplete 81 consumers 83,480 vertically related 503 country 321 Marshallian curves (see Demand curves families 89 and Supply curves) groups 372,502 Maximum (ceiling) prices (see Prices) per capita 85 Mathematical-programming producers 474 methods 463, 482, 486, 508, 510 Lump-sum transfer 42,44,90 non-negativity constraint 444 Mean-squared error criterion 47 M Meat 71,154,321 Macroeconomic adjustments 38 (see also Beef, Lamb, etc.) Maintenance research consumer preference for (see Research) lean meat 243 Maize 28 Mechanical rules 493 Malnutrition 85,365 Micromanagement of scientists (see also Nutrition) xix, xx, 509 Malsters and brewers 243 Middle East 37 Manure 154, 165 Military (see also Fertilizer) vulnerability to conflict 481 Marginal analysis security 80,87 of program changes 486 wars 38 Marginal research benefit 442 Milk 71,231,283,317 Marginal social cost of government (see also Dairy) Subject Index 573 Minimum target prices 269 n-factor cases 264-266 with deficiency payment in a closed- notation used in 255 economy model 270, 271 proportional, parallel, or pivotal Ministry shift assumptions 255 of Agriculture 4,5,7,90 solutions 255,257-261 of Finance 90 three-factor cases 264 Model specification 109 Modeling, degree of disaggregation 66 N Modified efficiency index 489 National economic pie Money-metric measure 53,234 size and scope of 92 Monocropping systems 176,297 National self-reliance 502 Monopoly Natural endowments of resources 221 exploitation of foreign markets 289 Natural predators 297 power in international markets 285 Natural resources Monopsony 299 conservation 6,502,510 power in international markets 285 excessive exploitation 297 Monte Carlo simulation 366 flow of services from vs stock of 297 MOT AD approach 453,455 management 506 Mules 157 management and conservation 76 Multicollinearity 100, 108, 109, rate of exploitation 33 188-189,504 Net efficiency index 469,483,487,489 Multigroup model 228 for a commodity research program 485 Multilocational trials 338 Net present value (NPV) 32, 54, 110, Multiple 325,362-364,369-371,377,383, criteria 463 441-489,502,507 cropping 231 economic surplus accruing to displacements 234 particular groups 458 factors 207 expected 37 Multiple markets 207,225,234 per dollar of research spending 452 models, horizontal per unit of constraint 307, 442, 508 disaggregation 65,212-230 ranking 32,369-371 models, vertical scale of investment 33 disaggregation 65, 246-266 size of research programs 442 related in consumption 237-239 total domestic economic surplus 458 related in production 240 vis-a-vis congruence 489- 490 situations 230 Net transfer from abroad 234 Multiple objectives 16,87-92,369, Netherlands, The 487 372-377,442,458,459,464,465 Nominal group technique 490 (see also Priority setting) Nonparametric approach 98, 101, dealing with 474 116-120 programming model 443 advantages of 119,120 reconciliation of 80 biased and neutral technical social 509 changes 120 trading off 87-91 functional forms 117 Multistage production system lack of invariance to scaling 120 208,225,246 "power" of 119 Muth model serious flaw 119 and fixed proportions 262 Normal goods 47,84 description of 253,255 Normative economics 22,43 incidence of benefits from North Carolina State 470 technical change 263 Numerical solutions 441 574 Subject Index Nutrition 309,473 Output (see also Malnutrition) aggregation issues 154 animal nutrition 307,461,476,479 -augmenting technical benefits 471 change 101, 114 human 85 controls, as a complement to implications of research 85 a subsidy scheme 283 value of commodities 85 controls, in a closed-economy o model 274, 275 controls, in small-country Objective function 441 trader model 283 in mathematical programming data 153-155 models 444 effective vs observed 115 multiplicative 489 price ceilings in small-country Objectives 15,19,310,457,458 trader model 277,279,280 (see also Multiple objectives, price supports in large-country Research objectives, and Priority trader model 285, 286 setting) prices 98, 194, 197 analysis, for 303 prices and aggregation issues 155 criteria related to quality of data 154 distribution 479,480 quantities and prices 153-155 criteria related to security 480 sources of growth 99 distinguishing weights from subsidies in small-country trader measures 467,468 model 276, 278 distributional 15,16,21,29,453, -supply equations 104 458,465,482,483 Output-constrained input-demand economic efficiency 16,21,369-372 functions 241 453,458,465,502 eliciting initial weights 474 p equity 21,502 Paired-comparison method 490 growth 16 Parameters, restrictions 103 means and measures of achieving 309 Parametric approach 97,98, 102-104 meaningful 467,473 Parametric variations national 37 constraints on the solution 444,445 regional distributional 483 weights on the objective research system 303 function 444,446 security 16,29,82,86,87,89,310, Pareto 459,465,502,503 criterion 42 specifying 473 optimal 52 stated vs measurable 473 Parity model "value-adding" 72 (see Congruence) OECn 80,237 Partial factor productivity Opportunism 51 changes in factor mix 130 Opportunity cost 82,374,474,508 factor bias 137 government funds 77-78,333,491 quality changes 130 land 332 Partial equilibrium 511 resources used in production analysis 50,231 and consumption 227 approach 236 ORANI model 80 framework 20,512 Ordinal utility function measures 233 cardinal representation 44 model 28,40,78,80 Organic farming 76 Pasture 71 Subject Index 575 (see also Forage) environment 38,510 Patents 12,26,51,69 factors 511 Patterns of trade 81 trade-offs 88 Peer review 464, 487, 490, 509 trouble 481 as a complement to formal visibility 471 economic approaches 492 Pollution 15,51,76 Peer-group discussion 479 (see also Externalities) Per capita income 317 air 22 demand-shifting chemical 133 effects 245, 325, 326, 387, 388 costs 22 income elasticities 321 pesticide 76 Per unit cost reductions (see Supply shift) water 22,294 Perennial crops 152, 153, 163, 164, Population 177,322 demand-shifting Perfect knowledge 103 effects 245,325,326,387,388 Peru 470 pressures 15 Pests 104,112 Porcine somatotropin 243 and diseases 231, 340, 341, 357 Pork 322 control 17, 100, 334 (see also Hogs, Swine, and Livestock) crop vulnerability to 481 Portfolio risk 453 infestation 455 Positive economics 21,43 management 294,455,506 Postharvest 69 resistance 341 handling and storage technology 72 Pesticides 31,164,192,273,297,358, research 72 360,431 sector 69 pollution 76 Poultry 152,322 Philippines 5,87 (see also Chicken and Livestock) Plants Poverty 364 , (see also Crops and specific plants) countries 14 breeding 8, 11,307,341,348,351, incidence 489 372,433,457,461,476,484 people 82 material 9 urban 458 pathology 5, 11 Precedence 464,488,489,493 protection 307,476 Precision 101 Policy Preference 236 agricultural 469,505 changes in 511 Cobb-Douglas preference function 43 decision makers' personal preference distortions and research benefits, vs professional judgment 459 a more general approach 289 decision makers' preference in domestic 213 mathematical programming government policy effects models 444 on rates of return 150 strength of 490 international trade 213 Price 15,27,66,106,113,115,134, makers 9,512 150,157,168,317-318,491 market-distorting 20,78,221,299 (see also Input and Output) price-distorting 211,212 actual vs effective 101 reform 18 aggregate 125 trade-distorting 208 ceilings (maximum prices) 269 variables 166--167 ceilings in a closed-economy Political model 271 economy model 16,75,88 domestic 86 576 Subject Index dynamic price elasticities 285 objectives 372-377 elasticity, constant unitary 73 measurement issues 374 elasticity of demand 491 nonresearch policies 372 elasticity of demand optimization algorithms 372 for home consumption 229 ranking programs 369-372 elasticity of supply 491 reality check 376 expected 114 reallocation among programs 369-372 factor 24, 78, 106, Ill, 112, 122, shutting down programs 369-372 123, 125, 130, 135, 138, 147, 150 Priority-setting analyses -fixing schemes 269 clients 305-306 for priority-setting studies 317-318 defining the problem 305-314 future 152 degree of detail 312-314 indexes for R&D 173 general-equilibrium models 314 international 75 market-related data 314-326 paid 28,85 measures of benefits 309-311 policy 17, 88, 266, 268 objectives 306-307 reduction due to research 210 research programs vs program (see also Supply shift) alternatives 307-308 received by producers 28 scope 307-308 relative to change in equilibrium 223 Pristine wilderness 22 risk 455 Private spillovers vis-a-vis externalities 294 industry 470 stabilization schemes 39 research 466 substitution effects 101 sector 9,12,25,69,81,340,491 supports in a closed- support 7 economy model 270, 271 Probabilities of research success 22, 35 supports (minimum prices) in a small- 467-469,477,485,490,491,502 country trader model 276-278 (see also Research risk) target 269 Processing 12, 69 target prices with deficiency industry 71 payment in a closed-economy inputs 247 model 270, 271 levels 510 wedge due to policies or research 69,70 transportation costs 222, 225, 226 technology 224 world 86, 102, 111, 227 technology, changes in 250,252 Primalapproach 97,98,99,102, Processors 10,69,507 104-111 Producers 9,12,17,42,69,507 rate of technological change 134 benefits, domestic 214 vis-a-vis research benefits 110 (see also Research benefits) Priorities -funded research 466 ex ante assessment 68 gross income, variance- regional vs national 68 covariance matrix 456 short and long run 442 groups 83 Priority setting 4, 17, 19 number of 471 achieving a balance 377 subsidy equivalent 319 conceptualizing trade-offs 372-374 surplus 40,48,207,315,331, discrete vs continuous 361,385 alternatives 373, 375 surplus and externalities 51 guidelines 472,487,488,491, 509 surplus and trade 66 maximizing efficiency 369-372 surplus and value judgements 43 maximizing with multiple surplus disaggregation 228-231 Subject Index 577 Production and the value of production 489 domestic share of global Protein 471 production 491 Proxies 107, 142, 187 economics vis-a-vis evaluating time index 100,101 effects of agricultural R&D 99 variables 166-167 productivity 68 Public goods 13,51 productivity, change in 23 Public sector 12, 18 risk 455,456 crowding out 14 risk and food security 37 Publication 501 shared factors of 231, 240 citation-adjusted 26 scaling by output or farm numbers 107 Purchasing-power-parity indexes 318 variability 481 Production function 97,102,105,110, Q 133-135,241,504,505 Q-sort method 490 Cobb-Douglas 143-144 Quadratic flexible forms 144 approximation to cost function 148 parallel vs aggregate function 128 nonparallel supply shift 110 forms 109 research in 26 production function 109 state of technology indexes 133 profit function 115 translog 109,145,146,199 programming model 453 vis-a-vis economic surplus Quality approach 109 change 243-245 Productivity 53, 55, 473, 504 change effects of research 74 (see also Partial and Total/actor demand vs supply shift productivity) representation 74 environmental and natural product 503,510 resources 133 vs discrete variation 244 functions 97, 189 Quantitative gains 490 genetics 35 input quality 132 restrictions on inputs, measurement 129-132,133 outputs, or trade 269 new technology 133 skill 464 output quality 132 Quantities 66 research and extension 133 actual vs effective 101 partial factor 129, 130 consumed and priority-setting sources of growth 132, 133 studies 316-317 Profit 146, 178, 185, 191, 194 demanded share-weighted sum Marshallian factor-demand of relative change 222 equations 111 produced and priority-setting maximization 103,117-118,146, studies 316 147, 189 supplied share-weighted sum output-supply functions III of relative change 222 Profit function 98, 10 I, 103, 104, Ill, traded and priority-setting 115,120,146,147, 155, 189, 191, studies 316-317 194,504,505 Quasi-fixed factors 112,115 Profitability 23 Quasi-rents 207,250 Program alternatives 465 farming inputs 210 Project evaluation 312, 490 off-farm processing and Property rights 12, 14 marketing inputs 210 Proportionality rules between research Queueing 272 578 Subject Index Quota applied 4,8,26,31,67,503 owners 283 as a blunt instrument (see Blunt rents 275,282,393 instrument) basic 4,8,12,18,31, R 67,463,487,503 Rainfall 37,356 basic vs applied 26 Rainforests, clearing of 294 before and after 32,420-421 Ranchers 13 boards 513 (see also Fanners) budget 442 Ranking 65, 364, 470, 492 capital 24, 488 alternatives for allocating causality 189-191 research resources 371 commodity 5, 6,475,476 across research programs 32, 307, complementarities with education 369-371,463,469,492 and extension 14 cardinal 486,487,490 constraints 441 circular process 475 cost 22,32,35,333-334,469, commodities 478 477,478,485,486,491,503 commodity research councils 513 programs 481-483 cross-country transferability 344 one-dimensional methods 490 depreciation 358-359,468,491 ordinal 371,469,483,484,487, diminishing returns 34 490,507,508 directors 479,509,513 ordinal vs cardinal 473 disciplines 5,6,476,501 programs, according to scores 466 distributional impact 364-365 project proposals 492 domestic vs foreign 26 projects 32,364 dynamics 8 research program economic evaluation vs components 484,485 explanation 172 rough cardinal 469 economies of scope 13,14,25 using aggregated scores 466 economies of size 13,14,25 vs congruence 490 effects, two-step estimation vs funding and staffing procedure 108 decisions 486 environmental 16 Rapeseed 150 evaluation xxxi, 4, 17, 19 Rate of technical change 133-136 extension (see Extension) dual 135 factor mix 501 dual vs primal 136 functional distribution of benefits 364 production-function approaches 134 institutional setting 4 Rates of return 14 institutional variables 24 (see also Internal rate of return) intensity ratios 489 average vs marginal 34 investments, diminishing returns 488 government policy effects 151 maintenance research II, 31, 32, 176 Rational expectation formulations 152 management 7 RE4 359 mean-variance trade-off 365 Regional employment 81 molecular biology 8 Religion 356 multidisciplinary 6 Rent 207,229 net present value (see Net present Research 4 value) adaptive 8,177 noncommodity 475,476,479,506 administrators 16,86,87,90,482 on-farm 9 and extension 174 on-station 9 Subject Index 579 personal distribution of benefits 365 maximizing total research benefits 450 policy, joint determination of 17 open economy 212-228 portfolios 15,474 policy distortions, a more pretechnology 18, 26, 29 general approach 289 pretechnology vs applied vs relative to costs 494 development 105 retail level 210 private vs public 26 sensitivity analysis 369 problem orientation 501 size of 114,469 production function (see Knowledge small-country-in-trade model 226-228 production function) supply functions 204-206 projects 443, 457, 490 vis-a-vis market-distorting public vs private vs foreign 173 policies 266 public vs private vs foreign with input substitution 256,262,264 investment 105 Research benefit streams 193-206 regional focus 501 cost functions 203-204 resource scarcity 21 production functions 193, 194-199 resources, short-run fixities 443 productivity functions 199-202 resources, suboptimal allocation 488 simulating alternative response function 442 scenarios 198-199 response function with diminishing supply functions 204-206 marginal returns 452 Research expenditures 172,505 socioeconomic 463 (see also Knowledge stock) temporal aggregation 175-185 ad hoc lag structures 167 with and without 32,342,343, core funded 168, 169 420,421 donor funded 170 Research benefits 32 government funds 169 algebra for the closed-economy lagged 27 case 211 measurement 167-173 algebra for the open-economy vis-a-vis quality of research 172 case 216,217 vs service flows 172 among consumer groups 230 Research-induced supply shift among producer groups 229,230 (see Supply shift) annual flow of 31 Research lags 26,29,99,168,177 between producers and consumers 29 349-358,366 closed economy 208-210 binomial distribution 183 critical assumptions when modeling 58 development 29,99 determinants of size and finite infinite 179, 181 distribution 57-77 form-free structures 184,185 distribution 469 Hatanaka and Wallace method 184 distribution between producers inappropriate priors 179 and consumers 114 institutional environment 177 distribution, international 72 inverted-V or Deleeuw environmental externalities (see structure 183 Externalities) length 202 exchange rates 291-293 logistic structure 357,358 farm level 210 more flexible structures 182-185 functional distribution 71 multicollinearity problems 179 gross annual 29,201,202 normalized temporal weights 197 gross or net 468-469,507 Pascal distribution 183 Hicksian measures 80 polynomial or Almon large vs small farms 83 structure 181,182,183 580 Subject Index smoothness priors 183 in a domestic setting 348-349 symmetric vs nonsymmetric in an international setting 343-348 inverted V 181, 182 in small country temporal weights 178 model 227 trapezoidal structure 181,183,357 matrix 224 truncation 185 n-market model 221-225 typical, finite structures 178-182 present in a two- vs extension lags 174 market model 219,220 with endpoint constraints 182 price 6,67,68,212,213 Research objectives (see Objectives and rules of thumb 346 Multiple objectives) spatial aggregation 185-188 Research priorities 6,369-379,450- spatial weights 186 456,482-487 spill-in 326 national 6 technology 6,67,68,212,213 noncommodity program 485 technology vs price 313 regional 6 technological proximity 187, 188 sensitivity analysis 443 vis-a-vis externalities 294 short vs long term 39,461 Researchable problems Research programs 476 number and severity 478 all-or-nothing 308 Resource depletion (see Natural alternatives, marginal effects 507 resources) areas 507 Response functions 97,338 closing down 34,369,370,486 Retaliatory policies 285 evaluation 512 Returns (see also Internal rate o/return) priority setting 512 inappropriability of returns to ranking 369-371 investment 81 review 512 negative 286 size of 489 to R&D 102 Research risk 13-15,34,349-355, to research 40, 54, 113 365-369,458,473 to research and distortions analyst's uncertainty 35 in commodity markets 286 diversification strategies 37 to scale 108,109 expected vs most likely outcome 37 Revealed preference 88 mean-variance trade-offs 36 axioms 119 pooling 36 Revenue production risk 37 function 241 program size 36 industry 28 public vs privately borne 36 -pooling price-discrimination triangular distribution 366 schemes in small-country Research spillovers 6,7,65,67,92, 168, trader models 283, 284 224,313,326,343-349,432- Rewards 509 433,485 Rice 13,148,154,170,176,186,192, absent in a two-market 322,339,457,486,507 model 213-215,217,218 dry-stalk paddy 154 agroecological zones 186,347 high-yielding varieties 82 cross-commodity 74 irrigation 37,83, 163,476 double counting 345 milled 154 effects in priority-setting quality 243 implications for resource swamp 476 allocations to research 68 upland 83, 476 analysis 343-348 Rich countries 14 Subject Index 581 Risk 83,85,450,481,510 Self-sufficiency 87,473,481 agriculture 365 Seed 132, 164 aversion, degree of 510 certification and distribution income 80,82,86,455-456,473 systems 178 famine 456 Seemingly unrelated regression multi-variate 455 procedure vs maximum of ruin 453 likelihood methods 149 pooling 38 Selecting projects or experiments 508 production vs price 456 Self-reliance 473 -reducing research 39 Semiarid areas 38 reduction 38 Sensitivity analysis 369 research 81,473 Separability research portfolio 461 assumptions among inputs 246 Roads 23,346 between inputs and outputs 241 ROW producers assumptions vis-a-vis general- as net losers 219 equilibrium models 234 Rubber 163 assumptions vis-a-vis mathematical Rules of thumb 479,487,488, programming models 443 490,512 Serendipity 509 Ruminants, small 168,476 Shared factors of production 231 (see also Goats) Sheep 151,154,157,231 Rural infrastructure 14,346 (see also Lamb and Wool) Rural Urban North South nutrition 457 (RUNS) model 80,237 Shephard's lemma 113,135,149 Short- and long-run plans 461 S Simultaneity 189-191 Salt 294 between inputs and outputs 100 Scientific Single-equation supply-response context 8 model 104 disciplines 24 Small countries 80 entrepreneurship 509 Social merit 492 accounting matrix (SAM) 79 Scientists 35,452,467,477-479, policy 14 482,505,509,512 science 303,461,476 time 501 research 11,17,18,511 Scores welfare function 88 ordinal vs cardinal 472 Socioeconomics 478 vis-a-vis weights 472 Soil 169,187,356 Scoring 464 acidification 297 as a measure 466 decline in fertility 133 as a weight 466 degradation 76 common practice 465-467 erosion 15,76,294,297 implementation 482,483 fertility 133,231 methods 508 management practices 170 principles 467 salinization 297 Second-best 52 science 5,461,478,484,506 problem, vis-a-vis distorted structure 231 exchange rates 293 type 163 world 270 Sorghum 150, 154 Secrecy 12 Southeast Asia 344 Security objectives 82 Soybeans Ill, 148, 150 582 Subject Index Species general-equilibrium 233 depletion 294 general-equilibrium vs partial- preservation 293 equilibrium 210,212 Specific (per unit) tariffs Marshallian 236 fixed or variable 279 vertical 54 Specification error 186, 187 Supply elasticity (see Elasticity of supply) Spillage 275 Supply function 27,388,401,505 Spillovers (see Research spillovers) choice of a functional form 35,49, Spreadsheet 362, 368, 380-385 101-103,110,146,150,151,189, Stability 191,193,194,204-206 income 481 commodity 327 price 481 constant-elasticity form with production 481 positive intercept 62 yield 481 constant-elasticity model 61 Stabilization programs directly estimated 103, 113-114, output vs prices vs income 38 331,334-335 Standard gamble 490 domestic 323 Staple food crops 82,84,85 dynamics 114,153 State oftechnology 112,114 estimating 102-103 Stock of knowledge expected prices 114 (see Knowledge stock) expected vs actual prices 152 Strategic interaction among flexibility 114 governments 285 functional forms 64 Subjective judgments 508 industry 327,329,330,413,415 Subsidies 75,269,391-392,510 K-shift 337 ad valorem 221 kinked 60,61 cost 276 linear 42,54,61 on inputs in a closed-economy linear vs constant-elasticity forms 61 market 273 linear vs nonlinear 49 on output in a closed-economy negative intercept on the price axis 60 market 273 single-equation supply producers 42 models 150-153 Substitutability among inputs 246 Supply shift 25,326-361,388-393, Substitution 395,429,491,505,511 effects 244,485,511 assumption 61 elasticity of 101,108,116,143-145, conceptual issues 328-332 148, 150, 191 cost and profit functions 336 in consumption 231,233,240,244 directly estimated supply in production 73,233 functions 333,335 Substitutability 70, 73 disaggregation 66 Successive comparisons 490 econometric measures of 335-338 Sugarcane 86 endogenous 232 Sunk costs 51,307 eliciting K from Supply scientists 340-349 controls in large-country experimental data 338-340 trader models 289,290 horizontal vs vertical (K vs J elasticity (see Elasticity of supply) estimates) 61,65,322,338,348, response 102, 103, 114, 152, 155 361,397,411-418,505 response in agricultural R&D 97 individual firms vs industry 64 Supply curve 506 industry 327,336,339,340 disaggregating 83 measuring 326-361 Subject Index 583 nature 20, 49, 63, 64, 208, 502 benefits 386 parallel 42,54,58,61,386 surplus 269 pivotal vs parallel 49, 151,209 Technical Advisory Committee practical measurement 332-349 (TAC) 471 production functions 335-336 Technical change 27 productivity functions 336-338 biased vs neutral components proportional (or pivotal) 61,383 in Muth models 254 proportional, parallel, or pivotal 223 disembodied 18, 503 proxy 338 embodied 18 ROW 219 factor bias 117,137-142 size 20,502 factor bias and homothetic unavoidable assumptions 50 technology 137-139 vis-a-vis constant elasticity 61 factor bias and nonhomothetic Supply-and-demand curve technology 139-141 linear vs nonlinear 49 Hicks-neutral 118, 135, 138, 139 Surplus transformation curve 373 immiserizing 138 Sustainability 15,76,325,502,510 modeling options 114-116 agriculture 471 output-neutral and output-biased 137 and interest rates 33 pairwise and overall factor as a research objective 309 bias 142 criteria 294 scale effects 137-142 issues 293 Solow's representation 134 vis-a-vis price-distorting policies 293 time-index proxy 100,101 Sustaining environmental quality 455 Technological regression 138 Sweden 487 Technology Swine 151 environmentally sensitive 67 (see also Hogs and Pork) index 98,101,104, lIS, 116 index over time vs over space 101 T obsolescence 31 Tanzania 470,476 searching for 25 Tariffs 75 transfer 9 ad valorem 398 transferability 67 import 391,398,401 types 507 modeling as excess demand or Temperature 356 excess supply shift 289 Tenants 83, 230 optimal 285 Tenure situations 83, 356 revenues 235 Transportation inputs 247 Tastes, changes in 511 Three postulates 208 Tax 65,77-78,91,391-392,510 Time required to complete ad valorem 221,318 research 327, 355, 420, 433 export 391 Tobacco 244,322 on commodities 319,386 Tomato harvester 243, 299 on consumers 42 Topography 356 on production, consumption, or Total elasticity 232 factor returns 235, 391 Total factor productivity 130, 131,200 on production, input, or trade 269 (see also Productivity) trade 234 aggregation bias 131 policy 317 index numbers 121 -subsidy schemes 44, 362 nonparametric methods 118 wedges 225 Tornqvist-Theil index 131 Taxpayer 211,212,268 rate of change 131 584 Subject Index Tractors 131 Variability of income (see Income, Trade variability) international and interregional 65 Variable factors 115 patterns 81,469,502 proportions in vertical market status 17 models 251,253,255 Transaction costs 14, 16,51-52 Variable import levies 269 Translog 100,109,145,146,179 Varieties 17,175,343,352 aggregator function 128-129 (see also Com, Rice, etc.) approximation potential 131 development 177 cost function 112, 148,203,204 green revolution 342 model 131 new 167,178,183,338,342,343, production function 109, 145, 346,355,357,421,433, 501 146, 199 resistant 31,86,341,342,455 production function and trials 329, 338 growth accounting 193 Vegetables 212,476 Transportation 106, 112, 356 Vehicles 157 costs 221,222,226,317 Venezuela 470,477 Trees 322 Triangular w distribution 354,366-369,439 Wages 156 and research risk 353,366,369 Wars 38 (see also Military) u Water 6,196 Unbiasedness 102 pollution 133 Uncertainty 20, 104 storage 164 Uncontrolled factors 104 -use efficiency 296 Underinvestment in research 81 Weak axiom of cost minimization Unit cost, reduction in (see Supply shift) (WACM) 98,116,118-120 United Kingdom (U.K.) 443,487 and nonparametric approaches 117 United States (U.S.) 3,72,77, Ill, 150, Weak axiom of profit maximization 156,157,158,185,241,253,270,322 (WAPM) 98,117-118 agricultural experiment stations 172 and nonparametric approaches 117 Department of Agriculture Weather 100,104,108, 112, 190, (USDA) 470 199,316,338,340,365,429 Universities 4,5,470 crop vulnerability to 481 Urban poor 365,458 index 166 Uruguay 13,470 variables 152 Utility 235 Weighted -average method 490 v score 496 welfare function 268 Value (see also Net present value) Weighting 492,493,503 -added 489 objectives 16,457,459,460 judgments 43,53 empirical approach 459 of consumption 317 Weights 373,376,467,508,510 of experimental yield gain 352 distributional 236 of human life 22 effective 496 of production 307,311,477,481, effective vs actual 494 491,502 elicited 470,484 of public resources 324 eliciting 474 of technology 359 empirical determination 475 Subject Index 585 normalizing 494,496 165,186,243,244,341,346,355,486 on objectives 450 rust 176 on objectives vs criteria 486 Wilderness areas, preservation 293 the problem of units 494-496 Willingness to pay 222, 255 unequal welfare weights 44 methods 22 vs measures 467,468 Wool 111,154,264,364 welfare weights 42,88 (see also Sheep) Welfare analysis 317,318,329 Australian Wool Welfare change Corporation 443, 457, 475 correct measures of 233 Australian Wool Research exact Hicksian measures 47 Council 366 Hicksian money-metric industry 226 measures 53,234 research 475 incidence of 239 World Bank 80,237 total effect 238 World price (see Price) Welfare economics and value judgments 43 y Welfare effects 393-394 Yield 32,35,72,477,503 Welfare measures change 479,485 deadweight loss 47 commercial 360 double counting 232 equation 107 precision vs bias 44,45 expected changes 469 surplus in the Muth experimental (see Experimental model 256,262-264 yields) Welfare weights 42,88 experimental vs commercial 339 unequal 44 gap 339,471 Well-being 21,309 improvements 505 economic and physical 473 -response function 107 West Africa 344,470 risk 455 Wheat 35, 129, 150, 153, 154, 163, variability 365,481 Food Systems and Agrarian Change Edited by Frederick H. Butte!, Billie R. DeWalt, and Per Pinstrup-Andersen Science under Scarcity: Principles and Practice for Agricultural Research Evaluation and Priority Setting by Julian M. Alston, George W. Norton, and Philip G. Pardey Hungry Dreams: The Failure of Food Policy in Revolutionary Nicaragua, 1979-1990 by Brizio N. Biondi-Morra Research and Productivity in Asian Agriculture by Robert E. Evenson and Carl E. Pray The Politics of Food in Mexico: State Power and Social Mobilization by Jonathan Fox Searching for Rural Development: Labor Migration and Employment in Mexico by Merilee S. Grindle The Global Restructuring of Agro-Food Systems edited by Philip McMichael Structural Change and Small-Farm Agriculture in Northwest Portugal by Eric Monke et al. Diversity, Farmer Knowledge, and Sustainability edited by Joyce Lewinger Moock and Robert E. Rhoades Local Knowledge and Agricultural Decision Making in the Philippines: Class, Gender, and Resistance by Virginia Nazarea-Sandoval Child Growth and Nutrition in Developing Countries: Priorities for Action edited by Per Pinstrup-Andersen, David Pelletier, and Harold Alderman Networking in International Agricultural Research by Donald L. Plucknett, Nigel J. H. Smith, and Selcuk Ozgediz Adjusting to Policy Failure in African Economies edited by David E. Sahn The New Economics of India's Green Revolution: Income and Employment Diffusion in Uttar Pradesh by Rita Sharma and Thomas T. Poleman Agriculture and the State: Growth, Employment, and Poverty in Developing Countries edited by C. Peter Timmer The Greening of Agricultural Policy in Industrial Societies: Swedish Reforms in Comparative Perspective by David Vail, Knut Per Hasund, and Lars Drake Transforming Agriculture in Taiwan: The Experience of the Joint Commission on Rural Reconstruction by Joseph A. Yager