IFPRI Discussion Paper 02260 June 2024 Two Decades After Maputo, What’s in the CAADP Ten Percent? Determinants and Effects of the Composition of Government Agriculture Expenditure in Africa Samuel Benin Development Strategies and Governance Unit INTERNATIONAL FOOD POLICY RESEARCH INSTITUTE The International Food Policy Research Institute (IFPRI), established in 1975, provides evidence-based policy solutions to sustainably end hunger and malnutrition and reduce poverty. The Institute conducts research, communicates results, optimizes partnerships, and builds capacity to ensure sustainable food production, promote healthy food systems, improve markets and trade, transform agriculture, build resilience, and strengthen institutions and governance. Gender is considered in all of the Institute’s work. IFPRI collaborates with partners around the world, including development implementers, public institutions, the private sector, and farmers’ organizations, to ensure that local, national, regional, and global food policies are based on evidence. IFPRI is a member of the CGIAR Consortium. AUTHOR Samuel Benin (s.benin@cgiar.org) is Deputy Division Director for Africa and Senior Research Fellow in the Development Strategy and Governance Unit of the International Food Policy Research Institute (IFPRI), based in Davis, California. Notices 1 IFPRI Discussion Papers contain preliminary material and research results and are circulated in order to stimulate discussion and critical comment. They have not been subject to a formal external review via IFPRI’s Publications Review Committee. Any opinions stated herein are those of the author(s) and are not necessarily representative of or endorsed by IFPRI. 2 The boundaries and names shown, and the designations used on the map(s) herein, do not imply official endorsement or acceptance by the International Food Policy Research Institute (IFPRI) or its partners and contributors. 3 Copyright remains with the authors. The authors are free to proceed, without further IFPRI permission, to publish this paper, or any revised version of it, in outlets such as journals, books, and other publications. mailto:s.benin@cgiar.org iii Contents ABSTRACT .............................................................................................................................................................. V ACKNOWLEDGMENTS ........................................................................................................................................... VI ABBREVIATIONS AND ACRONYMS ....................................................................................................................... VII 1. INTRODUCTION .................................................................................................................................................. 1 2. CONCEPTUAL FRAMEWORK ............................................................................................................................... 4 3. EMPIRICAL APPROACH ....................................................................................................................................... 6 3.1. THE EMPIRICAL MODEL ............................................................................................................................................. 6 3.2. ECONOMETRIC METHODS AND ESTIMATION ISSUES ........................................................................................................ 7 3.3. DATA, SOURCES, AND DATA QUALITY ISSUES ................................................................................................................. 7 3.4. OUTCOME AND EXPLANATORY VARIABLES .................................................................................................................. 12 4. COMPOSITION OF AGRICULTURE EXPENDITURE AND OUTPUT ........................................................................ 16 4.1. TRENDS IN TOTAL GOVERNMENT EXPENDITURE ON AGRICULTURE .................................................................................... 16 4.2. SUBSECTOR COMPOSITION OF AGRICULTURAL EXPENDITURE AND OUTPUT ........................................................................ 17 5. DETERMINANTS AND EFFECTS OF AGRICULTURAL EXPPENDITURE .................................................................. 21 5.1. SAMPLE SIZE AND SUMMARY STATISTICS OF THE VARIABLES............................................................................................ 21 5.2. DETERMINANTS OF GOVERNMENT AGRICULTURE EXPENDITURE ...................................................................................... 21 5.3. EFFECTS OF THE COMPOSITION OF GOVERNMENT AGRICULTURE EXPENDITURE ON AGRICULTURAL LAND PRODUCTIVITY ............. 27 6. CONCLUSIONS AND IMPLICATIONS .................................................................................................................. 34 REFERENCES ......................................................................................................................................................... 36 ANNEX TABLES ..................................................................................................................................................... 39 iv Tables TABLE 1: GOVERNMENT AGRICULTURE EXPENDITURE, % OF TOTAL GOVERNMENT EXPENDITURE (COUNTRIES WITH UNLIKELY OR UNREALISTIC DATA), 2015-2020 ................................................................................................................................ 9 TABLE 2: GOVERNMENT AGRICULTURAL SUBSECTOR EXPENDITURE, % GOVERNMENT AGRICULTURE EXPENDITURE (COUNTRIES WITH UNLIKELY OR UNREALISTIC DATA), 2015-2020 ............................................................................................................... 9 TABLE 3: GOVERNMENT AGRICULTURE EXPENDITURE, % AGRICULTURE VALUE ADDED (COUNTRIES WITH UNLIKELY OR UNREALISTIC DATA), 2015-2020 .......................................................................................................................................................... 10 TABLE 4: GOVERNMENT AGRICULTURE RESEARCH EXPENDITURE, % AGRICULTURE VALUE ADDED (COUNTRIES WITH UNLIKELY OR UNREALISTIC DATA), 2015-2020 .............................................................................................................................. 10 TABLE 5: AGRICULTURE VALUE ADDED GROWTH RATE, % (COUNTRIES WITH UNLIKELY OR UNREALISTIC DATA), 2015-2020 ................. 11 TABLE 6: AGRICULTURE SUBSECTOR VALUE ADDED, % AGRICULTURE VALUE ADDED (COUNTRIES WITH UNLIKELY OR UNREALISTIC DATA), 2015-2020 .......................................................................................................................................................... 11 TABLE 7: TOTAL GOVERNMENT AGRICULTURE EXPENDITURE MINUS SUM OF GOVERNMENT AGRICULTURE SUBSECTOR EXPENDITURE, % DIFFERENCE (COUNTRIES WITH UNLIKELY OR UNREALISTIC DATA OR NON-ZERO DIFFERENCE), 2015-2020 ................................ 12 TABLE 8: TOTAL AGRICULTURE VALUE ADDED MINUS SUM OF AGRICULTURE SUBSECTOR VALUE ADDED, % DIFFERENCE (COUNTRIES WITH UNLIKELY OR UNREALISTIC DATA OR NON-ZERO DIFFERENCE), 2015-2020 ......................................................................... 12 TABLE 9: DESCRIPTION OF VARIABLES USED IN THE REGRESSIONS................................................................................................ 14 TABLE 10: GOVERNMENT AGRICULTURE EXPENDITURE IN AFRICA, 2014-2020 ............................................................................ 16 TABLE 11: SUMMARY STATISTICS OF THE VARIABLES USED IN THE ECONOMETRIC ESTIMATIONS, 2014-2020. .................................... 22 TABLE 12: FACTORS AFFECTING GOVERNMENT AGRICULTURE EXPENDITURE, 2014-2020 .............................................................. 24 TABLE 13: FACTORS AFFECTING SUBSECTOR COMPOSITION OF GOVERNMENT AGRICULTURE EXPENDITURE, 2014-2020 ...................... 26 TABLE 14: EFFECT OF GOVERNMENT AGRICULTURE EXPENDITURE (GAE) ON TOTAL AGRICULTURE VALUE ADDED PER UNIT AREA, 2014- 2020 ................................................................................................................................................................... 29 TABLE 15: EFFECT OF GOVERNMENT AGRICULTURE SUBSECTOR EXPENDITURE ON TOTAL AGRICULTURE VALUE ADDED PER UNIT AREA, 2014-2020 .......................................................................................................................................................... 30 TABLE 16: EFFECT OF AGRICULTURE SUBSECTOR EXPENDITURE ON SUBSECTOR VALUE ADDED PER UNIT AREA WITHOUT CROSS-SUBSECTOR SPENDING EFFECTS, 2014-2020 ............................................................................................................................... 31 TABLE 17: EFFECT OF AGRICULTURE SUBSECTOR EXPENDITURE ON SUBSECTOR VALUE ADDED PER UNIT AREA WITH CROSS-SUBSECTOR SPENDING EFFECTS, 2014-2020 ............................................................................................................................... 32 Figures FIGURE 1: ANNUAL AVERAGE AGRICULTURAL TOTAL FACTOR PRODUCTIVITY GROWTH IN THE WORLD, 2010-2019 ............................... 1 FIGURE 2: SCATTERPLOT OF AGRICULTURE ORIENTATION INDEX (AOI), 2014-2020 ..................................................................... 17 FIGURE 3: SHARE OF SUBSECTOR IN AGRICULTURE VALUE ADDED AND GOVERNMENT EXPENDITURE IN AFRICA, 2014-2020 ................. 18 FIGURE 4: AGRICULTURE ORIENTATION INDEX (AOI) IN AFRICA BY SUBSECTOR, 2014-2020 .......................................................... 18 FIGURE 5: SCATTERPLOT OF AGRICULTURE ORIENTATION INDEX (AOI) AND SHARE IN VALUE ADDED BY SUBSECTOR, 2014-2020 ........... 19 FIGURE 6: AGRICULTURE ORIENTATION INDEX (AOI) AND EXPENDITURE SHARES IN AFRICA BY DOMINANT SUBSECTOR, 2014-2020 ANNUAL AVERAGE ................................................................................................................................................... 19 v ABSTRACT This paper analyzes the determinants of the composition of government agriculture expenditure (GAE) in Africa and estimates the effect of the composition on agricultural productivity using cross-country annual data from 2014 to 2020 and structural equations modeling methods. It includes different specifications of the explanatory variables to assess the sensitivity of the results to different assumptions of the conceptual variables that are hypothesized to affect the composition and pathways of impact of government expenditure. The results show that there is a wide variation in GAE across African countries, and few have achieved the 10 percent CAADP agriculture expenditure target. Most African countries spend much smaller proportions of the national budget on agriculture than the sector’s share in the economy, and total agriculture expenditure seems to be allocated across subsectors according to their relative contribution to the sector’s output, with forestry and fisheries being slightly favored compared with crops and livestock, which dominate the sector. The allocation is also affected by several factors, such as past output and size of the subsector, official development assistance, education, irrigation, and state of agricultural transformation, although there are cross-subsector differences in their influence. There are also subsector differences in the estimated effect of GAE on land productivity: 0.06 to 0.08 for expenditure on the total sector, 0.02 for research, 0 to 0.09 for crops, 0 to 0.08 for livestock, and 0 to 0.07 for fisheries. The lower bound of zero means that the estimated effect is not statistically significant in some of the model specifications, such as whether cross-subsector expenditure effects are considered. We discuss implications of the results and suggestions for future research. vi ACKNOWLEDGMENTS This work is funded by the Bill and Melinda Gates Foundation via the Akademiya2063–IFPRI joint project on Biennial Review and Data Systems Strengthening and by the United States Agency for International Development. vii ABBREVIATIONS AND ACRONYMS AOI agriculture orientation index BR biennial review CAADP Comprehensive Africa Agriculture Development Programme CLFE-SEM cross-lagged fixed-effects simultaneous equations modeling GAE government agriculture expenditure GDP gross domestic product ODA official development assistance R&D research and development SEM simultaneous equations modeling 1 1. INTRODUCTION Though two decades have passed since the African heads of state and government committed to allocate at least 10 percent of their annual national budgets to the agriculture sector, popularly known as the Maputo Declaration, very few African countries have reached or surpassed the target—for example, four in 2021 (AUC 2022) and three in 2023 (AUC 2024). Furthermore, the average achieved for the continent has persistently declined over time, from more than 6 percent per year in the 1980s to a little over 2 percent in 2020 (Benin 2022; ReSAKSS 2022). Failure to meet the target may not be of concern, as the 10 percent goal seems arbitrary. However, because public agriculture expenditure has high returns in terms of growth (Fan 2008; Mogues et al. 2015), and agricultural growth may have been more effective at reducing poverty than growth originating from other sectors (World Bank 2007, 2015), the persistent decline in the GAE share, whether measured relative to total government expenditure or agriculture value added, is of concern. In the last decade (2010–2019), for example, agricultural productivity growth in Africa has been the slowest compared with other developing regions of the world, with about one-half of African countries achieving a negative annual average agricultural total factor productivity growth rate (Figure 1). Figure 1 Annual average agricultural total factor productivity growth in the world, 2010–2019 Source: 2021 Global Agricultural Productivity Report (Steensland 2021). Furthermore, the contribution of agriculture to gross domestic product (GDP) has remained at about 15 percent on average for the continent since the 1980s (ReSAKSS 2022), whereas the expected role of agriculture in development has broadened beyond providing food security to addressing malnutrition, reducing poverty, and building resilience among other outcomes, while facing challenges like climate change, degrading natural resources, and widespread pests and diseases (AU 2014). Making a case to convince policymakers to increase the share of the national budget allocated to agriculture to reverse the declining trend would require new evidence on the productivity and outcomes of recent trends in government expenditure on agriculture relative to other sectors. The https://globalagriculturalproductivity.org/map/ 2 trends and composition of public agriculture expenditure associated with the periods analyzed in the studies that are commonly cited in reference to the evidence on the returns to public agriculture spending in Africa are different from recent trends, especially those following the Maputo Declaration (Benin 2015; Goyal and Nash 2017). In addition, African leaders have signed on to various charters that demand hefty shares of the national budget for other sectors. For example, the “Abuja Declaration on HIV/AIDS, Tuberculosis and Other Related Infectious Diseases” calls for 15 percent of the national budget to be spent on the health sector (AU 2001), and the “Nairobi Declaration and Call for Action on Education” calls for at least 15 to 20 percent of total public expenditure for education (AU 2018). Thus, convincing policymakers to increase the share of the national budget allocated to agriculture will also depend on political economy factors such as the incentives and constraints of the actors (politicians, bureaucrats, interest groups, and donors) in favor of or lobbying for agriculture and the Maputo Declaration, compared with those in favor of or lobbying for other sectors or declarations (Mogues 2015). Using cross-country annual data from 2014 to 2020 from the third cycle of the Comprehensive Africa Agriculture Development Programme (CAADP) biennial review (BR) process, this paper contributes to providing new evidence on the effect of recent trends in different types of GAE as well as the political economy factors associated with the recent trends. Specifically, our objective is to analyze the determinants of the GAE composition and estimate the effect of the composition on agricultural growth. We use structural equations modeling (SEM) methods to estimate the relationship between expenditure and growth, including different specifications of the explanatory variables to assess the sensitivity of the results to different assumptions of the conceptual variables that are hypothesized to affect the composition and pathways of impact of government expenditure. By integrating the analysis of the determinants of composition of government expenditure with the estimation of the effect of spending on growth, this paper’s contribution is unique to the extent that it discerns the underlying causative process of government spending. Briefly, the results show that there is a wide cross-country variation in government agriculture expenditure in Africa, and few African countries have achieved the 10 percent CAADP expenditure target. Overall, at the sector level, most African countries spend much smaller proportions of the public budget on agriculture than the sector’s share in the economy. At the subsector level, the distribution of total agriculture expenditure is more equal to its relative contribution to the sector’s output, with forestry and fisheries being slightly favored compared with crops and livestock, which dominate the sector. The subsector allocation is also affected by several factors, such as past output and size of the subsector, net official development assistance (ODA), education, irrigation, and state of agricultural transformation, although there are subsector differences in their influence. There are also subsector differences in the estimated effect of GAE on land productivity: 0.06 to 0.08 for expenditure on the total sector, 0.02 for research, 0 to 0.09 for crops, 0 to 0.08 for livestock, and 0 to 0.07 for fisheries. The lower bound of zero means that the estimated effect is not statistically significant in some of the model specifications, such as whether cross-subsector expenditure effects are considered. 3 In the next section, we present the conceptual framework on the allocation of government expenditure to various activities or sectors and how different types of government expenditure affect growth. We present the estimation methods and data in section 3, followed by the results in sections 4 (trends of composition of expenditures and output) and 5 (determinants and effects of composition of expenditures). We conclude with a discussion of the implications in section 6. 4 2. CONCEPTUAL FRAMEWORK Following the modeling approach that is well established in the literature (Aschauer 1989; Barro 1990; Fan 2008; Mogues and Benin 2012; Thirtle et al. 2003), the conceptual framework used here considers public expenditure in agriculture as contributing to a stock of public capital in the sector, which in turn contributes to agricultural growth. The growth linkage can arise through various channels, namely productivity effects that advance technology, enhance human capital, reduce transaction costs, and crowd in (Benin 2019; Benin and Odjo 2018). The technology- advancing productivity effects derive from yield-enhancing technologies that are generated from public expenditure in agricultural research and development (R&D). The human capital– enhancing productivity effects derive from public expenditure in agricultural education, extension, and information, which all help increase the knowledge and skills of farmers and those engaged in agricultural production (Schultz 1982). The transaction cost-reducing productivity effects derive from public expenditure on infrastructure in the agricultural sector (such as storage facilities, market information, and feeder roads), which in turn contributes to improved access to input and output markets and thus reduces the cost of agricultural inputs and technologies (Sadoulet and de Janvry 1995). The crowding-in productivity effects of public expenditure in agriculture is a second-order effect in which the increase in public capital in the agricultural sector causes an increase in private capital in the sector. In addition to the various channels through which the productivity effects of public agriculture expenditure can arise, the literature also shows that the effects are not the same for all types of expenditure, and the productivity effects often materialize with a lag rather than contemporaneously. Furthermore, the public expenditure decision may be endogenous, as the amount spent or invested in a sector or activity may depend on the performance of the sector or returns to investment in the activity. So is the notion that growth in public capital is an endogenous process, that is, an outcome of growth in income rather than only a cause of it. These various effects and reverse causality issues can be captured in a model in which agricultural subsector output is specified as a function of the stocks of both public and private capital in the respective agricultural subsectors, as well as the use of resources such as land, labor, and modern inputs in the respective subsectors, and exogenous factors such as rainfall and shocks (e.g., floods and droughts), which may be distinguishable by subsector or not. The stocks of public and private capital depend, respectively, on accumulated public and private spending over time, with relevant time lag structures and depreciation rates. The stocks of public capital in the various subsectors can also be disaggregated by type of spending (for example, R&D, extension, training, storage facilities, market information, and feeder roads). Including the lag value of the respective subsector output is useful for addressing the endogeneity of growth in the public capital stock. The stock of private capital and the use of other resources (land, labor, and modern inputs) in the various subsectors can be captured in a second set of equations, where they are specified as a function of the stock of public capital, along with other variables such as prices, policies, terms of trade, infrastructure, access to financial and technical services, access to input and output markets, land pressure, and so on. These explanatory variables may or may not be distinguished 5 by subsector. The endogeneity of the public expenditure decision process is captured in a third set of equations that explain the level of public spending each year in the various subsectors and activities, typically using variables on political processes and institutional arrangements. Including the lag value of the respective subsector expenditure, public capital stock, and output is also useful. 6 3. EMPIRICAL APPROACH The literature on the determinants and effects of the composition of public spending as presented in the conceptual framework is dominated by the analysis of public expenditure that is disaggregated by the broader functions of government or sector (e.g., infrastructure, health, education, agriculture, and defense) or economic use (e.g., capital vs. recurrent), as found in Devarajan et al. (1996), Easterly and Rebelo (1993), and Fan (2008), for example. Studies on the determinants and effects of public spending on agriculture that focus on a single type of agricultural expenditure, such as research, extension, or farm support subsidies, dominate this subset of the literature, in addition to studies on irrigation and marketing infrastructure, among others. The lack of studies on the determinants and effects of the composition of public spending in agriculture as presented in the conceptual framework is attributed to the lack of time series and cross-country data on public agriculture expenditure that is disaggregated by subsector (crops, livestock, fishery, forestry), commodity (e.g., cereals, vegetables, poultry, and dairy), and activity (e.g., R&D, extension, pest control, training, storage facilities, market information, and feeder roads), for example. Benin’s (2019) study on Ghana, which disaggregates GAE into cocoa and noncocoa subsectors, is representative of the conceptual framework presented earlier. The study finds that whereas the estimated elasticities of government spending on output in the noncocoa subsector are mostly positive and statistically significant, those for the cocoa subsector are mostly statistically insignificant, with a mix of positive and negative effects. Another paper with a twist to the conceptual framework is Abdullahi’s (2021) study on Nigeria, which estimates the effect of total GAE on subsector output (crops, livestock, fishing, and forestry). The study finds that while the estimated effect of total GAE on output in the crops and fishing subsectors is positive and statistically significant (with the effect being larger for the crops subsector), the estimated effect in the livestock subsector is positive but statistically insignificant, and the estimated effect in the forestry subsector is negative and statistically significant. These results are in line with the literature showing that the effects of public spending are not the same for all types of expenditure. 3.1. The empirical model There are no data on agricultural public capital stock, whether total or disaggregated by subsector (crops, livestock, fisheries, and forestry). However, there are data on total (public and private) agricultural capital stock (more on the data later). Thus, we estimate a reduced-form model of the one presented in the conceptual framework by including the agricultural subsector expenditures along with the total agricultural capital stock in their respective agricultural subsector output equations. Equations 1 and 2 show the reduced-form model, , where 𝑌𝑌𝑠𝑠𝑠𝑠 is the agricultural output in the sth subsector at time t, 𝐺𝐺𝑠𝑠𝑠𝑠 is the amount of GAE allocated to the sth subsector at time t, and 𝐾𝐾𝑠𝑠 is total (public and private) agricultural capital stock at time t. 𝑌𝑌𝑠𝑠𝑠𝑠 = 𝛼𝛼𝑠𝑠𝑌𝑌 + 𝑮𝑮𝑠𝑠𝑠𝑠′ 𝜹𝜹𝑠𝑠 𝑌𝑌,𝑮𝑮 + 𝜑𝜑𝑠𝑠𝑌𝑌𝐾𝐾𝑠𝑠 + 𝒁𝒁𝑠𝑠𝑠𝑠′ 𝛽𝛽𝑠𝑠𝑌𝑌 + 𝜏𝜏𝑠𝑠𝑌𝑌𝑡𝑡 + 𝜀𝜀𝑠𝑠𝑠𝑠𝑌𝑌 (1) 𝐺𝐺𝑠𝑠𝑠𝑠 = 𝛼𝛼𝑠𝑠𝐺𝐺 + 𝜑𝜑𝑠𝑠𝐺𝐺𝐾𝐾𝑠𝑠 + 𝑾𝑾𝑠𝑠𝑠𝑠 ′ 𝛽𝛽𝑠𝑠𝐺𝐺 + 𝜏𝜏𝑠𝑠𝐺𝐺𝑡𝑡 + 𝜀𝜀𝑠𝑠𝑠𝑠𝐺𝐺 (2) 7 The variables 𝒁𝒁𝑠𝑠𝑠𝑠′ and 𝑾𝑾𝑠𝑠𝑠𝑠 ′ capture the vector of factors (with the notion that there may be some common elements to them) that determine or influence 𝑌𝑌𝑠𝑠𝑠𝑠 and 𝐺𝐺𝑠𝑠𝑠𝑠, respectively, and 𝜀𝜀𝑠𝑠𝑠𝑠𝑌𝑌 and 𝜀𝜀𝑠𝑠𝑠𝑠𝐺𝐺 are the error terms for 𝑌𝑌𝑠𝑠𝑠𝑠 and 𝐺𝐺𝑠𝑠𝑠𝑠, respectively. From the parameters of the model, the total effect of government agricultural subsector expenditure on agricultural subsector output is given by 𝜹𝜹𝑠𝑠 𝑌𝑌,𝑮𝑮 + 𝜋𝜋𝑠𝑠 ∗ 𝜑𝜑𝑠𝑠𝑌𝑌, where 𝜋𝜋𝑠𝑠, which is to be obtained from outside of the model estimation, is the contribution or share of government agricultural subsector expenditure to the total (public and private) agricultural capital stock. The first term, 𝜹𝜹𝑠𝑠 𝑌𝑌,𝑮𝑮, may be denoted as the direct effect of GAE, and the second term, 𝜋𝜋𝑠𝑠 ∗ 𝜑𝜑𝑠𝑠𝑌𝑌, as the indirect effect via the capital stock. Note that 𝜹𝜹𝑠𝑠 𝑌𝑌,𝑮𝑮 is an sxs matrix of the estimator for the effect of GAE in one subsector on agricultural output in the same subsector, as well as on agricultural output in the other subsectors. For the other estimators of the model, 𝛼𝛼𝑠𝑠𝑌𝑌 and 𝛼𝛼𝑠𝑠𝐺𝐺 are the intercepts in the respective equations, 𝜑𝜑𝑠𝑠𝑌𝑌estimates the effect of the total capital stock on subsector expenditure, 𝛽𝛽𝑠𝑠𝐺𝐺 is the estimator for the effect of other factors (W) considered in the subsector expenditure allocation decision process, 𝛽𝛽𝑠𝑠𝑌𝑌 measures the effect of the factors (Z) on agricultural subsector output, and 𝜏𝜏𝑠𝑠𝑌𝑌 and 𝜏𝜏𝑠𝑠𝐺𝐺 measure the effect of time on the respective dependent variables. 3.2. Econometric methods and estimation issues We used annual data on African countries from 2014 to 2020 to estimate the system of equations using a cross-lagged fixed-effects structural equations modeling (CLFE-SEM) method, with country and year fixed effects to control for unobserved cross-country heterogeneity and to eliminate bias due to omitted time-invariant variables. The error terms 𝜀𝜀𝑠𝑠𝑠𝑠𝑌𝑌 and 𝜀𝜀𝑠𝑠𝑠𝑠𝐺𝐺 are assumed to have a mean of zero and constant variance, uncorrelated with the explanatory variables, and uncorrelated across equations or over time (Greene 1993; Zellner and Theil 1962). The main estimation issues are identification of the system or endogeneity of the subsector GAE (𝐺𝐺𝑠𝑠𝑠𝑠) and potential reverse causality of agricultural output and expenditure. We address endogeneity by using exclusion restrictions. The rule of thumb for using exclusion restrictions is that the number of exogenous variables excluded from equation 1 must be at least as large as the number of endogenous variables (i.e., 𝐺𝐺𝑠𝑠𝑠𝑠) included in equation 1. Let us denote this subset of exclusion restriction variables or instruments by 𝑾𝑾𝑠𝑠𝑠𝑠 𝐺𝐺 ⊂ Wst and use variables on political processes and institutional arrangements for it. To address reverse causality, we include the lagged value of output in both Zst and Wst and use the lagged value of the subsector expenditure instead of the contemporaneous value in equation 1 (i.e., replace 𝐺𝐺𝑠𝑠𝑠𝑠 with 𝐺𝐺𝑠𝑠𝑠𝑠−1) (Moral-Benito et al. 2018). Including the lagged value of output in Zst also helps to eliminate bias due to time-variant omitted variables. Replacing 𝐺𝐺𝑠𝑠𝑠𝑠 with 𝐺𝐺𝑠𝑠𝑠𝑠−1 in equation 1 also addresses the endogeneity of 𝐺𝐺𝑠𝑠𝑠𝑠, which may render strict exclusion restrictions unnecessary. Last, as total agricultural capital stock is derived from public and private sources, 𝐾𝐾𝑠𝑠−1 is used in place of 𝐾𝐾𝑠𝑠. 3.3. Data, sources, and data quality issues We obtained the core data on subsector-disaggregated agricultural output and government agriculture expenditure from the third cycle of the CAADP BR on 46 indicators from 2014 to 8 2022 (AUC 2022). We combined these with data from the World Development Indicators (World Bank 2023) and other publicly available databases on agricultural land use (FAO 2023), rainfall (World Bank 2022), and agricultural transformation classification (Benin 2021). Because there are a maximum of 385 observations in the BR database (55 countries for seven years) and we must control for several factors in the econometric estimation, a fundamental objective is to include the relevant variables on which there are data for as many countries and years as possible. This helps avoid degrees-of-freedom problems that arise from estimating several equations and having variables with a small number of observations. There are several data quality issues with the CAADP BR data that severely reduce the number of reliable observations to use in the econometric estimation. See Benin et al. (2022), for example, for an assessment of the CAADP BR data based on analysis of missing observations; internal consistency checks of outliers, illogical responses, implausible values, and so on; and external consistency checks in comparison with data on the similar indicators from other publicly available databases. For the main variables of interest in this paper—GAE and agricultural output and growth—Tables 1 to 8 illustrate some of the data quality issues with the CAADP BR data. For the GAE measured as a percentage of total government expenditure (BR Indicator 2.1i), for example, Table 1 shows that for five countries (Angola, Chad, DRC, South Africa, and South Sudan), the entire series or data for some years do not seem correct. The values for South Africa, for example, are unrealistically low, and it seems that the units of measurement used for the numerator and the denominator are not the same (e.g., millions are used for the numerator, and thousands or hundreds for the denominator). The issue is similar for Angola for all years and Chad for 2019 and 2020. For DRC, however, the values are unrealistically high. A variant of BR Indicator 2.1i that is of primary interest in this paper is the disaggregation by subsector (crops, livestock, fisheries, forestry), measured as government agricultural subsector expenditure as a percentage of GAE. As Table 2 shows, the data for several countries—Djibouti, Ethiopia, Mauritania, and Tunisia—are static for all the years, or several years for the other 10 countries identified. This is likely a result of using one- or two-years’ worth of data as an approximate of the values for the other years. Similar issues are identified with other BR indicators or their variants, such as GAE measured as a percentage of agriculture value added (BR Indicator 2.1ii, Table 3), government agriculture research expenditure measured as a percentage of agriculture value added (BR Indicator 3.1v, Table 4), agriculture value-added growth rate (BR Indicator 4.1i, Table 5), and agriculture subsector value added measured as a percentage of total agriculture value added (Table 6). Another issue is that for the subsector disaggregation of expenditure or value added, the sum of the parts does not add up to the total value (Table 7 for GAE and Table 8 for agriculture value added). 9 Table 1 Government agriculture expenditure, percentage of total government expenditure (countries with unlikely or unrealistic data), 2015–2020 Country 2015 2016 2017 2018 2019 2020 Angola 0.6 0.5 0.4 0.5 0.3 0.3 Chad 3.9 1.5 3.6 3.2 0.0001 0.000006 DR Congo 99.6 35.4 28.7 77.7 41.8 69.9 South Africa 0.0002 0.0001 0.0001 0.0001 0.0001 0.0001 South Sudan 0.2 0.1 0.1 0.03 6.7 16.0 Source: Author’s calculations based on AUC (2022). Note: DR Congo = Democratic Republic of the Congo. Table 2 Government agricultural subsector expenditure, percentage of government agriculture expenditure (countries with unlikely or unrealistic data), 2015–2020 Country Subsector 2015 2016 2017 2018 2019 2020 Benin Crop 73.2 73.2 73.2 73.5 82.2 85.5 Livestock 11.8 11.8 11.8 11.6 10.6 7.0 Fisheries 12.4 12.4 12.4 12.0 6.1 6.3 Forestry 2.6 2.6 2.6 2.9 1.1 1.2 Burundi Crop 70.0 70.0 70.0 70.0 73.5 62.0 Livestock 19.0 19.0 19.0 19.0 13.1 34.4 Fisheries 3.0 3.0 3.0 3.0 1.9 0.2 Forestry 8.0 8.0 8.0 8.0 11.5 3.4 Djibouti Crop 20.0 20.0 20.0 20.0 20.0 20.0 Livestock 45.0 45.0 45.0 45.0 45.0 45.0 Fisheries 35.0 35.0 35.0 35.0 35.0 35.0 Forestry 0.0 0.0 0.0 0.0 0.0 0.0 DR Congo Crop 53.3 55.0 55.0 55.0 55.0 55.0 Livestock 29.2 30.0 30.0 30.0 30.0 30.0 Fisheries 11.7 10.0 10.0 10.0 10.0 10.0 Forestry 5.8 5.0 5.0 5.0 5.0 5.0 Egypt Crop 45.0 45.0 45.0 45.0 23.0 23.0 Livestock 25.0 25.0 25.0 25.0 35.0 35.0 Fisheries 23.0 23.0 23.0 23.0 35.0 35.0 Forestry 7.0 7.0 7.0 7.0 7.0 7.0 Equatorial Guinea Crop 52.5 49.1 49.1 48.4 48.4 48.4 Livestock 18.6 24.4 24.4 24.0 24.0 24.0 Fisheries 28.9 12.9 12.9 15.5 15.5 15.5 Forestry 0.0 13.6 13.6 12.0 12.0 12.0 Ethiopia Crop 36.1 36.1 36.1 36.1 36.1 36.1 Livestock 51.6 51.6 51.6 51.6 51.6 51.6 Fisheries 0.3 0.3 0.3 0.3 0.3 0.3 Forestry 11.9 11.9 11.9 11.9 11.9 11.9 Gambia Crop 40.0 40.5 45.0 40.5 40.0 40.5 Livestock 25.0 27.0 25.0 27.0 25.0 27.0 Fisheries 15.0 20.3 10.0 20.3 15.0 20.3 Forestry 20.0 12.2 20.0 12.2 20.0 12.2 Mauritania Crop 27.0 27.0 27.0 27.0 27.0 27.0 Livestock 34.0 34.0 34.0 34.0 34.0 34.0 Fisheries 28.0 28.0 28.0 28.0 28.0 28.0 Forestry 11.0 11.0 11.0 11.0 11.0 11.0 10 Country Subsector 2015 2016 2017 2018 2019 2020 Sierra Leone Crop 27.5 25.0 25.0 25.0 25.5 11.2 Livestock 21.8 19.9 19.9 19.9 27.4 29.4 Fisheries 32.2 38.2 38.2 38.2 40.0 53.1 Forestry 18.5 16.9 16.9 16.9 7.1 6.3 Sudan Crop n.d. 76.8 76.8 76.8 76.8 76.8 Livestock n.d. 17.2 17.2 17.2 14.1 17.2 Fisheries n.d. 2.0 2.0 2.0 3.0 2.0 Forestry n.d. 4.0 4.0 4.0 6.1 4.0 Tanzania Crop 43.7 62.4 43.8 43.8 43.8 43.8 Livestock 29.3 18.7 26.3 26.3 26.3 26.3 Fisheries 1.6 1.2 1.6 1.6 1.6 1.6 Forestry 25.4 17.7 28.3 28.3 28.3 28.3 Tunisia Crop 50.0 50.0 50.0 50.0 50.0 50.0 Livestock 27.0 27.0 27.0 27.0 27.0 27.0 Fisheries 8.0 8.0 8.0 8.0 8.0 8.0 Forestry 15.0 15.0 15.0 15.0 15.0 15.0 Zimbabwe Crop 61.2 99.0 93.3 89.6 25.3 99.5 Livestock 30.4 0.0 5.9 9.0 0.5 0.2 Fisheries 2.3 0.3 0.2 0.4 n.d. 0.3 Forestry 6.1 0.7 0.6 1.0 1.1 0.1 Source: Author’s calculations based on AUC (2022). Note: n.d. = no data. Table 3 Government agriculture expenditure, percentage of agriculture value added (countries with unlikely or unrealistic data), 2015–2020 Country 2015 2016 2017 2018 2019 2020 Congo 94.0 93.2 94.0 23.4 93.4 93.4 Liberia 13.5 2.6 2.0 2.4 80.9 86.2 Mauritania 10.0 10.0 10.0 10.0 10.0 10.0 Nigeria 0.5 0.5 0.5 0.6 0.7 0.6 South Sudan 10.2 10.5 10.3 10.6 100.0 100.0 Sudan 0.5 6305.9 5947.6 9110.2 14729.9 4496.9 Zimbabwe 7.2 50.1 72.3 44.5 372.3 970.5 Source: Author’s example based on AUC (2022). Table 4 Government agriculture research expenditure, percentage of agriculture value added (countries with unlikely or unrealistic data), 2015–2020 Country 2015 2016 2017 2018 2019 2020 Burundi 308.7 339.0 265.2 242.2 273.8 249.7 Cameroon 56.8 38.5 0.4 0.3 0.2 0.3 Nigeria 0.0 0.0 0.0 0.0 0.0 0.0 Seychelles 2.3 2.1 2.6 3.3 7.4 16.9 Sudan n.d. 210.6 213.9 134.5 165.9 166.2 Zimbabwe 7.5 53.5 11.1 0.8 2.3 1.1 Source: Author’s example based on AUC (2022). Note: n.d. = no data. 11 Table 5 Agriculture value added growth rate, percentage (countries with unlikely or unrealistic data), 2015–2020 Country 2015 2016 2017 2018 2019 2020 Angola 23.3 −70.9 28.8 −44.0 −27.0 69.8 Burundi 959.9 1.8 18.3 −1.6 −0.5 −6.7 Congo −43.0 53.3 10.7 115.6 −72.2 −24.7 DR Congo n.d. −88.3 812.3 11.9 35.0 7.2 Guinea-Bissau 2.9 5.3 −0.3 0.7 −76.6 13.0 Liberia −87.1 734.7 19.4 74.9 −99.1 0.8 Malawi −1.1 0.1 6.3 0.9 306.3 6.7 Senegal 13.7 7.2 11.9 10,288.7 5.3 16.7 Sierra Leone n.d. 0.6 9.0 −90.4 1043.3 14.8 Source: Author’s calculations based on AUC (2022). Note: DR Congo = Democratic Republic of the Congo. n.d. = no data. Table 6 Agriculture subsector value added, percentage of agriculture value added (countries with unlikely or unrealistic data), 2015–2020 County Subsector 2015 2016 2017 2018 2019 2020 Djibouti Crop 20.0 20.0 20.0 20.0 20.0 20.0 Livestock 45.0 45.0 45.0 45.0 45.0 45.0 Fisheries 35.0 35.0 35.0 35.0 35.0 35.0 Forestry 0.0 0.0 0.0 0.0 0.0 0.0 Mauritania Crop 27.0 27.0 27.0 27.0 27.0 27.0 Livestock 34.0 34.0 34.0 34.0 34.0 34.0 Fisheries 28.0 28.0 28.0 28.0 28.0 28.0 Forestry 11.0 11.0 11.0 11.0 11.0 11.0 Tunisia Crop 57.0 57.0 57.0 57.0 57.0 57.0 Livestock 40.0 40.0 40.0 40.0 40.0 40.0 Fisheries 2.0 2.0 2.0 2.0 2.0 2.0 Forestry 1.0 1.0 1.0 1.0 1.0 1.0 Zambia Crop 100.0 100.0 88.4 n.d. 39.6 32.0 Livestock 0.0 0.0 n.d. n.d. n.d. n.d. Fisheries 0.0 0.0 n.d. n.d. n.d. n.d. Forestry 0.0 0.0 n.d. n.d. n.d. n.d. Zimbabwe Crop 63.8 61.9 71.9 68.8 6,534.4 6,255.1 Livestock 31.6 35.3 25.8 28.7 3,195.6 2,465.7 Fisheries 0.1 0.2 0.1 0.1 11.0 4.8 Forestry 4.5 2.7 2.2 2.3 259.0 172.0 Source: Author’s calculations based on AUC (2022). Note: n.d. = no data. 12 Table 7 Total government agriculture expenditure minus sum of government agriculture subsector expenditure, percentage difference (countries with unlikely or unrealistic data or nonzero difference), 2015–2020 Country 2015 2016 2017 2018 2019 2020 Botswana 17.8 55.5 52.5 60.5 8.4 6.5 Cameroon 53.6 56.1 64.5 49.4 55.0 Central African Republic 99.9 99.7 Lesotho 2.5 Liberia 20.9 52.6 Zimbabwe 73.0 Source: Author’s calculations based on AUC (2022). Note: A blank cell means the difference is zero or there are no data. Table 8 Total agriculture value added minus sum of agriculture subsector value added, percentage difference (countries with unlikely or unrealistic data or nonzero difference), 2015–2020 Country 2015 2016 2017 2018 2019 2020 Angola 100.0 99.9 100.0 100.0 Benin 0.4 0.9 −0.5 −3.7 2.1 Botswana 24.6 23.1 24.2 25.5 60.3 Congo 75.0 Liberia 87.5 85.8 87.5 Madagascar 8.9 Nigeria 14.3 South Africa 99.9 99.9 100.0 100.0 Zambia 11.6 60.4 68.0 Zimbabwe −9,900.0 −8,797.6 Source: Author’s illustration based on AUC (2022). Note: A blank cell means the difference is zero or there are no data. Looking at the data quality issues for one indicator at a time may not seem as concerning, as only a few countries may be affected, and only those countries or data points may need to be dropped from the analysis. However, when the data quality issues for all the relevant indicators of interest to the paper or regressions are considered together, then the issues are concerning, as many more countries must be dropped. For the main variables of interest reflected in Tables 1 to 8, a total of 29 countries are affected. We managed to fix some of the data issues, especially those that seem to have different units for numerators and denominators or had values that could be cross- checked or verified with national statistical abstracts that were available online. In the end, we kept 24 countries (Benin, Burkina Faso, Burundi, Cameroon, Cote d’Ivoire, Egypt, Ethiopia, Gabon, Gambia, Ghana, Guinea, Kenya, Madagascar, Malawi, Mali, Mozambique, Namibia, Rwanda, Senegal, Sudan, Tanzania, Togo, Tunisia, and Uganda) with a minimum of five years’ time series data that gave a total of 132 observations for the econometric estimation. 3.4. Outcome and explanatory variables Table 9 describes the variables used for the econometric estimation. The outcome variable is agricultural land productivity, which is measured as agriculture value added per unit area for the entire sector (AGVAy_tot). This is disaggregated by subsector—crops (AGVAy_crp), livestock (AGVAy_liv), fisheries (AGVAy_fis), and forestry (AGVAy_for), where the value added for the 13 subsector is divided by the area for the subsector. GAE, which is also disaggregated by subsector and for research, is measured in three ways. The first one is the common indicator, where GAE is measured as a share of total government expenditure for the entire sector (GAEsh_tot). This is also disaggregated by subsector, where the subsector expenditure is measured as a share of GAE for crops (GAEsh_crp), livestock (GAEsh_liv), fisheries (GAEsh_fis), forestry (GAEsh_for), and research (GAEsh_res). The second definition, referred to as government agriculture expenditure intensity, is where total GAE is measured relative to the size of the sector (GAEint_tot) and subsector GAE is measured relative to the size of the subsector for crops (GAEint_crp), livestock (GAEint_liv), fisheries (GAEint_fis), forestry (GAEint_for), and research (GAEint_res)1. The third definition of GAE is the agriculture orientation index (GAEaoi), which sort of combines the first two definitions. For the entire agricultural sector, it is measured as GAE’s share of total expenditure relative to agriculture’s share of GDP (GAEaoi_tot). This is also disaggregated by subsector and measured as the subsector’s expenditure share of GAE relative to the subsector’s share of agriculture value added for crops (GAEaoi_crp), livestock (GAEaoi_liv), fisheries (GAEaoi_fis), forestry (GAEaoi_for), and research (GAEaoi_res)2. The second and third definitions of GAE (i.e., GAEint and GAEaoi) provide better measures for cross-country comparison of government commitment to the agricultural sector. Using the three measures in the econometric estimation increases reliability of the results to the extent that their estimated effect on agricultural land productivity is in line with their hypothesized effect and statistically significant. The explanatory variables are selected to capture an array of exogenous factors (economic, sociopolitical, demographic, environmental or natural factors, and shocks such as wars, floods, or droughts) that influence agricultural land productivity and GAEs. They include lagged value of agricultural capital stock, lagged value of agriculture value added share, rural population growth rate, rainfall, population affected by natural disasters, years of basic education, life expectancy, infrastructure, governance, official development assistance, geographic location of country, agricultural transformation classification of country, and time trend. An inverse interaction term between irrigation and rainfall (LOIRRIG*RAIN) is included to help address an abnormal effect of rainfall on productivity that is likely to arise from a low rainfall–high productivity relationship in irrigated areas. The lagged values of the dependent variables (agricultural land productivity and government agriculture expenditure) are also included to address potential reverse causality as well as omitted time-variant factors. 1 As agriculture value added from research is unknown, agriculture value added for the entire sector is used. 2 See footnote 1. 14 Table 9 Description of variables used in the regressions Variable Description Agriculture output per hectare of land (AGVAy) AGVAy_tot Total agriculture value added divided by area under agriculture, 2015 US$ AGVAy_crp Crops value added divided by the area under crops, 2015 US$ AGVAy_liv Livestock value added divided by the area under livestock, 2015 US$ AGVAy_fis Fisheries value added divided by the area under fisheries, 2015 US$ AGVAy_for Forestry value added divided by the area under forestry, 2015 US$ Government agriculture expenditure as share of total expenditure (GAEsh) GAEsh_tot Agriculture expenditure, % of total government expenditure GAEsh_crp Crops expenditure, % of government agriculture expenditure GAEsh_liv Livestock expenditure, % of government agriculture expenditure GAEsh_fis Fisheries expenditure, % of government agriculture expenditure GAEsh_for Forestry expenditure, % of government agriculture expenditure GAEsh_res Research expenditure, % of government agriculture expenditure Government agriculture expenditure intensity (GAEint) GAEint_tot Agriculture expenditure, % of agriculture value added GAEint_crp Crops expenditure, % of crops value added GAEint_liv Livestock expenditure, % of livestock value added GAEint_fis Fisheries expenditure, % of fisheries value added GAEint_for Forestry expenditure, % of forestry value added GAEint_res Forestry expenditure, % of agriculture value added Government agriculture expenditure orientation index (GAEaoi) GAEaoi_tot GAEsh_tot divided by share of agriculture value added in total GDP GAEaoi _crp GAEsh_crp divided by share of crops value added in agriculture value added GAEaoi _liv GAEsh_liv divided by share of livestock value added in agriculture value added GAEaoi _fis GAEsh_fis divided by share of fisheries value added in agriculture value added GAEaoi _for GAEsh_for divided by share of forestry value added in agriculture value added GAEaoi _res GAEsh_res divided by share of agriculture value added in total GDP Explanatory variables TOTGEXP Lag of total government expenditure per capita, 2015 US$ AGCAPSTK Lag of agricultural capital stock divided by agriculture value added AGVAx_tot Lag of agriculture value added, % of total GDP AGVAx_crp Lag of crops value added, % of agriculture value added AGVAx_liv Lag of livestock value added, % of agriculture value added AGVAx_fis Lag of fisheries value added, % of agriculture value added AGVAx_for Lag of forestry value added, % of agriculture value added GOVERN Governance, average of six worldwide governance indicators, −2.5 to 2.5 NETODA Net official development assistance received, % of gross national income AGLNDLAB Ratio of agricultural land to labor RUPOPGR Rural population growth rate, % EDUCATE Compulsory education, years LIFEEXP Life expectancy, years INFRASTUR Quality of trade and transport-related infrastructure (1 = low to 5 = high) IRRIGATION 1 = country with 1% or more of area equipped with irrigation, 0 = otherwise RAINFALL Rainfall, mm LOIRRIG*RAIN Reverse interaction term where value = RAINFALL if IRRIGATION = 0; value = 0 if IRRIGATION = 1 AGTR-HHPD 1 = country with high initial (2003) agricultural employment and GDP shares and where labor productivity is declining (2003–2018), 0 = otherwise 15 Variable Description AGTR-HHPI 1 = country with high initial (2003) agricultural employment and GDP shares and where labor productivity is increasing (2003–2018), 0 = otherwise TIMETRND 1 = 2014, 2 = 2015, …, 2020 = 7 Source: Author’s definitions based on AUC (2022), World Bank (2022, 2023), FAO (2023), and Benin (2021). Note: Using the FAO land use classification (FAO 2023), area under agriculture = area under crops + area under livestock + area under fisheries + area under forestry. Area under crops is the total of areas under arable land and permanent crops. Area under livestock is the total of areas under temporary and permanent meadows and pastures. Area under fisheries is the total of areas under inland and coastal waters. Area under forestry is forest land. GDP = gross domestic product. 16 4. COMPOSITION OF AGRICULTURE EXPENDITURE AND OUTPUT 4.1. Trends in total government expenditure on agriculture Between 2014 and 2020, government expenditure on agriculture constituted an average of 6.3 percent of total government expenditure (which is below the CAADP 10 percent target) and 7.7 percent of agriculture value added (Table 10). There is a wide variation in spending across the countries, and few have achieved the 10 percent target, including Benin, Burundi, Ethiopia, and Mali, which have surpassed it in most of the years from 2014 to 2020. Because country conditions and, thus, spending contexts differ widely across the continent, the Agriculture Orientation Index (AOI), defined as agriculture’s share of public spending relative to its share in the economy, provides a better measure for cross-country comparison of government commitment to the sector.3 Table 10 Government agriculture expenditure in Africa, 2014–2020 Annual Indicator 2014 2015 2016 2017 2018 2019 2020 Average Expenditure on agriculture, % of total government expenditure 6.20 6.70 6.34 6.17 6.49 5.89 5.90 6.26 Expenditure on agriculture, % of agriculture value added 6.86 6.78 6.21 5.96 6.81 11.15 11.27 7.67 Agriculture orientation index (AOI) 0.33 0.32 0.30 0.28 0.31 0.31 0.29 0.30 Source: Author’s calculations based on AUC (2022) and World Bank (2022). Note: AOI is the ratio of agriculture share in total government expenditure to agriculture share in total gross domestic product. Overall, most African countries spend much smaller proportions of the public budget on agriculture than the sector’s share in the economy. Of the 47 countries for which the AOI can be computed, the majority have an index in the 0.1 to 0.5 range, with agriculture’s share of GDP in the 15 to 40 percent range and no clear pattern of the relationship between the two indicators (see conglomerate of plots in the middle of the first chart in Figure 2). Only one country in Africa, Namibia, has an AOI of 1 or more in any year from 2014 to 2020, although some countries like Senegal and Tunisia have values close to 1. Agriculture’s share of GDP in these countries is lower than 15 percent. The other outlier cases are Gabon, Republic of Congo, and Sudan, which have a similar low share of agriculture in GDP, but the AOI is lower than 0.4 (see circled plots to the lower lefthand side of the first chart in Figure 2). Chad, on the other hand, has low AOI and a high share of agriculture in GDP (see circled plots to the lower righthand side of the first chart in Figure 2). The second chart in Figure 2 shows that there is also no clear relationship between AOI and the diversity of the agricultural sector. Together, these results seem to suggest that government expenditure allocation to the agricultural sector is not directly related to the sector’s contribution to GDP or its diversity in terms of crops, livestock, fisheries, and forestry. Other factors may be more important in the allocation decisions, which we investigate later in the regression analysis. Although there is no reason expenditure must be allocated exactly in 3 An AOI value of 1 would indicate that the share of the total budget that the government spends on agriculture is equal to agriculture’s contribution to GDP. 17 proportion to each sector’s contribution to the economy, large deviations signal a need for deeper analysis by policy makers. Figure 2 Scatterplot of agriculture orientation index (AOI), 2014–2020 Source: Author’s illustration based on AUC (2022) and World Bank (2022). Note: Data are annual values. AOI is the ratio of agriculture share in total government expenditure to agriculture share in total gross domestic product (GDP). The agricultural subsector diversity index is the sum of squares of the subsector (crops, livestock, fisheries, and forestry) share in agriculture value added. The diversity of the sector declines with increasing index values and vice versa. 4.2. Subsector composition of agricultural expenditure and output The trends shown in Figure 3 indicate that the subsector share of GAE over time is very close to subsector share agriculture value added. The crops subsector dominates, however, with an average of 58 and 63 percent of agriculture expenditure and value added over 2014 to 2020, respectively, followed by livestock (17 and 19 percent, respectively), forestry (15 and 11 percent, respectively), and fisheries (10 and 7 percent, respectively). Looking at the subsector AOIs, we see that they are greater than 1 but very variable for forestry and fisheries, compared with those for crops and livestock, which are just under 1 but stable (Figure 4). Together, these results suggest that although overall agriculture share in total government expenditure is much less than the sector’s contribution to GDP (AOI is 0.3), the distribution of total agriculture expenditure across the subsectors is more equal to its relative contribution to the sector’s output, with forestry and fisheries being slightly favored compared with crops and livestock, which dominate the sector. Regarding the relationship between AOI and value added at the subsector level, the inverse relationship seems stronger there than at the aggregate level, with the fitted trend in the forestry subsector having the largest r-squared value, followed by crops, fisheries, and livestock (Figure 5). Delving deeper into the subsector composition of agricultural expenditure, Figure 6 reveals differences in overall agriculture expenditure allocation decision by the dominant subsector in the share of agriculture value added. For example, when the dominant subsector is fisheries (as in Gambia and Namibia), the overall AOI for agriculture is highest at 0.6, compared with where livestock dominates (as in Mauritania and South Sudan) or where crops dominate, as in most y = -0.0024x + 0.355 R² = 0.0111 0 0.2 0.4 0.6 0.8 1 1.2 0 10 20 30 40 50 60 AO I ( ag ric ul tu re ) Agriculture share in GDP (%) y = -0.0438x + 0.3266 R² = 0.0011 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 AO I ( ag ric ul tu re ) Agriculture subsector diversity index 18 places. These seem to reflect the need for more public goods and services to protect fisheries and grazing resources that are deemed more common property and outside the control of fishers and livestock producers, compared with the crops subsector, where the resources and production process may be relatively more under the control of farmers. A striking difference in the trends is with overall research expenditure, which is only 0.4 percent of total agriculture expenditure where the livestock subsector dominates, compared with 9.5 and 5.8 percent where crops and fisheries dominate, respectively (Figure 6). Figure 3 Share of subsector in agriculture value added and government expenditure in Africa, 2014–2020 Source: Author’s illustration based on AUC (2022) and World Bank (2022). Note: agVAD = agriculture value added. agEXP = agriculture government expenditure. Figure 4 Agriculture orientation index (AOI) in Africa by subsector, 2014–2020 Source: Author’s illustration based on AUC (2022) and World Bank (2022). Note: For all, AOI is the ratio of agriculture share in total government expenditure to agriculture share in total gross domestic product. For the subsectors, AOI is the ratio of subsector share in agriculture government expenditure to subsector share in agriculture value added. 62 59 64 56 64 56 63 59 62 58 61 56 62 59 21 19 19 17 19 16 19 16 18 17 18 15 18 17 7 9 7 10 7 12 7 11 7 11 9 12 9 9 9 13 10 17 9 15 11 15 13 14 12 17 11 15 0% 20% 40% 60% 80% 100% agVAD agEXP agVAD agEXP agVAD agEXP agVAD agEXP agVAD agEXP agVAD agEXP agVAD agEXP 2014 2015 2016 2017 2018 2019 2020 Crops Livestock Fisheries Forestry 0.0 0.5 1.0 1.5 2.0 2014 2015 2016 2017 2018 2019 2020 All Crops Livestock Fisheries Forestry 19 Figure 5 Scatterplot of agriculture orientation index (AOI) and share in value added by subsector, 2014–2020 Source: Author’s illustration based on AUC (2022) and World Bank (2022). Note: AOI is the ratio of subsector share in agriculture government expenditure to subsector share in agriculture value added. Figure 6 Agriculture orientation index (AOI) and expenditure shares in Africa by dominant subsector, 2014–2020 annual average Source: Author’s illustration based on AUC (2022) and World Bank (2022). Note: AOI is the ratio of agriculture share in total government expenditure to agriculture share in total gross domestic product (GDP). Crop-dom, Lvst-dom, and Fish-dom indicate that crops, livestock, or fisheries, respectively, is the dominant subsector. There were no cases in which the forestry subsector dominated. y = -0.0054x + 1.2891 R² = 0.1257 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 20 40 60 80 100 AO I ( cr op s) Crops share in agriculture value added (%) y = -0.022x + 1.4291 R² = 0.0772 0 1 2 3 4 5 6 0 20 40 60 80 100 AO I ( liv es to ck ) Livestock share in agriculture value added (%) y = -0.1433x + 4.0772 R² = 0.0856 -5 0 5 10 15 20 25 30 0 20 40 60 80 100 AO I ( fis he rie s) Fisheries share in agriculture value added (%) y = -0.1323x + 4.0364 R² = 0.1659 -5 0 5 10 15 20 0 20 40 60 80 100AO I ( fo re st ry ) Forestry share in agriculture value added (%) 0.31 0.36 0.58 6.67 7.69 5.32 9.49 0.40 5.80 0 2 4 6 8 10 0.0 0.2 0.4 0.6 Cr op -d om Lv st -d om Fi sh -d om Cr op -d om Lv st -d om Fi sh -d om Cr op -d om Lv st -d om Fi sh -d om AOI (left axis) Agriculture expenditure, % of total expenditure (right axis) Research expenditure, % of agriculture expenditure (right axis) Pe rc en t AO I ( ra tio ) 20 The observation that the distribution of total agriculture expenditure across the subsectors is more equal to their relative contribution to the sector’s output (Figure 3), compared with the overall agriculture share in total government expenditure, could be due to data quality issues. In several countries, we found that the subsector shares were the same for all the years that the data were available (see section 3). This seems unrealistic, and we suspect that total GAE data may have been disaggregated proportionally to the subsector share in agriculture value added or some other unknown formula. This needs to be investigated to identify the actual reasons for the anomalies. In the econometric estimation below, however, we exclude countries with such unrealistic data. 21 5. DETERMINANTS AND EFFECTS OF AGRICULTURAL EXPPENDITURE This section presents the results of the econometric estimations. We estimate the system of equations at three levels: (1) aggregate for the entire agricultural sector, (2) research, and (3) subsector—crops, livestock, fisheries, and forestry. For the latter, the forestry subsector is dropped and used as the reference point. With fixed-effects methods, time-invariant variables drop out of the estimation. Here, this would apply to the variables on education, infrastructure, and irrigation (EDUCATE, INFRASTUR, and IRRIGATION) and agricultural transformation classification (AGTR-HHPD and AGTR-HHPI) because of how they are measured. To maintain them in the estimation, they were first interacted with the time trend variable. All variables with continuous values were first transformed by natural logarithm. 5.1. Sample size and summary statistics of the variables The data are on 24 countries (Benin, Burkina Faso, Burundi, Cameroon, Cote d’Ivoire, Egypt, Ethiopia, Gabon, Gambia, Ghana, Guinea, Kenya, Madagascar, Malawi, Mali, Mozambique, Namibia, Rwanda, Senegal, Sudan, Tanzania, Togo, Tunisia, and Uganda), each with a minimum of five years’ time series data, for a total of 132 observations. Table 11 presents summary statistics of all the variables used in the econometric estimation. The annual data on government expenditure and agricultural value added by subsector for these countries are presented in the Annex. For the contribution or share of government agricultural subsector expenditure to the total agricultural capital stock (AGCAPSTK), we calculate the value of 𝜋𝜋𝑠𝑠 using information at the continental level. They include agricultural investment from various sources—on-farm, government, public research, ODA, and foreign direct investment (Lowder et al. 2015); composition of agricultural capital stock—land, livestock, and machinery (Lenné and Thomas 2006); composition of the value of natural capital—cropland, pasture, fisheries, mangrove, and forestry (AfDB 2023); and type of machinery in use—tractors, soil machines, harvesters, milking machines, and so on (FAO 2023). First, we used the data from Lowder et al. (2015) to estimate 𝜋𝜋𝑠𝑠 for total agriculture and research expenditure to obtain 𝜋𝜋𝑠𝑠𝑡𝑡𝑠𝑠 as 0.1120 and 𝜋𝜋𝑟𝑟𝑟𝑟𝑠𝑠 as 0.0238. Then, we used data from the other sources to determine the share of crops, livestock, fisheries, and forestry in the value of total agricultural capital, which we multiplied by 𝜋𝜋𝑠𝑠𝑡𝑡𝑠𝑠 to obtain 𝜋𝜋𝑐𝑐𝑟𝑟𝑐𝑐 as 0.0239, 𝜋𝜋𝑙𝑙𝑙𝑙𝑙𝑙 as 0.0691, 𝜋𝜋𝑓𝑓𝑙𝑙𝑠𝑠 as 0.0005, and 𝜋𝜋𝑓𝑓𝑡𝑡𝑟𝑟 as 0.0185. 5.2. Determinants of government agriculture expenditure The GAE determinants are the results from the estimation of equation 2 in the system of equations in the empirical model. We present the results starting with those for government total agriculture expenditure, followed by government agriculture research expenditure and then government subsector agriculture expenditure (crops, livestock, and fisheries, with forestry as the reference subsector). 22 Table 11 Summary statistics of the variables used in the econometric estimations, 2014–2020 Variable Mean Standard error Agricultural output per hectare AGVAy_tot 234.20 32.42 AGVAy_crp 554.28 58.90 AGVAy_liv 120.77 16.09 AGVAy_fis 269.13 41.16 Government agriculture expenditure share of total GAEsh_tot 6.73 0.36 GAEsh_res 0.46 0.05 GAEsh_crp 59.23 1.80 GAEsh_liv 17.33 1.14 GAEsh_fis 10.39 1.03 Government agriculture expenditure intensity GAEint_tot 8.18 0.75 GAEint_res 0.57 0.06 GAEint_crp 7.66 0.62 GAEint_liv 7.75 0.86 GAEint_fis 12.14 1.43 Government agriculture orientation index GAEaoi_tot 0.34 0.02 GAEaoi _res 0.03 0.00 GAEaoi _crp 0.95 0.03 GAEaoi _liv 1.11 0.09 GAEaoi _fis 1.66 0.16 Explanatory variable TOTGEXP 349.76 33.86 AGCAPSTK 76.56 3.52 AGVAx_tot 21.24 0.67 AGVAx_crp 64.37 1.63 AGVAx_liv 20.04 1.15 AGVAx_fis 9.17 1.04 GOVERN −0.54 0.03 NETODA 6.50 0.46 AGLNDLAB 29.45 6.11 RUPOPGR 1.51 0.08 EDUCATE 8.71 0.20 LIFEEXP 63.90 0.37 INFRASTUR 2.40 0.06 IRRIGATION 0.24 0.07 RAINFALL 935.82 42.76 LOIRRIG*RAIN 1,050.07 39.39 AGTR-HHPD 1.09 0.19 AGTR-HHPI 1.93 0.22 TIMETRND 4.42 0.15 Source: Econometric model estimation results. Note: See Table 9 for detailed description of the variables. For the variables and data disaggregated by subsector, forestry is omitted as the reference subsector. Data are on 24 countries and 132 annual observations. 23 Government total agriculture expenditure Table 12 shows detailed results for the three measures of governmental total agriculture expenditure: as share of total government expenditure for the entire economy (GAEsh_tot), relative to the sector’s GDP (GAEint_tot), and ratio of GAE’s share of total expenditure to agriculture’s share of total GDP (GAEaoi_tot). The overall model fit results for the three measures are statistically significant at 1 percent, with similar r-squared values of 0.3. The performance of the estimated coefficients is similar for GAEsh_tot and GAEaoi_tot, in terms of those that are statistically significant at 1 or 5 percent. Here, the variables of statistical significance at 1 or 5 percent are lag of output per hectare, education, and life expectancy, all of which are negatively associated with GAEsh_tot and GAEaoi_tot. These results suggest that rising agricultural land productivity is associated with a reduction in government total spending on the sector. The negative association with respect to an increase in education and life expectancy, both of which are indicators of increase in income and wealth, is also expected, as governments may shift focus to nonagricultural sources of growth and development. Land–labor ratio, rural population, infrastructure, and irrigation, on the other hand, are positively associated with GAEsh_tot and GAEaoi_tot, as expected. By reducing the transactions cost and improving access to remote areas, improvements in infrastructure, for example, help governments and politicians to increase their expenditure on agriculture to reach farmers, in return for their votes (Olson 1965). Similarly, farmers are more able to reach politicians to lobby and demand more spending on the sector. When government total agriculture expenditure is measured relative to the sector’s GDP (GAEint_tot), then total government expenditure, agriculture value added share in GDP, and governance are the most important determinants, in addition to lag of output per hectare and irrigation, as discussed above. Total government expenditure and agriculture value added share in GDP have a positive association, whereas governance has a negative association. The positive association with total government expenditure indicates that spending on agriculture increases as the government’s total budget increases, which is likely the same for all other sectors in the economy. However, the increase in spending on agriculture is much lower than the increase in total government expenditure (estimated coefficient is 0.025), which is consistent with the trend analysis results presented earlier, in which government spending on agriculture relative to the total budget is declining over time. The negative association with governance is interesting, suggesting that government spending on agriculture is likely more corrupt compared with spending on other sectors (De la Croix and Delavallade 2009). Thus, by reducing inefficiencies or removing excesses in expenditures, improvement in governance4 reduces spending on the sector. 4 Governance is measured as an average of six worldwide governance indicators—control of corruption, government effectiveness, political stability and absence of violence/terrorism, regulatory quality, rule of law, and voice and accountability (World Bank 2023). 24 Table 12 Factors affecting government agriculture expenditure (GAE), 2014–2020 Variable GAE measured as: GAE measured as: Share of total government expenditure (GAEsh_tot) Share of agriculture value added (GAEint_tot) Ratio of expenditure share to value-added share (GAEaoi_tot) Share of total government expenditure (GAEsh_res) Share of agriculture value added (GAEint_res) Ratio of expenditure share to value-added share (GAEaoi_res) Lag of dependent variable 0.077 −0.024 0.064 0.189 ** 0.083 0.186 ** TOTGEXP 0.146 0.359 *** 0.138 Lag of GAEsh_tot 0.214 * 0.064 0.225 * AGCAPSTK 0.498 0.326 0.529 −0.105 −0.198 −0.074 Lag of AGVAy_tot −0.853 *** −1.303 *** −0.897 *** −0.340 −0.903 ** −0.417 AGVAx_tot 0.001 0.061 ** −0.016 −0.078 ** −0.021 −0.089 ** GOVERN −0.639 −1.392 *** −0.714 * 1.773 *** 1.072 ** 1.718 *** NETODA −0.015 0.009 −0.021 * −0.016 0.006 −0.024 AGLNDLAB 0.904 ** 0.449 0.841 ** −1.362 ** −1.934 *** −1.395 ** RUPOPGR 0.372 ** 0.117 0.293 * 0.000 −0.238 −0.066 EDUCATE −0.091 *** −0.052 −0.091 *** −0.035 −0.001 −0.036 LIFEEXP −8.696 ** 7.206 −8.461 ** −5.534 8.884 ** −5.687 INFRASTUR 0.128 *** 0.013 0.123 *** 0.137 *** 0.045 0.138 *** IRRIGATION 0.135 *** 0.178 *** 0.122 *** −0.059 −0.028 −0.067 RAINFALL −0.969 −0.741 −1.040 −1.203 −1.153 −1.308 LOIRRIG*RAIN 1.017 1.068 1.104 1.056 1.250 1.173 AGTR-HHPD 0.107 *** 0.147 *** 0.126 *** −0.065 −0.015 −0.040 AGTR-HHPI −0.040 0.041 −0.038 −0.022 0.052 −0.021 Intercept 0.000 0.000 0.000 0.000 0.000 0.000 Overall model fit: R-squared 0.38 0.36 0.39 0.25 0.31 0.27 F-test 4.42 *** 4.04 *** 4.63 *** 2.44 *** 3.32 *** 2.71 *** Source: Econometric model estimation results. Note: See Table 9 for a detailed description of the variables. Data are on 24 countries and 132 annual observations. * p < .10; ** p < .05; *** p < .01. The group of countries with agricultural transformation classified as having high initial agricultural employment and GDP shares and declining labor productivity (AGTR-HHPD— including Burundi, Gambia, Ghana, Madagascar, and Mauritania) is positively associated with all three measures of government total agriculture expenditure. Recognizing the importance of the agricultural sector to the overall economy but untapped potential in these countries, the governments there spend relatively more on the sector compared with those of the other groups, which makes sense. Government agriculture research expenditure Table 12 presents detailed results for the determinants of government agriculture research expenditure. There are differences in the variables that have a statistically significant association with government agriculture research expenditure compared with those of government total agriculture expenditure, as discussed above. This seems consistent with the notion that the total allocation for the sector is first made, followed by the allocation to the various subcomponents. The variables with a 1 or 5 percent statistical significance are lag of research expenditure (positive effect), lag of output per hectare (negative effect), lag of agriculture value-added share 25 in GDP (negative effect), governance (positive effect), agricultural land–labor ratio (negative effect), life expectancy (positive effect), and infrastructure (positive effect). As with the determinants of government total agriculture expenditure, there are some differences with the statistically significant variables across the research expenditure measures. The lag research expenditure, lag of agriculture value-added share in GDP, and infrastructure are statistically significant in both GAEsh_res and GAEaoi_res only, whereas lag of output per hectare and life expectancy are statistically significant in GAEint_res only. Governance and agricultural land– labor ratio are statistically significant in all three measures. Past spending on agriculture research also seems important, although it is only statistically significant at 10 percent for GAEsh_res and GAEaoi_res only. Of these results, two stand out the most (lag of agriculture value added share in GDP, agricultural land–labor ratio, and governance) in terms of having an opposite effect on government agriculture research expenditure when compared with their effect on government total agriculture expenditure. The negative effect of the lag of agriculture value-added share in GDP is consistent with the “small-country problem” (Fuglie and Rada 2013; Ruttan 1982), as countries with larger agricultural GDP shares tend to be those with lower resources or returns to investment to justify larger investment in agricultural research. The negative effect of the agricultural land–labor ratio suggests that agricultural research expenditure increases with rising scarcity of agricultural land relative to labor, which likely reflects differences in the complexity of research solutions and technologies in land-abundant versus land-scarce agriculture. It is consistent with the results of Judd et al. (1986) that costly land expansion induces expansion in agricultural research and extension expenditure. The positive effect of governance, which is statistically significant in all three measures of agriculture research expenditure, seems to suggest that agriculture research expenditure is less corrupt compared with spending on other functions, such as fertilizer subsidies (Omuru and Kigwell 2006). Thus, improvements in governance that seek to reduce inefficiencies in spending tend to benefit agriculture research expenditure more than the other functions. Government agriculture subsector expenditure Table 13 shows details of the regression results on the factors affecting the subsector composition (crops, livestock, and fisheries, with forestry as the reference subsector) of GAE for the three expenditure measures. The allocation of government agriculture expenditure by subsector strongly depends on past spending on the subsector. The estimated coefficients are positive and statistically significant at 1 percent for all the subsectors and the three expenditure measures. Other variables that are statistically significant (mostly at 5 percent) have mixed effects, either statistically significant in some subsectors but not others or statistically significant with some expenditure measures but not others. This means that once the total agriculture expenditure budget is decided, the subsector allocations mostly follow past spending, with the fisheries subsector having the largest dependency, followed by crops and livestock. 26 Table 13 Factors affecting subsector composition of government agriculture expenditure (GAE), 2014–2020 Variable Subsector expenditure share of GAE Subsector expenditure as share of subsector value added Ratio of subsector expenditure share to subsector value added share Crops (GAEsh_crp) Livestock (GAEsh_liv) Fisheries (GAEsh_fis) Crops (GAEint_crp) Livestock (GAEint_liv) Fisheries (GAEint_fis) Crops (GAEaoi_crp) Livestock (GAEaoi_liv) Fisheries (GAEaoi_fis) Lag of dependent variable 0.296 *** 0.282 *** 0.416 *** 0.380 *** 0.269 *** 0.455 *** 0.305 *** 0.264 *** 0.435 *** Lag of GAEsh_tot −0.117 −0.108 0.500 *** −0.421 *** −0.449 *** −0.056 −0.012 −0.120 0.443 *** AGCAPSTK −0.121 −0.610 0.534 −0.203 −0.512 0.441 −0.272 −0.687 0.649 Lag of AGVAy_crp −0.012 −0.970 *** 0.115 AGVAx_crp 0.008 0.029 ** −0.008 Lag of AGVAy_liv 0.096 −0.968 ** −0.195 AGVAx_liv −0.089 *** −0.056 −0.103 *** Lag of AGVAy_fis −0.001 −0.284 0.078 AGVAx_fis 0.047 * 0.022 0.018 GOVERN −0.075 0.244 −0.602 −0.894 * −0.505 −1.569 * 0.065 0.691 −0.481 NETODA 0.044 ** −0.033 * −0.052 ** 0.010 −0.044 ** −0.058 ** 0.012 * −0.041 ** −0.052 ** AGLNDLAB −0.298 −0.267 −0.490 −0.209 −0.227 −0.079 −0.507 ** −0.610 −0.477 RUPOPGR 0.229 −0.028 −0.415 0.097 0.143 −0.513 0.000 0.022 −0.524 EDUCATE −0.072 −0.094 * 0.135 ** −0.096 ** −0.142 ** 0.045 −0.035 * −0.087 * 0.107 * LIFEEXP 3.608 −0.668 0.506 6.305 3.926 −2.932 2.770 −0.936 −2.561 INFRASTUR 0.030 0.023 −0.020 0.039 0.056 −0.019 0.004 0.049 −0.010 IRRIGATION −0.105 ** −0.071 0.092 0.043 0.039 0.129 −0.063 *** −0.068 0.050 RAINFALL −0.240 −0.060 0.164 −0.676 −0.861 −0.256 0.113 0.110 0.560 LOIRRIG*RAIN 0.154 −0.083 −0.796 0.887 1.074 −0.071 −0.178 −0.162 −1.168 AGTR-HHPD −0.078 0.044 −0.136 0.095 * 0.116 * 0.002 −0.020 0.023 −0.092 AGTR-HHPI 0.047 0.113 ** 0.016 0.056 0.148 ** 0.081 0.021 0.117 ** 0.040 Intercept 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Overall model fit: R-squared 0.34 0.30 0.32 0.39 0.31 0.35 0.44 0.38 0.34 F-test 3.93 *** 3.42 *** 4.14 *** 4.91 *** 4.02 *** 3.94 *** 5.75 *** 5.00 *** 3.95 *** Source: Econometric model estimation results. Note: See Table 9 for a detailed description of the variables. Data are on 24 countries and 132 annual observations. * p < .10; ** p < .05; *** p < .01. 27 Regarding the other variables, the major ones (those that are statistically significant in more than one equation) are lag of total agriculture expenditure, lag of subsector output per hectare, share of subsector in agriculture GDP (agGDP), net ODA, education, irrigation, and the group of countries with agricultural transformation classified as having high initial agricultural employment and GDP shares and increasing labor productivity (AGTR-HHPI). The lag of total agriculture expenditure, for example, is positively associated with expenditure on fisheries (GAEsh_fis and GAEaoi_fis) and negatively associated with expenditure on crops (GAEint_crp) and livestock (GAEint_liv). Net ODA is positively associated with expenditure on crops (GAEsh_crp and GAEaoi_crp) and negatively associated with expenditure on livestock (all three measures) and fisheries (all three measures). This suggests that ODA complements government expenditure on the crops subsector but crowds out government expenditure on the livestock and fisheries subsectors. Education is positively associated with government expenditure on fisheries (GAEsh_fis and GAEaoi_fis) and negatively associated with government expenditure on crops (GAEint_crp and GAEaoi_crp) and livestock (all three measures). This seems to suggest that the fisheries subsector may have more people with higher levels of education, which, in addition to the fisheries subsector having more holistic cognition (Uskul et al. 2008), is a key factor in lobbying governments and politicians (Binswanger and Deininger 1997). Irrigation is negatively associated with government expenditure on crops (GAEsh_crp and GAEaoi_crp) only, whereas the group of countries with agricultural transformation classified as HHPI is positively associated with government expenditure on livestock (all three measures) only. The negative effect of irrigation on government expenditure on crops, especially since irrigation has a positive effect on government total agriculture expenditure (see Table 12). The data show that other subsectors, especially livestock and fisheries, are also important in countries where irrigation development is higher. Comparing the subsector agriculture expenditure for countries with higher irrigation development to those with lower irrigation, the data show that the share of agriculture expenditure for crops (GAEsh_crp) is lower by about 5 percentage points in the countries where irrigation development is higher. On the other hand, the share of agriculture expenditure for livestock (GAEsh_liv) is higher by about 8 percentage points in the countries where irrigation development is higher. 5.3. Effects of the composition of government agriculture expenditure on agricultural land productivity The effects of the composition of GAE on agricultural land productivity are the results from the estimation of equation 1 in the system of equations presented in the section on the empirical model. We present the results of the effects on total agricultural land productivity starting with those for government total agriculture expenditure (details in Table 14), followed by government agriculture research expenditure (Table 14) and then government subsector agriculture expenditure (crops, livestock, and fisheries, with forestry as the reference subsector) (Table 15). For government agriculture subsector expenditure, we also present results of the effects on their subsector productivity, without cross-subsector expenditure effects (Table 16) and with cross- subsector expenditure effects (Table 17). The statistically significant effects of other variables are also presented. 28 Effect of government total agriculture expenditure Government total agriculture expenditure has a statistically significant direct effect on total agricultural output per unit area (land productivity) as well as an indirect effect via agricultural capital stock, with a total effect or elasticity of 0.06 to 0.08 that is statistically significant at 1 percent for all three expenditure measures (Table 14). This means that a 1 percent increase in government total agriculture expenditure is associated with a 0.06 to 0.08 percent increase in total agricultural land productivity, with the effect being higher when measured as an orientation index (GAEaoi_tot) or share of total government expenditure (GAEsh_tot) as opposed to share of agriculture added (GAEint_tot). These results are similar to the findings of the few comparable cross-country studies, including those of Fan et al. (2008), which estimate the elasticity at 0.08 using data on 44 developing countries, including 17 African countries. Other variables that have a statistically significant effect on total agricultural land productivity are lag of total agricultural land productivity, governance, life expectancy, infrastructure, irrigation, and rainfall, all of which have an expected positive effect. The statistical significance of rainfall is relatively weak, at only 10 percent. The lag of the size of the sector as a share of GDP (AGVAx_tot) has a negative effect, which suggests more intensive production systems with declining size of the sector and is consistent with the inverse farm size–productivity relationship that is well established in the literature (Hazell et al. 2010). The inverse interaction between irrigation and rainfall (LOIRRIG*RAIN) has a negative effect, which depicts a lower agricultural land productivity in countries with moderate to high annual rainfall but low irrigation development, compared with those with low annual rainfall areas but high irrigation development. The average difference in total agricultural land productivity between the groups is about 40 percent. Effect of government agriculture research expenditure The estimated effect of government agriculture research expenditure on total agricultural land productivity is mixed. The direct effect is not statistically significant, resulting in a total effect or elasticity of 0.02 that is statistically significant at 10 percent for only two expenditure measures (GAEsh_res and GAEaoi_res). These results are close to the findings of Fan et al. (2008), which estimate the elasticity at 0.04 using data on 44 developing countries, including 17 African countries. The estimates are lower, however, than those from other studies that estimate the effect of overall public agriculture research expenditure on either agricultural land or labor productivity, such as those of Thirtle et al. (2003), which estimate the elasticity at 0.26 to 0.36 using data on 22 African countries south of the Sahara. Other variables that have a statistically significant effect on total agricultural land productivity are the same as presented above, except for rainfall, where the effect was weak but is now no longer statistically significant. 29 Table 14 Effect of government agriculture expenditure (GAE) on total agriculture value added per unit area, 2014-2020 Variable GAEsh_tot GAEint_tot GAEaoi_tot GAEsh_res GAEint_res GAEaoi_res GAEsh_tot 0.042 ** GAEint_tot 0.029 * GAEaoi_tot 0.046 ** GAEsh_res 0.016 GAEint_res 0.019 GAEaoi_res 0.017 AGCAPSTK 0.295 ** 0.290 ** 0.291 ** 0.304 ** 0.300 ** 0.302 ** Lag of AGVAy_tot 0.582 *** 0.618 *** 0.586 *** 0.571 *** 0.595 *** 0.573 *** AGVAx_tot −0.015 *** −0.016 *** −0.013 ** −0.015 *** −0.015 *** −0.014 *** GOVERN 0.348 *** 0.334 *** 0.351 *** 0.293 *** 0.291 *** 0.292 *** NETODA −0.004 −0.004 −0.004 −0.003 −0.003 −0.003 AGLNDLAB 0.038 0.072 0.035 0.092 0.106 0.092 RUPOPGR −0.028 −0.014 −0.027 −0.022 −0.015 −0.021 EDUCATE 0.001 −0.001 0.001 −0.002 −0.002 −0.002 LIFEEXP 2.447 *** 1.789 * 2.466 *** 2.455 *** 2.119 ** 2.463 *** INFRASTUR 0.014 * 0.016 ** 0.014 * 0.015 * 0.015 ** 0.015 * IRRIGATION 0.019 ** 0.017 * 0.019 ** 0.021 ** 0.020 ** 0.021 ** RAINFALL 0.275 * 0.268 * 0.279 * 0.248 0.252 0.250 LOIRRIG*RAIN −0.397 ** −0.383 ** −0.402 ** −0.359 ** −0.359 ** −0.360 ** AGTR-HHPD −0.008 −0.008 −0.008 −0.006 −0.006 −0.006 AGTR-HHPI 0.002 −0.001 0.002 0.001 0.000 0.001 Intercept 0.000 0.000 0.000 0.000 0.000 0.000 Overall model fit: R-squared 0.89 0.89 0.89 0.89 0.88 0.89 F-test 61.75 *** 60.33 *** 61.96 *** 59.71 *** 59.58 *** 59.75 *** Total effect of GAE† 0.075 *** 0.062 *** 0.079 *** 0.023 * 0.026 0.024 * Source: Econometric model estimation results. Note: See Table 9 for a detailed description of the variables. † Estimated total effect of GAE = ∂AGVAy_tot/∂GAE_j + πj*∂AGVAy_tot /∂AGCAPSTK, where πj is 0.112 and 0.024 for total and research expenditure, respectively. Data are on 24 countries and 132 annual observations. * p < .10; ** p < .05; *** p < .01. Effect of government agriculture subsector expenditure Results of the regression presented in Table 15 show that the total (direct and indirect) effect or elasticity is estimated at 0.03 to 0.09 for crops and 0.02 to 0.03 for fisheries, which are statistically significant mostly at 1 or 5 percent for all three expenditure measures. The estimated total effect for livestock is not statistically significant. Other variables that have a statistically significant effect on total agricultural land productivity are the same as those presented in the discussion on the effect of government total agriculture expenditure. The results on the effect of government agriculture subsector expenditure on agricultural subsector productivity, without cross-subsector expenditure effects (Table 16) and with cross- subsector expenditure effects (Table 17), are interesting. Looking first at the results without the cross-subsector expenditure effects, we see that the direct effects are not statistically significant in any of the subsectors. Also, agricultural capital stock is not statistically significant in the fisheries subsector equations. The estimated total (direct and indirect) effect is statistically significant for government expenditure on livestock only, which is 0.05 for all three expenditure measures. When estimating the cross-subsector expenditure effects, government expenditure on 30 fisheries influences productivity in the crops and livestock subsectors, resulting in an estimated total effect of 0.08 to 0.09, which is statistically significant for all three expenditure measures. The effect of government expenditure on livestock is statically significant for the own effect only at 0.4 to 0.7. Table 15 Effect of government agriculture subsector expenditure on total agriculture value added per unit area, 2014–2020 Variable GAEsh GAEint GAEaoi GAEsh_crp 0.024 GAEsh_liv 0.002 GAEsh_fis 0.028 ** GAEint_crp 0.030 * GAEint_liv -0.011 GAEint_fis 0.022 ** GAEaoi_crp 0.087 ** GAEaoi_liv 0.003 GAEaoi_fis 0.025 ** AGCAPSTK 0.293 ** 0.262 ** 0.305 ** Lag of AGVAy_tot 0.574 *** 0.633 *** 0.545 *** AGVAx_tot −0.017 *** −0.015 *** −0.017 *** GOVERN 0.314 *** 0.366 *** 0.319 *** NETODA −0.003 −0.004 −0.004 AGLNDLAB 0.044 0.052 0.080 RUPOPGR −0.021 −0.017 −0.017 EDUCATE −0.004 0.000 −0.002 LIFEEXP 1.996 ** 1.459 1.858 ** INFRASTUR 0.017 ** 0.016 ** 0.019 ** IRRIGATION 0.021 ** 0.017 * 0.024 *** RAINFALL 0.255 * 0.300 ** 0.278 * LOIRRIG*RAIN −0.369 ** −0.411 *** −0.387 ** AGTR-HHPD −0.006 −0.010 −0.005 AGTR-HHPI −0.002 −0.003 −0.002 Intercept 0.000 0.000 0.000 Overall model fit: R-squared 0.89 0.89 0.89 F-test 55.81 *** 56.50 *** 56.61 *** Total effect of GAEi_crp† 0.031 * 0.036 ** 0.094 *** Total effect of GAEi_liv† 0.023 0.007 0.024 Total effect of GAEi_fis† 0.028 ** 0.022 ** 0.025 ** Source: Econometric model estimation results. Note: See Table 9 for a detailed description of the variables. † Estimated total effect of GAE = ∂AGVAy_tot/∂GAE_j + πj*∂AGVAy_tot /∂AGCAPSTK, where πj is 0.0239, 0.0691, and 0.0005 for crops, livestock, and fisheries expenditure, respectively. Data on 24 countries and 132 annual observations. * p < .10; ** p < .05; *** p < .01. 31 Table 16 Effect of agriculture subsector expenditure on subsector value added per unit area without cross-subsector spending effects, 2014–2020 Variable Effect of subsector expenditure as share of GAE Effect of subsector expenditure as share of subsector value added Effect of ratio of subsector expenditure share to subsector value added share Crops AGVAy_crp Livestock AGVAy_liv Fisheries AGVAy_fis Crops AGVAy_crp Livestock AGVAy_liv Fisheries AGVAy_fis Crops AGVAy_crp Livestock AGVAy_liv Fisheries AGVAy_fis GAEsh_crp -0.012 GAEsh_liv 0.017 GAEsh_fis 0.026 GAEint_crp 0.028 GAEint_liv 0.013 GAEint_fis 0.023 GAEaoi_crp 0.021 GAEaoi_liv 0.022 GAEaoi_fis 0.025 AGCAPSTK 0.347 ** 0.491 *** 0.294 0.332 ** 0.464 *** 0.290 0.340 ** 0.508 *** 0.284 Lag of AGVAy_i 0.393 *** 0.563 *** 0.404 *** 0.409 *** 0.573 *** 0.407 *** 0.369 *** 0.549 *** 0.426 *** AGVAx_i −0.013 *** −0.004 0.017 *** −0.013 *** −0.005 0.018 *** −0.012 *** −0.003 0.017 *** GOVERN −0.047 −0.145 0.279 0.015 -0.124 0.310 * −0.019 −0.146 0.285 NETODA 0.000 0.011 *** 0.004 −0.001 0.011 *** 0.003 0.000 0.011 *** 0.004 AGLNDLAB 0.171 0.196 0.083 0.159 0.173 0.076 0.167 0.195 0.087 RUPOPGR 0.107 * 0.064 0.089 0.116 ** 0.071 0.100 0.112 ** 0.074 0.086 EDUCATE 0.026 ** −0.031 *** 0.010 0.028 *** −0.031 *** 0.012 0.027 ** −0.032 *** 0.010 LIFEEXP 3.799 *** 3.638 *** 5.830 *** 3.068 ** 3.462 *** 5.613 *** 3.541 *** 3.669 *** 5.665 *** INFRASTUR 0.020 * 0.020 * 0.007 0.023 ** 0.021 ** 0.009 0.023 ** 0.021 ** 0.006 IRRIGATION 0.066 *** 0.026 ** 0.054 *** 0.065 *** 0.023 ** 0.055 *** 0.067 *** 0.026 ** 0.053 *** RAINFALL 0.254 0.150 −0.137 0.284 0.169 −0.119 0.264 0.147 −0.132 LOIRRIG*RAIN −0.328 −0.373 * 0.053 −0.365 −0.394 * 0.033 −0.342 −0.370 * 0.050 AGTR-HHPD −0.030 ** 0.018 −0.019 −0.030 ** 0.017 −0.021 −0.028 ** 0.019 −0.020 AGTR-HHPI −0.012 −0.010 −0.025 * −0.013 −0.009 −0.026 * −0.012 −0.009 −0.025 * Intercept 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Overall model fit: R-squared 0.78 0.81 0.73 0.79 0.81 0.73 0.78 0.82 0.73 F-test 27.73 *** 34.24 *** 21.37 *** 28.18 *** 34.07 *** 21.10 *** 27.77 *** 34.34 *** 21.35 *** Total effect of GAE† -0.003 0.051 ** 0.026 0.036 * 0.046 ** 0.022 0.029 0.057 ** 0.025 Source: Econometric model estimation results. Note: See Table 9 for a detailed description of the variables. † Estimated total effect of GAE = ∂AGVAy_j/∂GAE_j + πj*∂AGVAy_j /∂AGCAPSTK, where πj is 0.0239, 0.0691, and 0.0005 for crops, livestock, and fisheries expenditure, respectively. Data are on 24 countries and 132 annual observations. * p < .10; ** p < .05; *** p < .01. 32 Table 17 Effect of agriculture subsector expenditure on subsector value added per unit area with cross-subsector spending effects, 2014– 2020 Variable Effect of subsector expenditure as share of GAE Effect of subsector expenditure as share of subsector value added Effect of ratio of subsector expenditure share to subsector value added share Crops AGVAy_crp Livestock AGVAy_liv Fisheries AGVAy_fis Crops AGVAy_crp Livestock AGVAy_liv Fisheries AGVAy_fis Crops AGVAy_crp Livestock AGVAy_liv Fisheries AGVAy_fis GAEsh_crp 0.001 0.042 * 0.020 GAEsh_liv −0.014 0.029 0.019 GAEsh_fis 0.027 0.027 * 0.032 GAEint_crp 0.017 −0.007 −0.027 GAEint_liv −0.015 0.007 −0.002 GAEint_fis 0.035 ** 0.014 0.031 GAEaoi_crp 0.088 * 0.108 ** −0.036 GAEaoi_liv 0.002 0.033 * 0.001 GAEaoi_fis 0.039 ** 0.029 * 0.027 AGCAPSTK 0.299 * 0.509 *** 0.337 0.300 * 0.450 *** 0.283 0.355 ** 0.556 *** 0.291 Lag of AGVAy_i 0.404 *** 0.523 *** 0.395 *** 0.423 *** 0.577 *** 0.394 *** 0.331 *** 0.489 *** 0.420 *** AGVAx_i −0.013 *** −0.004 0.017 *** −0.013 *** −0.006 0.020 *** −0.011 ** −0.001 0.019 *** GOVERN −0.018 −0.128 0.274 0.054 −0.109 0.278 0.009 −0.136 0.267 NETODA 0.000 0.011 *** 0.004 −0.002 0.011 *** 0.004 −0.001 0.010 *** 0.004 AGLNDLAB 0.126 0.144 0.084 0.124 0.157 0.058 0.150 0.194 0.068 RUPOPGR 0.100 * 0.083 0.095 0.099 * 0.063 0.095 0.112 ** 0.100 * 0.092 EDUCATE 0.025 ** −0.033 *** 0.008 0.027 ** −0.032 *** 0.009 0.025 ** −0.033 *** 0.008 LIFEEXP 3.674 *** 3.248 *** 5.535 *** 3.085 ** 3.497 *** 6.061 *** 3.301 ** 3.268 *** 5.986 *** INFRASTUR 0.020 * 0.025 ** 0.008 0.021 ** 0.020 * 0.006 0.024 ** 0.026 ** 0.005 IRRIGATION 0.064 *** 0.030 *** 0.058 *** 0.062 *** 0.022 ** 0.056 *** 0.071 *** 0.034 *** 0.054 *** RAINFALL 0.266 0.167 −0.136 0.302 0.174 −0.145 0.296 0.182 −0.149 LOIRRIG*RAIN −0.338 −0.398 * 0.048 −0.377 * −0.396 * 0.063 −0.366 −0.401 * 0.067 AGTR-HHPD −0.033 ** 0.020 * −0.016 −0.032 ** 0.016 −0.023 −0.027 ** 0.021 * −0.022 AGTR-HHPI −0.012 −0.012 −0.027 * −0.014 −0.009 −0.024 * −0.014 −0.013 −0.024 * Intercept 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Overall model fit: R-squared 0.79 0.82 0.73 0.79 0.82 0.73 0.79 0.82 0.73 F-test 25.60 *** 31.45 *** 19.15 *** 26.19 *** 30.65 *** 18.88 *** 26.19 *** 32.50 *** 19.13 *** Total effect of GAE† 0.070 0.070 0.086 ** −0.009 0.021 0.080 ** 0.169 0.074 * 0.094 ** Own effect of GAE† 0.008 0.064 ** 0.032 0.024 0.038 * 0.031 0.097 * 0.072 *** 0.027 Cross effect of GAE† 0.062 0.005 0.054 ** −0.033 −0.017 0.049 ** 0.072 0.003 0.068 *** Source: Econometric model estimation results. Note: See Table 9 for a detailed description of the variables. † Estimated total effect of GAE is own effect and cross effects: own effect = ∂AGVAy_j/∂GAE_j + πj*∂AGVAy_j /∂AGCAPSTK, and cross effect = sum of ∂AGVAy_j/∂GAE_i ∀i≠j, where is πj is 0.0239, 0.0691, and 0.0005 for crops, livestock, and fisheries expenditure, respectively. Data on 24 countries and 132 annual observations. * p < .10; ** p < .05; *** p < .01. 33 The effect of the other variables on subsector productivity is also interesting. Compared with their effect on total agriculture productivity in the earlier findings, those that have the same sign across the various subsectors and expenditure measures are lag of the dependent variable or subsector productivity, life expectancy, and irrigation, all of which have a positive effect. The inverse size–productivity relationship continues to hold for the crops subsector but not for the livestock subsector and is opposite for the fisheries subsector. Governance is not statistically significant, and net ODA has a positive effect on livestock. Rural population has a positive effect on crops only. Education has a positive effect on crops but a negative effect on livestock and is neutral on fisheries. Infrastructure has a positive effect on crops and livestock only. The group of countries with an agricultural transformation classification of HHPD has a negative effect on crops and a negative effect on livestock, whereas those with a classification of HHPI have a negative effect on fisheries. 34 6. CONCLUSIONS AND IMPLICATIONS GAE measured relative to total government expenditure or agriculture value added has been declining in Africa since the 1980s. This is concerning because public agriculture expenditure has high returns in terms of growth, and agricultural growth may have been more effective at reducing poverty than growth originating from other sectors. However, a critical question relates to the quality of the expenditure and the returns to different types of public agriculture expenditure, which may help in convincing policymakers to increase the share of the national budget allocated to agriculture. Using cross-country annual data from 2014 to 2020, this paper analyzed the effect of recent trends in different types of GAE as well as the political economy and other factors associated with the composition of GAE, considering three measures of expenditure—share of total expenditure, share of agriculture value added, and orientation index. By integrating the estimation of the determinants of expenditure with the effect of spending on growth, paper’s contribution is unique to the extent that it discerns the underlying causative process of government expenditure. Analysis of the trends show a wide cross-country variation in GAE in Africa, with few African countries having achieved the 10 percent CAADP expenditure target. Overall, at the sector level, most African countries spend much smaller proportions of their public budget on agriculture than the sector’s share in the economy. At the subsector level, the crops subsector dominates, with an average of 58 and 63 percent of agriculture expenditure and value added, respectively, followed by livestock (17 and 19 percent, respectively), forestry (15 and 11 percent, respectively), and fisheries (10 and 7 percent, respectively). These findings suggest that although agriculture’s share in total government expenditure is much less than the sector’s contribution to GDP, the distribution of total agriculture expenditure across the subsectors is more equal to its relative contribution to the sector’s output, with forestry and fisheries being slightly favored compared with crops and livestock, which dominate