IFPRI Discussion Paper 02288 

October 2024 

Impacts of an Innovative Credit + 
Insurance Bundle for Marginalized Farmers 

Evidence from a Cluster Randomized Trial in Odisha, India 

Berber Kramer 

Subhransu Pattnaik 

Patrick S. Ward 

Yingchen Xu 

Markets, Trade, and Institutions Unit 



INTERNATIONAL FOOD POLICY RESEARCH INSTITUTE 
The International Food Policy Research Institute (IFPRI), a CGIAR Research Center established in 1975, 
provides research-based policy solutions to sustainably reduce poverty and end hunger and malnutrition. 
IFPRI’s strategic research aims to foster a climate-resilient and sustainable food supply; promote healthy 
diets and nutrition for all; build inclusive and efficient markets, trade systems, and food industries; 
transform agricultural and rural economies; and strengthen institutions and governance. Gender is 
integrated in all the Institute’s work. Partnerships, communications, capacity strengthening, and data and 
knowledge management are essential components to translate IFPRI’s research from action to impact. 
The Institute’s regional and country programs play a critical role in responding to demand for food policy 
research and in delivering holistic support for country-led development. IFPRI collaborates with partners 
around the world.  

AUTHORS 
Berber Kramer (b.kramer@cgiar.org) is a Senior Research Fellow in the Markets, Trade, and Institutions 
(MTI) Unit of the International Food Policy Research Institute (IFPRI), Nairobi, Kenya. 

Subhransu Pattnaik (subhransu.pattnaik@uni-goettingen.de) is a Research Analyst in IFPRI’s MTI Unit 
and a PhD candidate with the University of Göttingen, Germany.  

Patrick S. Ward (wardp@ufl.edu) is a Non-resident Fellow with IFPRI's Innovation and Policy Scaling 
Unit and an Associate Professor in the Food and Resource Economics Department of the University of 
Florida, Gainesville, FL. 

Yingchen Xu (yingchenxu@ufl.edu) is a PhD candidate in the Food and Resource Economics 
Department of the Universty of Florida, Gainesville, FL. 

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:b.kramer@cgiar.org
mailto:Subhransu.pattnaik@uni-goettingen.de
mailto:wardp@ufl.edu
mailto:yingchenxu@ufl.edu


iii 
 

Abstract 
 

Smallholder farmers often lack documented land rights to serve as collateral for formal loans, 
with livelihoods inextricably linked to weather conditions. Resulting credit and risk constraints 
prevent them from investing in their farms. We implemented a randomized evaluation of 
KhetScore, an innovative credit scoring approach that uses remote sensing to unlock credit and 
insurance for smallholders including landless farmers in Odisha, a state in eastern India. In our 
treatment group, where we offered KhetScore loans and insurance, farmers - and especially 
women - were more likely to be insured and borrow from formal sources without substituting 
formal for informal loans. Despite increased borrowing, treated households faced less difficulty 
in repaying loans, suggesting that insured KhetScore loans transferred risk and eased the burden 
of repayment. Moreover, the treatment enhanced agricultural profitability by increasing revenues 
during the monsoon season and reducing costs in the dry season. Positive and significant effects 
are found among both farmers with unconstrained baseline credit access, and quantity rationed 
farmers, suggesting that KhetScore helps address supply-side credit constraints. Finally, the 
treatment significantly enhanced women’s empowerment and mental health. In conclusion, 
remote sensing-enabled financial products can substantially improve landless farmers’ access to 
agricultural credit, risk management, resilience, and well-being. 

 

Keywords: credit, insurance, impact evaluations, gender parity, India 

  



iv 
 

Acknowledgements 
 

We thank Dvara E-Registry for working with us continuously and patiently to facilitate an 
impact evaluation of their innovation. We are grateful for funding, technical review, and support 
throughout the study from the International Initiative for Impact Evaluation (3ie). Additional 
funding was provided by the CGIAR Research Program on Policies, Institutions, and Markets 
(PIM), led by the International Food Policy Research Institute (IFPRI), and the CGIAR platform 
for Big Data in Agriculture. This work was undertaken as part of the CGIAR Research Initiative 
on Market Intelligence, which brings together strategic information on future crops, market 
segments and trait priorities aligned to the needs and preferences of farmers, agri-business and 
consumers. We would like to thank all funders who supported this research through their 
contributions to the CGIAR Trust Fund: https://www.cgiar.org/funders/. This paper has not gone 
through IFPRI’s standard peer review procedure. The opinions expressed here belong to the 
authors, and do not necessarily reflect those of 3ie, PIM, IFPRI, or CGIAR. 

https://www.cgiar.org/funders/


1 Introduction

In developing countries, financial inclusion and risk management are critical to fostering investments to

improve agricultural productivity and enhance rural livelihoods. Amid traditional barriers like limited access

to credit and vulnerability to agricultural risks, smallholder farmers often lack the wherewithal to expand

their operations or invest in profitable technologies and inputs. For smallholders with limited land holdings,

formal financial services like credit and insurance are often considered cost-prohibitive due to high transaction

and monitoring costs, along with limited available data on these farmers to identify the risks of insuring or

lending to these farmers. For instance, lacking documented land rights or lacking the necessary collateral

limits potential borrowers’ access to formal credit (Higgins et al., 2018), and consequently potential borrowers

are involuntarily limited in their borrowing, resulting in efficiency losses and potential entrapment in a low

productivity equilibrium (Croppenstedt et al., 2013). Even when individuals may have access to formal

credit, credit markets can be in a disequilibrium in which lenders restrict potential borrowers’ access to their

desired level of borrowed funds to finance agricultural investments (Stiglitz and Weiss, 1981). These forms

of supply-side credit rationing often arise due to information asymmetries between borrowers and lenders.

In particular, to the extent that lenders are unable to accurately assess the credit worthiness of potential

borrowers (e.g., if they do not have a credit report with a formal credit rating agency), lenders may restrict

their lending to avoid potential adverse selection. Further, since it is difficult to monitor borrowers’ actions

after a loan has been disbursed, lenders may restrict the amount lent to limit potential losses due to moral

hazard. In other cases, risk averse borrowers may voluntarily withdraw from the credit market, often because

they are afraid of losing their collateral if there is an adverse calamity and they are unable to repay the loan.

Such “risk rationing’ ’ (Binswanger and Sillers, 1983; Boucher et al., 2008) occurs in any situation in which

the lender shifts so much contract risk onto the borrower (in the form of excessive collateral requirements,

high interest rates, unfavorable repayment terms, etc.) that the borrower voluntarily opts out of the credit

market, even when otherwise qualifying for loans of a desired size.

Innovations in financial markets and financial products can potentially overcome both types of credit

rationing and expand investments in agriculture (J-PAL and CEGA, 2022). For example, technological

innovations may make it possible for lenders to overcome informational barriers to mitigate adverse selection

and moral hazard, which in turn could alleviate quantity rationing in credit markets. For example, using

emergent predictive modeling tools (such as machine learning and artificial intelligence), it may be possible to

assess a potential borrowers’ credit worthiness even in the absence of a formal credit report. Advancements

in monitoring technologies may also address concerns over moral hazard in credit markets. Monitoring is

essential in addressing moral hazard issues in these markets, since monitoring can reduce private benefits for

1



borrowers (Fedele, 2005) and allow firms to adjust charges based on monitored behaviors (Chen and Jiang,

2019). Traditional methods for monitoring borrowers’ behavior can be quite costly, so increasingly lenders are

leveraging technological innovations as a low-cost means of monitoring. Kouzez and Lecointre-Erickson (2019)

and Bülbül et al. (2019) point out that advancements in information technology and financial innovations have

enhanced the monitoring of credit risk in developing countries, highlighting the extent to which technological

progress has contributed to improving the oversight of credit markets in these regions. Additionally, Hyytinen

and Toivanen (2004) have noted the potential for monitoring technologies, originally designed for controlling

moral hazard, to be repurposed by banks for other uses beyond managing moral hazard, thus potentially

increasing the returns to technological innovations.

At the same time, it is also well known that crop insurance has the potential to de-risk agricultural

investments through de-coupling production risk from production decisions, often leading to increased

investments in modern technologies (e.g., Miranda and Farrin, 2012; Karlan et al., 2014; Elabed and Carter,

2014; Cai et al., 2015; Hill et al., 2019). Yet, the myriad challenges with agricultural insurance are also well

known (e.g., Hazell et al., 1986; Binswanger-Mkhize, 2012; Miranda and Farrin, 2012; Carter et al., 2017;

Kramer et al., 2022). In particular, traditional agricultural insurance suffers from high costs associated with

monitoring and assessing losses, resulting in high administrative loads; informational asymmetries manifesting

as both adverse selection and moral hazard; the highly covariate nature of production risks; and a mismatch

between insurance demand (which is frequently highest among wealthier farmers) and who would actually

most benefit from risk transfers (typically poor farmers without the ability to self-insure). Index insurance is

frequently advocated as an alternative to conventional agricultural insurance, since it eliminates the need for

monitoring and assessing individual losses; is objective; and is transparent. Although index-based insurance

programs promoted to address some of these challenges have their own shortcomings (namely, basis risk; the

risk that index measurements are incongruous with an insured individual’s actual experiences), they have

recently been able to achieve modest scales by leveraging remote sensing, for instance to predict livestock

mortality (Chantarat et al., 2013) or estimate food insecurity and humanitarian response costs, enabling risk

pooling and drought insurance for countries (Kramer et al., 2020). However, to harness the full potential of

remote sensing, it is essential to deepen our understanding of the extent to which it can enhance rural credit

access, help smallholders manage agricultural risks (Benami et al., 2021), and improve targeted populations’

wellbeing and livelihoods.

Recent financial innovations have led to packaging credit and insurance together in a bundled financial

product. The theoretical underpinnings of such bundles lie both in financial mediation and risk management,

and date back at least several decades. Stiglitz and Weiss (1981) noted that one method to ameliorate rationing

2



in credit markets with imperfect information is through risk-reducing instruments like insurance. Being able

to insure a loan, and thus potentially protect households from having to adopt costly income-smoothing

adaptations in the event of an environmental shock or other calamity, could improve livelihoods (Morduch,

1995). To that extent, credit + insurance bundles provide a safety net for farmers, at least partially protecting

them from adverse events. But there are other potential impacts of such products. Carter et al. (2016) have

demonstrated theoretically how linking credit with insurance can have numerous effects on the terms of the

products themselves, the overall agricultural credit market, and important downstream effects on borrowers.

Specifically, they demonstrate how interlinking insurance with credit can, compared to what would be feasible

either in the absence of insurance or if the insurance were standalone (not explicitly linked to the loan), (i)

produce lower interest rates; (ii) flatten lenders’ agricultural loan offer curves, similar to what would be

observed if loans were fully collateralized; and (iii) encourage adoption of advanced technologies. Clearly,

therefore, there is at least theoretically great potential for interlinked credit and insurance products to slacken

both supply- and demand-side credit constraints, while increasing liquidity to facilitate adoption of modern

productivity-enhancing inputs.

Combining credit and insurance as a means of simultaneously overcoming liquidity and risk constraints

in developing countries has been a topic of interest in academic research in recent years, though with some

mixed results (Giné and Yang, 2009; Karlan et al., 2014; Gallenstein et al., 2021; Ndegwa et al., 2020; Mishra

et al., 2023). A common thread that emerges among these studies is the importance of both price and

non-price factors in the adoption of bundled insurance products, and the challenges posed by incomplete

insurance and credit markets in developing countries. But there are also some important differences in the

products being offered that may in part contribute to the mixed results. In some cases (e.g., Giné and Yang,

2009; Ndegwa et al., 2024 and the “meso-insurance” product evaluated in Mishra et al., 2023), the insurance

products are designed to directly repay the lender, rather than borrower, while in some other cases (e.g., the

“micro-insurance” product evaluated in Mishra et al., 2023), insurance payouts go directly to the borrowers.

As Carter et al. (2016) have pointed out, this distinction has important theoretical implications for the

impacts that these different credit and insurance bundles may yield. In cases where the insurance contract

compensates the borrower rather than the lender, knowledge about the existence of the insurance contract is

asymmetric, and lenders will operate as if there were no insurance. On the other hand, when the insurance is

structured to compensate the lender, there is a direct linkage between the credit and insurance components

of the bundle, with the aforementioned theoretical impacts on product terms, market structure, and borrower

behavior. In this case, these insurance payouts are used to expunge borrowers’ outstanding debts, thereby

minimizing (if not fully eliminating) the need for collateral, but also likely reducing cyclical borrowing

3



and increased indebtedness. These bundled insurance and credit products are essentially contingent credit

products, since the amount borrowers ultimately need to repay depends on the realization of a stochastic

weather outcome, and whether it is so adverse as to trigger payouts from the associated index insurance

product. This contingency in repayment might be thought to increase credit demand and increase investments

in riskier agricultural inputs that could enhance productivity and profitability.

Consistent with these theoretical predictions, Ndegwa et al. (2024) find that bundling insurance with

credit in Kenya increased the proportion of farmers using chemical fertilizers by over 13 percent relative

to the status quo and led to an almost 2,000 Kenyan shilling increase in fertilizer expenditures per acre

(roughly USD 20 per acre). What’s more, the increase in fertilizer expenditures with the interlinked credit

and insurance product was more than 2.5 times what was observed when farmers had a traditional credit

product without accompanying index insurance. In the context of smallholder production in Malawi, Giné

and Yang (2009) surprisingly find substantially lower credit take-up when the loans (specifically group liability

contracts to purchase hybrid maize seeds) were packaged with an actuarially-fair index insurance product,

thus not only contradicting the expectation that the repayment contingency should increase credit demand

but also that it should increase investments in productivity-enhancing inputs. Mishra et al. (2023) is an

interesting case study in this regard in that it examines the impacts of a credit and insurance bundle in

which the insurance and credit are interlinked (i.e., insurance payments are directed toward the lender, what

the authors refer to as a “meso-insurance” bundle) as well as a bundle in which the linkages between credit

and insurance are less pronounced (i.e., insurance payments are directed toward the borrower, what the

authors refer to these as a “micro-insurance” bundle). Although they do not directly examine differences in

credit take-up based on the differences in insurance payment terms, they find that only the loans with the

“micro”-insurance yielded positive effects on compound fertilizer application, and that these effects were only

observed during the second year of lending operations.

This study assesses the impacts of a bundled credit and insurance product that leverages emergent

technologies like smart phones, remote sensing, and machine learning. In particular, we evaluate KhetScore –

a novel approach to credit scoring using remote sensing and crop analytics.1 KhetScore was designed to reduce

transaction costs and overcome information asymmetries and documentation requirements in the provision of

financial instruments for marginal and landless farmers. In the Indian context, female farmers frequently

lack documented land rights required by formal lenders as a condition for qualifying for a loan. KhetScore

aims to precisely address this issue by producing a lens into the production and profitability potential of a

plot of land regardless of who holds the title. Further, since these loans are bundled with insurance, there
1Technically, "KhetScore" refers only to the credit scoring approach, but throughout the remainder of this paper we will

reference "KhetScore loans" when discussing the loans that were disbursed at least partly on the basis of a farmer’s KhetScore.

4



is both less of a need for collateral to secure a loan and a lower risk of collateral loss, since the insurance

component would compensate the lender in the event of a catastrophic hazard that destroys a farmer’s crop.

Consequently, there is great potential for this kind of product to ease both supply-side rationing on the part

of the lender and demand-side rationing on the part of the potential borrower. Therefore, the evaluation pays

particular attention to gender and baseline credit rationing status, conditions that have historically been

associated with limited credit access and utilization.

Partnering with Dvara E-Registry (DER), a highly-regarded social enterprise operating in more than

12 different states in India, we implemented a cluster randomized trial with an initial pool of around 1,800

potential KhetScore clients from 58 villages in the district of Jajpur in eastern Odisha. These 58 villages were

randomized into a treatment arm (29 villages), where farmers were offered KhetScore loans bundled with

insurance to protect them from the risk of crop failure, and a control arm (29 villages), where farmers were not

offered loans based on the KhetScore credit scoring approach but may have accessed other sources of credit as

well as insurance coverage under the Indian government’s national crop insurance scheme (Pradhan Mantri

Fasal Bima Yojana, or PMFBY), which requires land titles. We find beneficial impacts of offering farmers loans

bundled with insurance over a wide range of outcomes, including financial inclusion, agricultural profitability,

gender parity, and mental health outcomes. In particular, the evaluation demonstrates the potential for this

novel financial product to have transformative effects on the rural economy by strengthening financial literacy,

improving credit and insurance uptake, increasing farm profits, enhancing women’s empowerment, and easing

some of the psychological and emotional health stresses that so often accompany near-subsistence agriculture.

The contributions of this paper are threefold. First, we demonstrate that this kind of credit and insurance

bundle – and particularly one that determines an applicant’s creditworthiness not based on traditional metrics

but based on a metric using emergent digital technologies to determine production and profit potential –

can crowd in marginalized groups into financial markets and improve their overall experience with credit.

This effect is especially pronounced among women, who are typically excluded from credit markets because

they often lack the documented land rights required by lenders as a condition for obtaining a loan. Not

only did women take up loans at an accelerated pace compared to their counterparts in the control villages,

but they also reported an increase in borrowing from formal sources and a decrease in borrowing from

informal sources, therein largely increasing the extent to which they are included in the formal financial sector.

We also demonstrate that this kind of credit and insurance bundle comes with terms that borrowers find

especially favorable, as there is a significant reduction in the proportion of borrowers who reported difficulties

in repaying their loans, and these effects persist even among members of these groups who might otherwise

face financial difficulties.

5



Second, we contribute to the evidence on strategies to alleviate constraints faced by smallholder farmers

to increase their agricultural output and productivity. In particular, local average treatment effects estimates

indicate that those farm households that received KhetScore loans bundled with insurance were both more

likely to cultivate and cultivated a greater area during the Rabi (dry) season – a season in which many farmers

in Odisha leave their land fallow due to various constraints, including limited irrigation capabilities. Further,

local average treatment effects estimates demonstrate that borrowers were able to attain higher profits per

acre across both the Kharif (monsoon) and Rabi agricultural seasons compared with other households that

did not procure KhetScore loans. Among those that borrowed funds, revenues and profits per acre during the

Kharif season were INR 8,500 (about USD 110) and INR 9,100 (about USD 118) higher than among farmers

who did not avail KhetScore loans, while profits during the Rabi season were about 12,800 (about USD 166).

The effects are largest among those farmers who were not credit rationed at the time of the project baseline,

but the beneficial impacts carry over to those farmers who were quantity rationed at baseline, implying

that this product successfully alleviates supply-side constraints to borrowing that could enhance agricultural

livelihoods.

Finally, we demonstrate that this innovative credit and insurance bundle has the potential to significantly

enhance livelihoods for women, regardless of whether they directly benefit from expanded credit access.

We find that women from villages randomly assigned to be offered KhetScore loans expressed enhanced

feelings of empowerment, thereby adding to the literature on improving gender parity in agriculture through

financial inclusion. We find positive intention-to-treat effects across several different domains of women’s

empowerment, but it is worth noting in particular that those women who were the primary clients (with the

loan and insurance bundle taken out in their name) reported a significant increase in their contributions to

household financial decisions, especially those related to the decisions whether to borrow and how to use

borrowed funds. But even those women who were not the primary clients – but who co-signed on loans

taken out by male members of their households (e.g., their husbands) – reported improvements in their

empowerment, particularly in terms of their contribution to household livelihood decisions, their control over

their own income, and their accumulation of assets. From anecdotal qualitative evidence, these latter results

seem to emerge as a consequence of male household members sharing more information with their female

household members because they were co-signees on the loans. Although women’s contributions in these

decision-making domains remain relatively low, these beneficial impacts hint at this products’ potential to

alter intrahousehold dynamics and improve gender parity in financial matters. We also find that women from

randomly assigned treatment villages reported significantly lower levels of perceived stress, with these effects

concentrated among women who were not the primary clients. This finding highlights the extent to which

6



women indirectly bear risks related to agricultural production through conjugal contracts, and how bundling

insurance with agricultural credit can help to alleviate their concerns about the costs of downside agricultural

risks.

The remainder of this paper is structured as follows. Section 2 describes the methods employed in

our impact evaluation, including an overall description of the intervention (section 2.1), the design and

timeline of the evaluation (section 2.2), sampling and data collection methods (section 2.3), and our empirical

specification (section 2.4). Section 3 provides an overview of the DER clients that comprise our sample,

disaggregated by randomly allocated treatment and control villages. Section 4 presents the results of this

impact evaluation across various outcome domains, including financial market development (sections 4.2

and 4.3), agricultural outcomes (section 4.4), intrahousehold gender dynamics (section 4.5), and women’s

perceived stress (section 4.6) at both the aggregate level, for male versus female farmers, and based on whether

clients were credit constrained (either quantity or risk rationed) at the time of project implementation, where

appropriate. Section 5 offers some insights into the cost effectiveness of the overall KhetScore program, and

section 6 concludes.

2 Methods

2.1 Context and intervention

We implemented this study in two blocks of Jajpur, a district in the Indian state of Odisha’s low-lying coastal

plains. In this district, agricultural activities are one of the primary sources of income, with paddy rice being

the dominant crop grown. Irrigation is available in Jajpur due to the vicinity of the Mahanadi River, the

largest river of the state. This creates opportunities for farmers to expand production during the Rabi season,

but credit and risk are often cited as a primary reason for not investing in crop production during this season.

Intercropping is not common, making it suitable for applications that use remote sensing for crop monitoring.

Despite relatively higher literacy and education levels compared to farmers in other districts, Jajpur has

a large concentration of sharecroppers and marginalized landless farmers who fall outside the purview of

state-sponsored credit and insurance schemes (including PMFBY), since they lack documented land rights

required by these schemes.2

2PMFBY was initiated in 2016 by the Ministry of Agriculture and Farmers’ Welfare in an attempt to provide affordable
insurance for farmers against a spectrum of disaster risks (including yield losses due to natural fire and lightening; storms,
hailstorms, cylones, and tornadoes; floods and landslides; droughts; and pests and crop diseases). Under this program, the
government subsidizes a portion of the insurance premiums for small and marginal farmers, with farmers’ share of insurance
premiums capped at 2 percent of the sum insured for Kharif (monsoon) crops, 1.5 percent of the sum insured for Rabi (dry
season) crops, and 5 percent of the sum insured for commercial and/or horticultural crops, regardless of which season they are
grown. Coverage is compulsory for farmers who take out agricultural loans under the Kisan Credit Card scheme (a separate
program implemented by the Government of India to provide financing to farmers for agricutlural operations). In order to be

7



To address this challenge, our implementing partner, DER, developed KhetScore, an innovation that

uses remote sensing and machine learning approaches to ascertain farmers’ creditworthiness. To provide

KhetScore loans, DER (i) digitizes information on a farmer’s land parcels (regardless of whether a farmer

owns, rents, or sharecrops the land) using georeferenced smartphone images for verification; (ii) creates

plot-level agricultural credit scores by estimating past and current productivity from satellite imagery; (iii)

provides loans for farmers with sufficiently high credit scores; and (iv) adds on crop insurance to de-risk these

loans using smartphone images taken at regular intervals throughout the crop season to verify any claims of

crop damage due to extreme weather or other natural hazards such as pests and disease.

One of the main aims of the KhetScore credit scoring approach is to unlock credit for individuals who

do not have an established credit report with formal credit bureaus. The score is an index ranging between

0 and 100 based on six variables with different weights, including productivity (40 percent), changes in

crop health (20 percent), moisture (15 percent), past instances of mechanical damage (13 percent), nutrient

availability (7 percent), and the area of cultivation (5 percent). A farmer’s KhetScore needs to be at least 40,

which is considered to indicate reasonable creditworthiness, in order to be eligible for a loan. Applications

may, however, still be rejected if an applicant’s identification card (Aadhaar, India’s unique identification

number) does not match bank account details; if the applicant has a poor credit report (as opposed to no

credit report) with the credit bureau; or if the farmer did not cultivate for three consecutive years during the

agricultural season (Kharif or Rabi) for which the farmer is requesting a loan.

KhetScore loans have been designed around agricultural operations, with the loan size explicitly based

on the cost of paddy rice production for an acre of land, which is an estimated 30,000 Indian Rupees (INR)

(roughly 400 USD).3. The fact that these loans are tailored more to agriculture is perhaps one aspect that

differentiates these loans from traditional microfinance loans. The interest rate on these loans is 14 percent,

well below the prevailing microcredit interest rate of 24 percent.4 Repayment is done in multiple installments,

with two to three payments before harvest, and a final bullet repayment soon after harvest. Future loans

are provided to farmers subject to timely repayment of prior loans and availability of funds, but farmers

are granted a certain degree of flexibility in loan repayment, to maximize the chances that a farmer can

repay. Though there is a nominal penalty (INR 5 per day) for late repayment, they can be granted additional

time to repay the loan in case they have a legitimate excuse for being unable to repay. In the rare case that

eligible for coverage under PMFBY, farmers must have a valid and authenticated land ownership certificate or a valid land
tenancy agreement and must have a valid bank account and valid identify proof at the time of enrollment.

3Over the duration of the study, the prevailing exchange rate flucuated between 0.012 INR/USD to 0.014 INR/USD.
Throughout the remainder of this paper, we use an exchange rate of 0.013 INR/USD for converting monetary sums from INR to
USD.

4Due to the use of technology to monitor these loans, DER incurred lower operational costs, and thus was able to charge a
lower interest rate and be more flexible in their lending terms.

8



the farmer does not repay a loan, DER staff work with the farmer for proper follow up and counselling.

Repayment rates were very high, at 83 percent.

Farmers whose applications were approved would receive a loan bundled with a hybrid crop insurance

covering the full value of the loan. The insurance policy covers farmers for any visible crop damage from

natural causes, including excess rainfall, hailstorms, winds, animal attacks, pests and diseases, landslides, and

localized un-seasonal rainfall. Several of these perils are characteristic of cyclonic activities that regularly

impact coastal Odisha. To verify claims of crop damage and ensure adequate crop management leading up to

the damage, DER staff took bi-weekly images of insured land until the time of harvesting. For farmers who

report damage, claims are calculated by agricultural experts at DER who review the images to declare a

percentage of loss. This loss assessment determines the insurance payout, which is then sent into the famer’s

bank account.

Within each of the 58 study villages (including both treatment and control villages) DER organized

farmer profiling meetings during which they compiled a list of prospective clients who attended these meetings.

During these profiling meetings, DER provided product training, introducing farmers to KhetScore and the

associated bundled credit + insurance product. After that, farmers in treatment villages could apply for a

KhetScore loan. Farmers interested in a KhetScore loan needed the approval of at least one other family

member (typically their spouse), and this person would serve as a co-signer on the loan. Any farmer with

a KhetScore loan was also enrolled in insurance, with premiums fully subsidized in this initial pilot phase.

Farmers who did not apply for a KhetScore loan, or whose applications were rejected, could still apply for

the insurance product separately. Farmers in control villages were not permitted to apply for KhetScore

loans from DER, but were not prohibited from applying for loans from other lenders nor from accessing crop

insurance under the state-sponsored PMFBY crop insurance scheme.

2.2 Design and timeline

To evaluate the impacts of this bundled financial instrument, we implemented a cluster randomized trial with

an initial target sample of about 1,800 potential KhetScore clients from 58 villages of two blocks in Jajpur,

selected due to their large population of marginalized and landless farmers and DER’s existing presence in

these areas. DER had lending operations in all villages in these two blocks, except for those deemed unfeasible

due to the high risk of flooding. Including these villages in the sample would have increased insurance costs

to unsustainable levels and potentially jeopardized the ability of farmers in these villages to repay their loans.

We randomized all 58 villages into a treatment arm (29 villages), where farmers were offered the

9



KhetScore loans bundled with insurance, and a control arm (29 villages), where DER did not offer any loans

or insurance products, though farmers may have accessed other sources of credit (e.g., from other formal

lenders or from informal sources) or insurance (e.g., under the national insurance scheme) if they could

produce the necessary documentation to overcome the administrative hurdles. The randomized trial was

implemented over three consecutive seasons, including the Kharif seasons of 2021 and 2022, and the Rabi

season of 2021/2022. A timeline of the study is provided in Figure 1 below.

2.3 Sampling and data collection

All farmers that attended the village-level farmer profiling meetings comprise our sampling frame. At baseline

(at the end of 2020 through early 2021), we invited the full list of 1,872 farmers for a phone survey (in-person

data collection was not possible due to COVID-19 related restrictions on mobility in Jajpur district). Based

on power calculations prepared prior to study implementation, we aimed to survey at least 1,800 farmers

at baseline. Assuming a 3 percent percent attrition rate from baseline to endline, this meant recruiting at

least 1,854 clients through product meetings. Some of the farmers (62) invited for the phone survey were not

reachable due to incorrect phone numbers or phones being switched off. Others (101) refused to consent to

the interview. Consequently, we were able to survey 1,706 households.

The endline survey was conducted in person in November/December 2022, and by then, pandemic-related

restrictions on mobility had been lifted. We aimed to re-survey all respondents at baseline, as well as one

other family member of the opposite sex (usually the client’s spouse, and typically a co-signer for loans issued

by DER). Of the 1,709 clients surveyed at baseline, we were able to track and re-survey 1,621 clients (about

90 percent). We were able to identify and interview another family member for 1,294 of those respondents,

for a total sample of 2,915 individuals.

Both baseline and endline surveys included modules to collect information about the respondent’s

demographics, household income, financial inclusion, credit practices, women’s empowerment, insurance

take-up and renewal, awareness and perceptions of insurance, income from insured crops, and women’s mental

health. Questions related to insurance take-up and renewal, awareness and perceptions of insurance, and

income from insured crops, were asked only of DER’s primary client, regardless of gender. Questions about

women’s mental health (perceived stress) were asked only to the female member of the household, regardless of

whether they were the primary client.5 Other questions (related to demographics, household income, financial
5Our registered pre-analysis plan (study ID RIDIE-STUDY-ID-6110cd09b6003) indicates that we will also evaluate impacts

on household food consumption and women’s dietary diversity. These outcomes were included at the request of the project donor,
to track such measures among all funded projects. We have not included these outcomes in the impact evaluation presented
in this manuscript, both because there were no meaningful impacts observed, but also because there is not an obvious causal
mechanism by which we would have expected there to be observed impacts in the time frame of this evaluation.

10



inclusion, and credit practices) were asked to both respondents in each household. At baseline, we also

included a series of questions to determine types of credit rationing in the sample. As alluded to previously,

credit rationing frequently takes one of two specific varieties, namely quantity (supply-side) rationing or risk

(demand-side) rationing. These are obviously not part of the vernacular for the average farmer in Odisha, so

classifying individuals as being quantity rationed, risk rationed, or unconstrained (as it pertains to credit)

requires some deduction.

In particular, the baseline questionnaire included two questions, one that asked how much the respondent

needed to borrow to finance household activities, and a second that asked how much credit was actually

granted. If these two sums were the same, the respondent was deemed to not be credit constrained. If the

desired loan amount was less than the loan amount granted, the respondent was deemed to be quantity

rationed, since this would indicate a supply-side limit on the amount the respondent could borrow. There

were also respondents who indicated that they did not borrow within the last 12 months, and a third question

attempted to elicit the reason(s) why the respondent did not borrow in the previous 12 months. If the

respondent indicated that they did not borrow because (a) the loan amount offered was insufficient, (b) they

were deemed to not be creditworthy, or (c) they lacked sufficient capital to secure a loan, they were classified

as quantity rationed, again because these barriers to borrowing stem from supply-side conditions. If, on the

other hand, the respondent indicated that they did not borrow in the previous 12 months because (d) they

were afraid of losing collateral, (e) the loan terms were not flexible enough, (f) the application costs were too

high, or (g) there was a high interest rate, they were deemed to be risk rationed, since these constraints reflect

demand-side considerations inhibiting borrowing.6 A summary of credit rationing status by experimental

arm is discussed in greater detail below in Section 3.1.

2.4 Empirical strategy

To estimate intention-to-treat (ITT) effects, we estimate a series of OLS regressions in which endline values

of various outcomes (dependent variables) are regressed on a binary variable indicating whether the farmer

was from a village randomly assigned to be in the treatment group controlling for baseline values of the

dependent variable and a vector of covariates, with standard errors adjusted for the clustered nature of the

experimental design (i.e., by village, since randomization was done at the village level; Abadie et al., 2023).7

6It should be noted that the risk aversion reflected in these conditions is not solely on the part of the (potential) borrower.
High interest rates, inflexible loan terms, and high application costs are ways in which the lender can (or at least attempts to)
transfer some of the contract risk on to the applicant, but ultimately it is the applicant that decides to voluntarily withdraw
from the credit market due to these conditions.

7For some outcomes, we do not have baseline values to include in “pure” ANCOVA regressions. In such cases, we include
baseline values of a close proxy in “pseudo”-ANCOVA regressions.

11



Our base econometric specification will be the following:

Yib,1 = α + βYib,0 + δ1Tib +
J∑

j=1
θjxijb,0 + εib (1)

where Yib,t is the measure of outcome Y for individual i from block b in period t ∈ 0, 1, with t = 0 and

t = 1 indicating baseline and endline, respectively; xijb,0 is a vector of covariates including block fixed effects

to account for stratification and variables with statistically significant differences between the treatment

versus control at the time of the baseline, as well as demographic controls from the endline survey (caste,

literacy, age terciles, and the primary client’s gender); εib is an identically distributed disturbance assumed

to be independently distributed across villages, but correlated among members from the same village; and

Tib is a binary variable equal to 1 if individual i was from a village that was randomly selected as part of

the treatment arm. This base specification will allow for us to estimate the average effect of the program

in treatment villages. These ITT effects will be most relevant for outcomes related to financial market

development, including overall credit uptake, the distribution of credit sources (formal vs. informal), insurance

uptake and renewal, and, importantly, whether borrowers faced difficulty in repaying loans. We also use this

equation to estimate the programmatic effects on (women’s) empowerment.

In addition to this base specification, we will be interested in examining heterogeneous effects by (i)

gender and (ii) baseline credit rationing status.8 To do so, we will modify equation (1) and estimate one of

the following two regressions:

Yib,1 = α + βYib,0 + γFemaleib + δ1Tib + δ2Tib × Femaleib +
J∑

j=1
θjxijb,0 + εib (2)

and

Yib,1 = α + βYib,0 + γ1QRib + γ2RRib + δ1Tib + δ2Tib × QRib + δ3Tib × RRib +
J∑

j=1
θjxijb,0 + εib (3)

where, in equation (2), Femaleib is a binary variable equal to 1 if individual i from village b was a female (and

0 otherwise), and in equation (3), QRib and RRib are binary variables equal to 1 if individual i from village b

was quantity or risk rationed, respectively, at the time of project baseline (with unrationed individuals as

the omitted category). In equation (2), Femaleib is interacted with Tib to test whether the program had

differential effects on male (δ1) and female respondents (δ1 + δ2) within the treatment group. In equation (3),
8In our registered pre-analysis plan, we did not specify that we would explore heterogeneous impacts on the basis of baseline

credit rationing status. This was a dimension of heterogeneity that was subsequently included based on conversations with
participants in a research seminar given by one of the authors.

12



QRib and RRib are each interacted with Tib, allowing us to test whether the program had differential effects

on unrationed (δ1), quantity rationed (δ1 + δ2), or risk rationed individuals (δ1 + δ3) within the treatment

group.

For outcome measures related to women’s mental and emotional health, we only have data on female

respondents since these questions were only asked of female respondents. Further, these outcome measures

were specified differently at the time of baseline and endline data collection, so there are no comparable

baseline measures for these outcomes to be included in an ANCOVA regression. For these outcome variables,

we therefore modify the regression equation by dropping Yib,0, reducing the estimator to a single-difference

comparison of simple (conditional) differences between treated and control samples. However, as a unique

feature of our data set, we have data not only on the client, but also on another household member who,

in the case of households with KhetScore loans, would have co-signed on this loan. Consequently, we can

still estimate effects separately for women in households where the primary client is male (meaning that the

female is the “other household member”) versus female (so that the female is the “client”) and estimate a

variation of equation (2):

Yib,1 = α + γOtherib + δ1Tib + δ2Tib × Otherib +
J∑

j=1
θjxijb,0 + εib (4)

where all terms are as they were introduced before, though specifically pertaining to female respondents, and

Otherib is a binary indicator equal to 1 if the respondent was classified as the “other household member” –

implying the female respondent was not a DER client, but rather a female member within the household of

a male client (typically his spouse, and a co-signer on the loan). So, the ITT effect among DER’s female

clients is captured by δ1, and the spillover effects from DER’s male client to his female household member

are captured by δ1 + δ2.

Equations (1) - (3) provide estimates of overall impacts of offering loans and insurance based on

the KhetScore credit-scoring methodology among farm households in randomly selected treatment villages,

regardless of whether a household took up the product. Given some farmers opted not to apply for a KhetScore

loan and others’ applications were rejected, and assuming a positive correlation between take-up and the

respective outcomes of interest, these ITT effects provide a conservative estimate of the overall program

impacts. Compared to the effects on just those households who took up KhetScore loans, such estimates of

the overall program impacts in treatment villages may be of greater interest to policymakers interested in the

economywide impacts of expanding credit market access, rather than just the effects on those households

who took up KhetScore loans. Consequently, for many of the outcomes we assess, we focus on ITT effects. In

13



other cases, such as for outcomes related to agricultural performance, we also estimate impacts for those

households in the sample that decided to take up these products if their applications were approved. We

estimated local average treatment effects, or LATEs (Imbens and Angrist, 1994) via instrumental variables

(IV) estimation by instrumenting for product take-up with random assignment to the treatment group.9 For

estimating heterogeneous treatment effects (i.e., among women clients or those that were credit rationed at

baseline), we instrument for the interaction between gender or credit rationing status and product take-up

with an interaction between gender or credit rationing and random assignment to the treatment group.

For models in which the outcome variable is binary (e.g., cultivating during either the Kharif or Rabi

season), a further complication arises in the case of the linear IV estimator for the LATE, namely that

predicted probabilities from a linear probability model can be outside the unit interval (0,1), which can

be especially pervasive in IV estimations with comparatively low compliance among the treatment group.

To circumvent these challenges in estimating LATEs, we implement a recursive bivariate probit estimator,

which is appropriate for estimation scenarios in which (1) the outcome variable is binary, (2) the endogenous

treatment is binary, and (3) the instrument is binary. Based on recursive bivariate probit estimates, we are

able to construct LATEs using the formula presented in equation (11) in Chiburis et al. (2011). To consider

heterogeneous LATEs for these binary outcomes (i.e., by gender or by baseline credit rationing status), we

estimate recursive bivariate probit models separately for the different subgroups and compute point estimates

for LATEs for each subgroup, with corresponding 95 percent confidence intervals for these LATEs constructed

using the parametric simulation procedure proposed by Krinsky and Robb (1986).

Regardless of whether the outcome variables in these LATE regressions are continuous or binary, the

fact that exposure to the treatment was strictly limited to the members of the randomly assigned treatment

villages ensures that there are no “always-takers”; i.e., there were no farmers from control villages who availed

KhetScore loans. As a consequence, the LATE estimates that we report can also be interpreted as the average

treatment effects among the treated (ATT).
9If product take up is relatively low (indicating a non-trivial degree of non-compliance with assignment), the resulting LATE

could over-state treatment effects among the treated. Consequently, LATE estimates based on IV regressions can be considered
as upper bound estimates of average treatment effects. In the case of a single endogenous variable (product take-up) and a single
excluded instrument (random assignment), the F -statistic on the first stage regression is 44.09, so our identification strategy
does not suffer from a weak instrument.

14



3 Descriptive statistics

3.1 Summary statistics

Table 1 provides summary statistics of 1706 DER clients surveyed at baseline. About 70 percent of clients are

female and an average respondent is around 42 years of age. About 85 percent know how to read and write,

with 75 percent having completed (at most) primary schooling and the remaining 25 percent having completed

secondary schooling. Social stratification through the caste system is still quite prevalent in India, despite

legislation barring outright discrimination on the basis of caste. In our sample, about 36 percent of clients

belong to general castes, with another 36 percent belonging to Scheduled Castes (SC), 18 percent belonging

to Other Backward Classes (OBC), and 10 belonging to Scheduled Tribes (ST).10 There are statistically

significant differences in the proportions of general caste and SC between the treatment and control groups,

with the treatment group consisting of a higher proportion of general caste than the control group (42 percent

to 31 percent, respectively), and the control group consisting of a higher proportion of SC (43 percent to 29

percent, respectively). More than half of households’ annual income fall below INR 90,000 (roughly USD

1210) per year, or INR 247 (about USD 3) per day, which puts a 2-member household below the poverty line

of USD 2 per person per day. About 30 percent have an annual income between INR 90,000 and INR 120,000

(about USD 1614), while only around 20 percent have an annual income of more than INR 120,000 (USD

1,613).

In terms of clients’ experience with credit and crop insurance, about half of the clients in the sample

reported having purchased on credit in the past 12 months, though there is a higher proportion of clients from

treatment villages that reported purchasing goods on credit (56 percent compared to 42 percent). Around 30

percent have borrowed money over the same period, with an average desired loan size of INR 3.762 × 104

(about USD 506). In contrast, insurance take-up rate is extremely low — only 3 percent have ever had crop

insurance in the past, only 1 percent currently have crop insurance, and fewer than NA percent report having

received a pay-out from such crop insurance.

The vast majority of clients (86 percent) cultivated in the most recent Kharif season, while a smaller

proportion (75 percent) cultivated in the most recent Rabi season. A significantly higher proportion of clients

in the control villages cultivated during the most recent Rabi season compared with clients in the treatment
10In India, these caste differentiations qualify or disqualify groups for participation in the "reservation system", a form of

affirmative action that reserves a certain number (or proportion) of opportunities in education, employment (especially civil
service employment), government programs, etc., for members of historically disadvantaged groups. The term "general caste" is
a blanket term referring to castes that do not qualify for the reservation system due to not being deemed disadvantaged (at
least on the basis of caste). SCs and STs are considered the most economically disadvanaged social groups in India, and were
specifically defined in the Constitution of India in an order instituted in 1950. OBC is another collective term referring to castes
that are "backward" relative to general castes, but who are not listed among the SCs or STs.

15



villages (82 percent vs. 67 percent). An average household cultivates about 2 acres of land, with 75 percent

of households cultivating fewer than 2.5 acres of land (marginal), 23 percent cultivating between 2.5-5 acres

of land (small), and only 2 percent cultivating more than 5 acres of land (with a maximum of 10. Virtually

all clients (99 percent) are paddy rice farmers and 26 percent reported having experienced crop damages and

loss. An average household reports making a total agricultural revenue of INR 22,940 (roughly USD 309)

per acre, incurring a total cost of about INR 13,777 (USD 185) per acre (excluding the implicit cost of own

labor), and earning profit of INR 9,164 (USD 123) per acre (again, excluding the cost of own labor). Despite

having agricultural revenues and costs statistically similar to those from control villages, households from

treatment villages earned greater agricultural profits than households in the control villages, by a margin of

INR 10,891 (USD 146) to INR 7,587 (USD 102). We control for these baseline differences in all regressions

discussed in section 4 below.

Panel D of Table 1 and Figure 2 summarize the distribution of DER clients’ credit rationing status at

baseline. Across both treatment and control arms, the plurality of clients responded to survey questions in

such a way that they were classified as being quantity rationed (48 percent of clients in the control group and

54 percent in the treatment group), having faced supply-side constraints in accessing their desired level of

credit. They could potentially benefit from the credit + insurance bundle to the extent that the KhetScore

credit scoring approach circumvents some of the other, more traditional means by which individuals could

qualify for loans (e.g., credit reports with formal credit bureaus or documentation of land rights), or to the

extent that the accompanying insurance component eases lenders’ concerns about borrowers’ potential to

repay loans, especially in the event of near-catastrophic crop losses that might otherwise impinge upon their

repayment ability. Roughly another 30 percent of clients across both treatment and control were classified as

being risk rationed, opting out of participating in the credit market because of lenders transferring contract

risk onto potential borrowers in the form of higher interest rates, fees, or collateral burdens. The bundled

insurance component may benefit them by absorbing some of that contract risk and ameliorating some of

these other less attractive terms. Finally, 24 percent of farmers in the control group and 16 percent of

farmers in the treatment group were neither quantity nor risk rationed at the time of project baseline, and

consequently are identified as unconstrained.

In Table 2, we report summary characteristics of DER clients that were classified as quantity rationed,

risk rationed, and unconstrained at the time of project baseline, with a series of t-statistics for the differences

in means between households that were deemed quantity- or risk-rationed and those that were deemed

unconstrained. As seen in columns (4) and (5), there are a large number of statistically significant differences

in the group means, and largely consistent with what we might expect. One dimension that is surprising,

16



however, is that there is a statistically significant difference in the proportion of DER’s unconstrained clients

who were female compared to the quantity rationed clients who were female. The proportion of unconstrained

clients who were female was roughly 80 percent, compared to a rate of about NaN percent female among

quantity rationed clients. A priori, we could reasonably expect female clients to be more likely to report

credit constraints, since they typically lack formal credit scores, lack documented land rights for the land

they would cultivate with additional liquidity, and have limited personal ownership of assets that could be

used as collaterial to secure an agricultural loan. It is surprising, therefore, that such a high proportion

of the unconstrained sub-population would be female. Part of this is likely due to the way in which we

characterize clients as being credit constrained or not. In particular, clients were classified as unconstrained

if they responded that they did not need to borrow on credit in the last 12 months. While normally this

would signal sufficient personal liquidity, it could also signal that the client simply did not have any financial

ventures to pursue that would require additional capital. It could reasonably be surmised that many women

in rural Odisha might fall into this latter category, and yet would be classified as unconstrained in credit

access, simply because their credit access was not tested in the preceding 12 months. Unfortunately, we are

unable to distinguish between these two potential reasons for not needing to borrow on credit in the past

year.

More consistent with what we would expect, there are also differences in the proportions of clients

completing primary and secondary schools, with quantity rationed and risk rationed clients having lower

proportions completing primary school and secondary school compared to the unconstrained clients. Uncon-

strained clients are also less likely to be classified as low income (below INR 90,000 per year) and more likely

to be classified as high income (more than INR 120,000 per year) than quantity- or risk-rationed clients.

Unconstrained clients also cultivate more area (2.3 acres) than quantity rationed (NaN acres) or risk rationed

clients (1.9 acres). While we are not attempting to ascribe causality with regards to credit rationing, it is not

unreasonable to at least envision some causal mechanisms whereby some of these factors might contribute to

credit constraints.

Overall, variables are fairly well balanced across treatment and control groups, with only a few exceptions.

Imbalanced variables include the following: the percentage of DER clients that belong to the General versus

Scheduled Castes, the percentage of clients purchasing goods and services on credit in the past 12 months,

the percentage of farmers that cultivated in the most recent Rabi season, agricultural profits per acre during

the most recent Kharif season, and the percentage of clients that are not credit constrained. While we might

reasonably expect some differences between treatment and control groups to emerge by chance, the fact that

there are six dimensions for which there are statistically significant differences between the treatment and

17



control groups suggests that we do not have ideally comparable groups with which to identify treatment effects.

This is confirmed by a joint orthogonality test in which we regress the binary treatment assignment indicator

on each of these variables (dropping categorical variables where necessary to avoid perfect multicollinearity).

The corresponding F test statistic associated with the null hypothesis that all coefficients in this regression

are simultaneously equal to zero is 3.712, leading to a rejection of this null hypothesis. Consequently, we

control for these experimental imbalances in our impact evaluation estimates in section 4 below.11

3.2 Uptake of the intervention

Table 2 reports monitoring and evaluation (M&E) data concerning the uptake of the intervention. Preceding

each season, enrollment initiatives occurred one to two months in advance. Dedicated personnel from DER

personally contacted all eligible farmers in each treatment village to encourage them to apply for the KhetScore

product. Column (1) indicates, per season, the number of farmers who applied for KhetScore loans. Column

(2) denotes the count of farmers eligible for KhetScore and who had the loan disbursed to their bank accounts.

Column (3) represents the number of farmers enrolled in insurance during the respective seasons (regardless

of loan uptake), and column (4) indicates the count of farmers who received payouts for those seasons.

On average, approximately 28% of farmers offered KhetScore loans applied for it each season. Initially,

uptake was subdued due to various challenges. COVID-19 lockdowns necessitated DER staff to contact farmers

via phone, leading to issues concerning farmer availability, low mobile penetration in Odisha, incorrect contact

information, network connectivity problems, and, perhaps most importantly, establishing and maintaining

rapport with potential clients. Over time, these challenges were ameliorated to some degree by transitioning to

in-person visits, facilitating continued engagement with farmers over multiple seasons. Additionally, financing

KhetScore loans during the pandemic was challenging as institutional lenders were cautious about microloans.

However, DER demonstrated the viability of its lending model using the relatively small number of initial

KhetScore loans, securing additional funding for subsequent seasons. Consequently, there was a progressive

increase in the number of farmers being invited to apply for, applying for, receiving KhetScore loans, and

enrolling in the accompanying insurance product.
11Some observations were missing for a few of these variables that demonstrated statistical imbalances between experimental

groups at baseline (specifically, whether a DER client had purchased on credit in the preceding 12 months, whether a DER client
had cultivated during the most recent Rabi season, and agricultural profits per acre during the previous Kharif season). We
imputed values for these missing observations using the village median value. In such cases, we also control for these imputations
with a series of binary variable indicating that these observations had been imputed. In addition to these variables, we also
include control variables to eliminate potential confounding factors and to increase the precision of our econometric estimates.

18



4 Empirical Results

4.1 Impacts of KhetScore loans

As mentioned previously, we are interested in measuring the effects of KhetScore loans on several important

outcomes pertaining to financial market development, enhancing agricultural livelihoods, gender equity and

increasing women’s empowerment, and perceived stress. Within each of these outcome domains, we consider

several different areas in which we test whether – and the extent to which – KhetScore loans have yielded

impacts. There are therefore several hypotheses that are being tested within each outcome domain, with a

nontrivial likelihood that the p-values for these different hypotheses tests could be correlated, which in turn

could result in a high likelihood of Type I error (false rejection of the null hypothesis). In the sub-sections

that follow, in addition to reporting naïve (unadjusted) p-values for the estimated treatment effects (which

treat each test as an independent test), we also report sharpened q-values (Benjamini and Hochberg, 1995;

Benjamini et al., 2006; Anderson, 2008), which are essentially adjusted p-values that account for the False

Discovery Rate (FDR) within each of the different outcome domains. Specifically, we use the two-step

procedure initially proposed by Benjamini et al. (2006) and described by Anderson (2008). This two-step

procedure detects whether a null hypothesis can be rejected at some significance level q, but alone it does

not report the smallest level q at which a null hypothesis would be rejected, which arguably is of greatest

practical importance. We follow the suggestion in Anderson (2008) and undertake this procedure for all

possible q levels between 1.000 and 0.001 (in downward increments of 0.001), recording the q-value at which

the null hypothesis ceases to be rejected. This approach is more forgiving and yields greater statistical power

than correcting for the family-wise error rate (e.g., Romano and Wolf, 2005; List et al., 2019), which limits

the probability of a family of hypotheses being subject to any Type I errors.12

4.2 Impacts on credit market development

We first examine the impacts of the program on credit uptake (Table 4) and report ITT effects of random

assignment to KhetScore treatment group. We focus on ITT effects rather ATT effects because policymakers
12Our pre-analysis plan specified that we would use the Bonferroni correction, which would either adjust the significance level

(i.e., the permissible probability of Type I error, α) by dividing it by the number of tests being conducted, or by adjusting
the p-value associated with a given test by multiplying it by the number of tests being conducted. Since some of our outcome
domains involve a large number of tests (multiple different outcomes, heterogeneity by gender and baseline credit rationing
status, ITT vs. LATE), we determined that the application of the Bonferroni correction would severly deplete statistical power
and increase the probability of Type II errors to an unacceptable degree. For example, when testing for treatment effects on
our agricultural outcomes, we test for two types of treatment effects (ITT and LATE) on six specific outcomes (whether they
cultivated in a particular season, area cultivated, revenue, cost of cultivation, profits, return on investment) across two seasons
(kharif and rabi), among the full sample, disaggregated by gender (male vs. female), and disaggregated by baseline credit
rationing status (unrationed vs. quantity rationed vs. risk rationed). This results in a "family" of 2 × 6 × 2 × (1 + 2 + 3) = 144
hypotheses being tested. With α = 0.1 – the most lenient conventional level of statistical significance – we would require p-values
of 0.0007 to reject the null hypothesis of no treatment effect, which would almost certainly lead to the conclusion of a large
number of false null effects.

19



are likely to be most interested in the the overall effects of such a program on credit market development in

rural communities on average.

Starting with the overall impact of the KhetScore program among the full sample (Figure 3 and Panel

A of Table 4), although there is limited evidence that the program had an overall effect on increasing credit

uptake, there is compelling evidence that the KhetScore credit + insurance bundle is effective in encouraging

borrowing from formal sources and deterring borrowing from informal sources, which are typically associated

with higher interest rates, less favorable loan terms, less transparency, and less regulatory oversight.13 In

particular, there was a 21 percentage point increase in formal borrowing and a corresponding 20 percentage

point reduction in informal borrowing, suggesting that on average clients in treatment villages effectively

substituted formal for informal borrowing. Formal borrowing is typically much preferred to informal borrowing,

since the former is typically associated with lower interest rates, more favorable terms, and because formal

lenders are registered with government regulatory authorities (the Reserve Bank of India, in this case), and

consequently are expected to operate under the rules and regulations of the regulators. The latter, on the

other hand, although perhaps filling a gap in rural credit markets, have a history of near-usurious interest

rates and may be less lenient with deliquencies. The KhetScore bundle also appears to come with more

favorable terms, as evidenced by the sizable reduction in the proportion of farmers in treatment villages

who indicated that they faced difficulties in repaying loans. Of note, only about 20 percent of borrowers in

treatment villages reported facing difficulty in repaying loans, compared with 76 percent of borrowers in

control villages. Note that this question was only asked among DER clients who indicated that they had

borrowed or purchased goods on credit over the preceding 12-month period. The net result is a 56 percentage

point reduction in the proportion of borrowers who reported difficulty in repaying loans. Part of this effect

is certainly due to the lower interest rate associated with KhetScore loans (14 percent compared to the

prevailing microcredit interest rate of 24 percent), but it certainly seems plausible that much of this could

also be attributed to the fact that these KhetScore loans do not require collateral, unlike most other types of

loans. Additionally, there is some anecdotal evidence that farmers liked the repayment schedule for KhetScore

loans, with loan repayments in periodic installments rather than in a single lump sum (which is often the

case for agricultural loans) or in weekly installments (which often applies to microfinance loans), though

additional anecdotal evidence suggests some borrowers may have had to work off-farm to enable them to be

able to repay the first two or three installments, which were due during the period before they had harvested
13It is worth noting that while the p-value associated with the treatment effect on taking up credit from any source (column

(1)) is 0.53, the sharpened q-value is only 0.13, close to an acceptable probability of Type I error. Sharpened q-values can
actually be smaller than naïve (unadjusted) p-values in cases in which many hypotheses within a family are rejected, since it
is likely that many of these rejections will be true rejections, and it is possible to tolerate several false rejections while still
maintaining a low FDR (McKenzie, 2020).

20



their crops. This did not, evidently, have a detrimental impact on how they perceived their abilities to repay

their loans.

In Figure 4 and Panel B of Table 4, we examine heterogeneous effects among male and female clients.

Among male clients, there is a 26 percentage point reduction in borrowing from informal sources, with a

17 increase in borrowing from formal sources, revealing that many male clients swapped out informal loans

for formal loans. The uptick in formal borrowing was not quite sufficient to offset the reduction in informal

borrowing, so we actually observe a small 8 percentage point reduction in overall credit takeup among male

clients (p = 0.1; sharpened-q = 0.03). Among female clients, on the other hand, there is compelling evidence

that the program crowded-in borrowing.14 In particular, we find evidence of a 26 percentage point increase

in formal borrowing among women clients, well exceeding the 10 percentage point reduction in informal

borrowing, resulting in a net increase in borrowing among women of roughly 16 percentage points. As

envisioned, these KhetScore loans appear to be particularly beneficial in expanding credit access among

women who are typically implicitly – if not explicitly – excluded from formal credit markets due to “thin”

credit files and lack of documented land rights.

In Figure 5 and Panel C of Table 4, we explore heterogeneity in program impacts among clients based

on their baseline credit rationing status. We find evidence that the program led to small and statistically

significant increases in credit takeup among unconstrained clients (a 6 percentage point increase; p = 0.23,

sharpened q =0.06) and small and marginally significant increases in overall credit uptake among quantity

rationed (total effect size of 2.5 percentage points; p = 0.54, sharpened q =0.13) or risk rationed (total effect

size of 3.2 percentage points; p = 0.45, sharpened q =0.11) farmers. We generally find that the program

facilitated a substitution of formal for informal credit among unconstrained, quantity rationed, and risk

rationed members of treatment villages, regardless of credit rationing status. In particular, we observe a 24

percentage point increase in formal borrowing among unconstrained farmers, a 20 percentage point increase

in formal borrowing among quantity rationed farmers, and a 21 percentage point increase in formal borrowing

among risk rationed farmers. At the same time, we observe a 18 percentage point reduction in informal

borrowing among unconstrained farmers, a 18 percentage point reduction in informal borrowing among

quantity rationed farmers, and a 21 percentage point reduction in formal borrowing among risk rationed

farmers. Consequently, although quantity and risk rationed borrowers may not disproportionately benefit

from the KhetScore credit scoring approach, the results clearly suggest that they benefit roughly as much
14As noted in the footnote to Table 4, the row labeled "Total effect | Female" represents the linear combination of regression

coefficients (reported in supplementary appendix A), Table A. Referring back to equation (2), this total effect is reflected by the
linear combination δ̂1 + δ̂2, and the standard error for this linear combination computed as σ̂δ̂1+δ̂2

=
√

σ̂2
δ1

+ σ̂2
δ2

+ 2σ̂δ1,δ2 .
Test statistics and standard errors for these linear combinations of regression coefficients were computed using the –glht–
(general linear hypothesis test) function from the –multcomp– package in R. The p-values from these tests have also undergone
the aforementioned adjustment to account for the FDR, reported in the line titled "Sharpened q-value".

21



as those individuals who were not credit rationed – and who presumably would have had a more secure

financial standing – at the time of the project baseline. Both quantity rationed and risk rationed individuals

from treatment villages reported facing difficulty in repaying their loans with a much lower frequency than

members of the control group (column 4), and despite the point estimate on this effect being greater among

those individuals who were not credit constrained at baseline, the effects are not statistically distinguishable

from one another.

4.3 Impacts on insurance market development

Figures 6-8 and Table 5 presents the results on the impacts of the program on outcomes pertaining to

insurance, including insurance uptake, insurance renewal, and financial literacy (i.e., understanding of the

terms and conditions of crop insurance measured via an insurance knowledge score). Although farmers

who took KhetScore loans universally received insurance coverage (since it was bundled with their loan),

there is also the possibility that farmers in randomly selected treatment villages – through product training

and increased familiarity with how crop insurance operates – could have purchased crop insurance coverage

even if they did not specifically take a KhetScore loan (e.g., under the PMFBY crop insurance scheme).

Consequently, here, too, we focus on ITT effects rather than ATT effects, for exactly the same reason as

elucidated above; namely that these program-wide effects are likely to be of greater policy interest than the

effects among those who took up KhetScore loans.

In short, we find overall positive and statistically significant effects of the program on insurance outcomes.

Random assignment of villages to the treatment increased the likelihood of clients taking up insurance at

least once during the study period by more than 60 percentage points, and the treatment effect does not vary

across gender nor baseline credit rationing status. The program also increased the likelihood that once-insured

farmers would renew insurance coverage by nearly 30 percentage points (Figure 6 or Panel A of Table 5),

though this effect is found to be mainly driven by women (Figure 7 or Panel B of Table 5), whose insurance

renewals increased by nearly 60 percentage points, compared an increase of only 7 percentage points among

men (p = 0.38; sharpened q = 0.07).

Lastly, we see that random assignment to the treatment group increased the number of correct responses

to a series of questions meant to test respondent’s awareness of the features and nuances of crop insurance,

which we use as a proxy for financial literacy (Figure 6 or Panel A of Table 5). In total, random assignment

increased the number of correct responses by about 0.5 questions. Although on average this effect is similar

between men and women (Figure 7 or Panel B of Table 5), there is much greater variation in the treatment

effect among women (p = 0.21; sharpened q = 0.05. We do not observe significant differences in treatment

22



effects on insurance outcomes among participants with different rationing status at baseline (Figure 8 or

Panel C of Table 5), which further suggests that the KhetScore product does not implicitly discriminate

against farmers who faced supply- or demand-side credit constraints at baseline.

4.4 Impacts on agricultural outcomes

In Table 6 and Table 7, we present results on ITT and LATE estimates of the KhetScore program on a series

of agricultural outcomes during Kharif (monsoon) and Rabi (winter) seasons, respectively. Figures 9 - 14

illustrate LATEs for the two seasons, among the full sample (Figures 9 and 12), disaggregated by gender

(Figures 10 and 13), and based on baseline credit rationing status (Figures 11 and 14). In the case of these

agricultural outcomes, both ITT effects and LATEs are likely to be of particular interest to policymakers.

LATE results are qualitatively similar to ITT estimates, but, as would generally be expected since they reflect

treatment effects among KhetScore loan recipients, LATEs are of considerably larger magnitude, suggesting

that the beneficial effects of the KhetScore credit + insurance bundle are primarily experienced among those

who decided to take up the loans. Here we mainly discuss the LATE results, which are also ATTs since

access to KhetScore loans was restricted to farmers from randomly selected treatment villages, so there is no

noncompliance among the control group (i.e., there are no “always takers”).

Starting with the Kharif season in Table 6, we find that farmers that took out KhetScore loans were

more likely to cultivate during the Kharif season, despite an already high proportion of farmers cultivating

during Kharif. Despite the increase in the likelihood of cultivation, we do not find any evidence that farmers

that took out KhetScore loans expanded their area cultivated during Kharif. Perhaps most importantly from

farmers’ perspective, we find that agricultural profits per acre increased substantially among those farmers

that took out KhetScore loans (on average, by about INR 9,100), or about USD 120 at prevailing exchange

rates at the time, and the effect was virtually the same between men and women (INR 9,300), though there

was much greater variation in the effect among men. This increase in agricultural profits was primarily driven

by an increase in revenues per acre, which increased by nearly INR 8,500 (about USD 110) compared to

revenues in the control group. The effects on revenues per acre are more pronounced among male clients,

with male clients observing an increase in revenues of INR 10,900 per acre (USD 142 ), compared with a

increase of only INR 6,500 per acre (USD 84.5 ) among women clients. There is little evidence that farmers

who took out KhetScore loans saw an increase in total costs of production, so it seems likely that farmers

must have been able to increase crop revenues in a cost-neutral fashion.

KhetScore loans also appear to have had beneficial impacts among those who were previously quantity

(supply-side) rationed or unrationed in their access to credit, as revenues per acre increased by approximately

23



INR 17,400 and INR 8,700 per acre, respectively (roughly USD 221 and USD 117, respectively). The positive

effect on agricultural profits per acre is experienced by farmers regardless of their baseline credit rationing

status, though farmers who were unconstrained at baseline experienced the largest boost in agricultural

profits (of about INR 17,000 or USD 221 per acre), followed by quantity rationed (INR 9,000 or USD 117 per

acre) and risk rationed (INR 7,400 or USD 91 per acre).

Turning to the Rabi season in Table 7, we observe that households that took out KhetScore loans were

about 22 percent [95 percent CI = (0.06,0.34)] more likely to cultivate paddy rice during the Rabi season

(Panel A, column 2), and households that took out KhetScore loans increased their Rabi paddy rice area by

about 1.3 acres, more than double the acreage cultivated among control group farmers during the Rabi season

(Panel A, column 4). The expansion of land cultivated during the Rabi season was considerably higher for

women (expansion of 1.62 acres) than for men (1.05 acres). There is also evidence that the expansion in Rabi

cultivation along the extensive margin was more pronounced among farmers who had previously been credit

constrained: previously quantity and risk rationed farmers increased their Rabi area by more than 1.53 and

1.06 acres, respectively, compared to an increase of 0.79 acres among those farmers who had not been credit

constrained at the time of project baseline. The total cost of production per acre fell by about INR 9,600

per acre (about USD 130) among farmers in the treatment group who took out KhetScore loans, perhaps as

treatment farmers generated economies of scale (Panel A, column 8), and the reduction in costs of production

were considerably larger among women borrowers than among men decrease than men (Panel B, column 8).

We also observe households that took out KhetScore loans earning significantly higher agricultural profits,

though the increases in agricultural profits are larger among male farmers and those farmers that had not

been credit constrained at baseline. All told, results in Panels B and C imply that this product could make

credit more accessible and inclusive, and could help alleviate both supply-side and demand-side constraints

to borrowing that could enhance agricultural livelihoods.

How do we explain these observed impacts on agricultural revenues, costs, and profits?15 In tables

8 and 9, we report treatment effects on various agricultural input expenditures, including seeds, fertilizer,

irrigation, machinery, land rentals, and hired labor (disaggregated by gender) for the Kharif and Rabi seasons,

respectively, while Tables 10 and 11 report treatment effects on household (unpaid) labor (including own-labor

and labor from other male and female household members) for Kharif and Rabi, respectively. The treatment

effects estimates are not measured very precisely, perhaps due to very heterogeneous investment behaviors

among borrowers, but there is rather compelling evidence that one of the important pathways by which

borrowers were able to increase their profits across both seasons is through investing in new seeds. Although
15In our pre-analysis plan, we did not indicate that we would analyze treatment effects on input use. This analysis was

undertaken in an attempt to better elucidate some of the outcomes discussed earlier.

24



harvested grains can be stored and used for seeds in subsequent seasons, seed viability and the nutritional

value of harvested grains tend to decline continually with extended storage time, and declining seed viability

tends to be accompanied by lower germination rates and lower yields. Consequently, occasionally purchasing

new seeds remains important for ensuring the use of the highest quality and most productive germplasm. In

our study, we find that farmers taking KhetScore loans spent an additional INR 1,430 on rice seeds for the

Kharif season and an additional INR 1,140 on rice seeds for the Rabi season, or about USD 19 and USD

15, respectively.16 While these amounts may not seem substantial at face value, they represent increases

of 67 percent and 61 percent over seed investments in the control group during Kharif and Rabi seasons,

respectively. In considering the gender disaggregation of these effects on seed expenditures, it seems much of

the overall effect can be attributable to increased seed purchases by male farmers, with scant evidence of

increased expenditures among women farmers.

In Table 10, we find evidence that male borrowers significantly reduce their allocation of labor to

agricultural production during the Kharif season, and apparently substitute for their own labor by hiring

in female workers. On average, men in the control group allocated about 58.7 days to providing labor for

agricultural production on their own farms, yet this amount was effectively eliminated once they took out a

KhetScore loan. Instead, they increased their spending on hiring female agricultural labor by INR 5,050,

which, at an average daily wage of about INR 240 per day, corresponds to an additional 21 days of farm labor.

During the Rabi season, male borrowers essentially maintain their same labor allocation (Table 11, Panel B,

column 1), reduce labor supplied by other family members (Table 11, Panel B, columns 2 and 3) and reduce

their reliance on hired laborers (Table 9, Panel B, columns 6 and 7), opting instead to supplement this reduced

labor supply with machinery (9, Panel B, column 4).17 We observe male borrowers increasing expenditures on

agricultural machinery by INR 3,220, roughly doubling the expenditures on machinery during the Rabi season

among farmers in the control group. Female farmers, on the other hand, increase their own labor supply

during the Rabi season by about 120 percent relative to the control group, nearly double the amount of labor

supplied by other household members, and more than double their expenditures on hiring male and female

laborers, respectively. These clear differences in agricultural labor use between men and women is consistent

with observed disparities in challenges faced by men and women farmers in accessing agricultural labor. As

Adhikari et al. (2023) observes, women face unique challenges in regards to agricultural labor, with more than

50 percent of women in their sample reporting that they faced challenges in hiring labor (compared to only 14

percent among men in their sample), 25 percent reporting that they had to pay extra for labor (compared to
16For the Rabi season effect on seed expenditures, the sharpened q-value is 0.25 compared to a naïve p-value of 0.09, so we

conservatively fail to reject the null hypothesis of no effect on Rabi season seed expenditures
17We do not have data on Rabi season input expenditures or labor for the 2020-21 Rabi season. Consequently, these regressions

do not control for baseline levels of input use or labor.

25



only 8 percent among men), and 30 percent reporting that laborers did not respect them (compared to only 8

percent among men). As a consequence of these challenges, women are often delayed in crop establishment,

which may push the crop growth cycle into periods of the year that are less hospitable. Women farmers

were about 26 percent more likely to cultivate during the Rabi season and increased their cultivated area

by 1.62 acres, so there is clearly a need for additional labor for land preparation, transplanting, weeding,

and harvesting activities. Although female farmers cut their expenditures on fertilizers, the additional labor

employed must be sufficiently productive to yield higher agricultural profits and a marginally significant

increase in return on agricultural investments, though admittedly these metrics ignore the value of unpaid

household labor.

4.5 Impacts on intra-household gender dynamics

In Table 12, we present results of program effects (ITT) on intra-household gender dynamics among DER

clients (Panel A) and non-client family members (Panel B). We pay particular attention to several important

indicators of empowerment, including whether they (1) contribute to household borrowing decisions, (2)

contribute to household decisions about how to spend borrowed money, (3) give input on household livelihood

decisions, (4) exercise control over uses of income, and (5) own land and/or other assets. We find sizable and

statistically significant improvements in these five empowerment indicators among households in treatment

villages, primarily among women. Indeed, we find virtually no evidence that men reported an increase

in empowerment, though this is almost certainly because men in the sample villages are all sufficiently

empowered. Interestingly, there is considerable heterogeneity in whether the significant effects are driven by

female DER clients or non-client female household members. Among female DER clients, random assignment

to treatment improves their ability to contribute to household financial decisions, specifically household

decisions around borrowing and the use of borrowed funds. Female DER clients were about 30 percentage

points more likely to contribute to household decisions about whether to borrow, and about and about

25 percentage points more likely to contribute to household decisions about how to use borrowed funds.

Other female household members (i.e., female partners of male clients who are co-signers on loans) reported

improvements in their ability to provide input on household livelihood decisions, control over household

income, and own land and/or other assets (by 38, 38, and 35 percentage points, respectively). In general,

women still trail far behind their male counterparts when it comes to household decision making power,

especially when it comes to household financial decisions. In the control group, only about 10 percent of

women reported that they contributed to household decisions about whether to borrow or how to spend

borrowed funds. Consequently, although there is still much progress to be made toward truly achieving

26



gender parity, our results provide strong evidence for the effectiveness of financial inclusion (perhaps not

limited to KhetScore loans) in empowering women and improving intra-household gender relations. As more

women are able to avail credit through inclusive credit scoring approaches and make profitable investments in

agriculture or other enterprises, they may be more widely recognized for making valuable contributions to the

household’s financial situation, and thus they may be given more agency in important household decisions.

4.6 Impacts on women’s perceived stress

Finally, we consider the effects of KhetScore loans on women’s mental and emotional health. Women in

rural settings frequently bear substantial risks in household production, even when they may not be directly

involved in agriculture. According to the “conjugal contract” model (Carter, 1997), men and women operate

as separate spheres in the household, with both aiming to maximize individual utility subject to separate

budget constraints. Men typically manage agricultural production, and the conjugal contract establishes

the rules by which men transfer resources to women to enable them to operate non-farm enterprises and

manage the household, including childcare. Because the agricultural sphere is subject to various production

risks (including, but not limited to climate or environmental conditions), there is a risk of men defaulting on

the conjugal contract, often leading women to liquidate their personal assets or reduce their consumption to

enable them to continue to satisfy household needs that rural society maintains are the woman’s responsibility

(Arteaga et al., 2024). There is, by now, a considerable body of literature that has documented how negative

agricultural shocks disproportionately impact women (see, e.g., Hoddinott and Kinsey, 2003; Quisumbing

et al., 2018). This indirect exposure to negative agricultural shocks may be even more severe when men

leverage credit to finance agricultural investments: not only is the household food supply and income stream

in jeopardy, but the household may lose collateral or liquidate assets or borrow from another lender to repay

borrowed funds. Consequently, bundling insurance with credit has the potential to provide peace of mind

that, even in the event of a negative agricultural shock, women’s exposure is limited.

In Table 13 we report the effect of the program on several measures of perceived stress, for both the

female client (Panel A) and another female household member, who is likely the co-signee on the loan if the

client was a male (Panel B).18 Using a Likert scale ranging from “never” (1) to “very often” (5), we asked

women to rate the frequency with which they felt (i) “unable to control the important things in life”, (ii)

“confident in their ability to handle personal problems”, (iii) “confident that things were going their way”,

and (iv) “difficulties were piling up so high that they could not overcome”. We evaluate responses to each of

these questions individually in columns (2) - (6). For questions (ii) and (iii), responding “very often” would
18The survey questions regarding perceived stress were only directed toward women respondents.

27



indicate that individuals felt relative comfort, since such answers would imply they frequently felt capable of

handling problems that arose or that things were generally going their way. Conversely, for questions (i) and

(iv), responding “very often” would indicate that individuals perceived substantial stress, since such answers

would imply they frequently felt incapable of controlling important things in their life or that challenges were

becoming almost insurmountable. In column (1) of Table 13, we evaluate program effects on a composite

stress score. To construct this composite stress score, answers to questions (ii) and (iii) were recoded so

that higher scores indicated a higher degree of stress, and then we took the arithmetic mean of respondents’

answers to the four questions.

It is worth noting that, in the absence of treatment, other female family members of DER’s male

clients (e.g., clients’ wives) reported higher levels of stress than DER’s clients on average, with composite

stress scores of 2.96 compared to 2.63 among DER’s female clients, suggesting greater mental and emotional

health problems among this subsample. This is consistent with the conjugal contracts model described above,

with these women likely reliant on somewhat unreliable income transfers from their husbands’ agricultural

activities, and thus expressing elevated levels of perceived stress. We find that the program had beneficial

impacts on some aspects of women’s perceived stress, but these effects were concentrated only among these

other female household members that were not directly targeted by DER. In particular, we see that the

program reduced the composite perceived stress score by 0.6 points. Examining the different perceived

stress indicators individually, we also find that the program had beneficial impacts among the other female

household member of male DER clients in several of these different dimensions. In particular, we find that

the program decreased the frequency with which women felt control over important things in their life (by

0.3 points), increased the frequency of feeling confident in their ability to handle personal problems (by 0.5

points), and decreased the frequency with which women felt that difficulties were piling up so high that they

could not overcome them (1.2 points). This latter effect is especially pronounced, as the treatment effect is

about 35 percent of the average score among the control group, signifying a substantial reduction in women

feeling overwhelmed. Indeed, this improvement in perceived stress (frequently feeling difficulties were piling

up) as a result of the KhetScore program placement is more than enough to overcome the gap in perceived

stress between clients (average score of 2.70) and other household members (average score of 3.36). Among

female DER clients (Panel A), we observe that the program increased the frequency of feeling confident in

their ability to handle personal problems, and an increase in the frequency with which women felt that things

were going their way, though neither of these effects are statistically significant at conventional levels. The

program had negligible effects – both in terms of effect size and statistical significance – on clients feeling

incapable of controlling things or feeling that difficulties were piling up.

28



4.7 Summarization of KhetScore treatment effects

In sum, the results from the program are highly encouraging. They suggest that the program has wide-ranging

beneficial impacts for women, in terms of increasing access and utilization of credit, improving their ability to

contribute to household decision-making, their control over income and assets, and reducing their overall

levels of stress, thereby improving well-being. These latter two sets of results (those presented in Table 12 and

Table 13) also highlight the value of interviewing multiple family members within beneficiary households. Had

this research only focused on interviewing the h