Research

This study constructs a novel dataset on the ethnicity of individuals in an ethnically diverse country in sub-Saharan Africa. We measure ethnicity using a machine-learning algorithm that exploits variation in surnames across ethno-linguistic groups. We apply this approach to voter registration data from Uganda's 2016 general election. The resulting data capture local variation in ethnicity over a wide geographic scale. We pair these data with election outcomes from polling stations throughout Uganda to estimate the relationship between ethnicity and voting behavior. Our regression analyses both control for location and include interactions between ethnic groups within the same polling station. Local variation in ethnicity is associated with voting behavior at the level of the polling station, and these relationships vary with the presence of other ethnic groups at the polling station. The results suggest the importance of studying ethnic voting using local variation in ethnicity at scale.

Published journal article

Accepted manuscript © 2025. This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/.

This paper studies whether the absence of locally adapted seed varieties constrains the productivity and incomes of farm households residing in small, agro-ecological niches. We empirically examine the disruption of the maize seed market in Western Kenya that took place when public sector foundation breeding and social impact investment capital came together and allowed a local seed company to expand and target a niche area with adaptively-bred maize varieties. The three-year randomized controlled trial reveals that these seed varieties increased farmer yields and revenues, both for better-resourced farmers (who used non-adapted hybrids and fertilizer prior to the intervention) as well as less well-resourced farmers (who did not). This theoretical and empirical evidence suggests news ways for thinking about seed systems in areas typified by high levels of agro-ecological heterogeneity.

Best Publication of 2022 Award -- Agricultural & Applied Economics Association Africa Section

Published journal article

Accepted manuscript © 2022. This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/.

ATAI Summary

AMA Policy Brief

We study the adoption of an electronic voucher subsidy for agricultural inputs by members of farmer organizations in Uganda. We randomly assign farmer organizations to have their members offered a status quo subsidy, a higher subsidy, or no subsidy. Adoption does not increase significantly due to the status quo subsidy but increases substantially due to the high subsidy. For farmers who at baseline did not use the inputs subsidized by the voucher, adoption increases with their farmer organization's leader's experience with the inputs and social similarity to the member. The results are consistent with material and information constraints limiting adoption of the electronic voucher subsidy, particularly for the intended beneficiaries of the subsidy program: farmers with limited experience with improved inputs, for whom a subsidy lowers the cost of learning by doing and may induce sustained technology adoption. While digital payment schemes can lower the costs of delivering payments from governments to people, they do not by themselves eliminate the need to tend to basic economic constraints of liquidity and information. 

Full paper

We show that large-scale public works programs targeted at rural women are not sufficient to offset the decline in women’s labor force participation driven by broader structural transformation. We examine the differential effects of agricultural mechanization on women’s employment by leveraging the staggered roll-out of India’s National Rural Employment Guarantee Scheme (NREGS), the world’s largest public works program. By exploiting spatial and temporal variation in district soil texture and national trends in mechanization, we find that NREGS did not mitigate women’s exit from the labor force spurred by agricultural mechanization. Instead, the mechanization-induced decline in women’s labor force participation is larger in NREGS than in non-NREGS districts. On the intensive margin, NREGS did not influence the reduction in days worked by women per week. These findings highlight the limits of public works programs in counteracting gender-biased impacts of structural change.

We study how religion influences economic behavior using a randomized control trial in Kisumu, Kenya. Participants attended workshops hosted by either Pentecostal or mainline Christian churches, followed by surveys and incentive-compatible choice experiments to measure altruism and investment behavior. Pooling both treatments, we find that attending the services led to decreased self-reported self control and grit while reducing contributions to a communal entrepreneurship fund by 16.3% relative to the control. Examining denominational differences, we find evidence that a shifting landscape toward Pentecostalism matters: Pentecostal messaging appears to have decreased the amount of financial transfers received and given by the respondent and further increased interest bearing savings among men and intrinsic religiosity among women. These findings provide causal evidence that Christian religious participation promotes individualism at the expense of collectivism, while demonstrating that sub-Saharan Africa's changing denominational composition may further amplify such effects with economically meaningful implications for investment behavior.

AEA RCT Registry

This paper studies whether the adoption of a technology that increases the production of a staple crop differs between households who are buyers, sellers, or self-sufficient with respect to the staple crop. I develop a theoretical model that shows that if buying or selling staples incurs large fixed costs, technology adoption varies with market participation; if fixed costs are small, however, technology adoption does not vary with market participation. Whether technology adoption varies by market participation has implications for how to target interventions promoting technology adoption such as input subsidy programs. I estimate how adoption varies with market participation using data from a randomized controlled trial of high-yielding varieties of maize developed for western Kenya. Treatment effects across market participation groups are neither large nor statistically significant. These results suggest that transaction costs in output markets are not large enough to shape the pattern of adoption of a production technology.

Full paper

This paper estimates how sales of staple food crops differ between female and male members of agricultural households. We estimate this effect by controlling for fixed effects for household and crop using panel data from the Government of Uganda for the Living Standards Measurement Survey – Integrated Survey of Agriculture from 2009-2015. Sales are greatest for harvests controlled by a single male, followed by harvests controlled by a female and male and harvests controlled by a single female. Some differences between genders are due to the condition in which the crop is harvested. We also estimate that restricting survey respondents to report a single decision-maker creates large measurement error in decision-maker gender as well as some bias when estimating the relationship between market participation and gender. Our results highlight differences in output market participation by gender as well as the importance of accounting for all household members involved in decision-making processes. 

Many governments in sub-Saharan Africa subsidize the use of agricultural inputs by smallholder farmers in order to increase food production and food security. These subsidy programs have been studied empirically in country-specific contexts, yet we lack a theoretical understanding of how program impacts vary with program targeting and subsidy level. I use numerical simulation to study a policymaker’s optimal targeting and design of a subsidy program given farmers’ best responses of input use under different subsidy levels. I find that the policymaker’s optimal targeting and subsidy level critically depend on the distribution of land and financial wealth across households.