Research

My primary fields are Education, Development, Labor, and Gender.

Job Market Paper

Costs or Returns? Why Students Specialize in Cognitive vs Socio-Emotional Skills
Teachers allocate scarce resources across multiple cognitive and non-cognitive skills. When a student is weak in a given skill, a key question is why: is it primarily because it's not valued, or is it because it's hard to learn? I develop a model where parents choose inputs to produce skills, maximizing utility over the child's resulting skill bundle. The model delivers a simple diagnostic—the return-level slope—that links skill levels to marginal returns for improvement. I implement the diagnostic using new data from 3,404 parents across five Indian private schools. Parents rate their child's level (0-100) across nine skills covering academic, emotional, and social traits, and then rank the skills by which they most want improved, at the margin. I regress ranks on levels to estimate the return-level slope (i) within each skill across students and (ii) within each student across skills. The slope is negative when parents most value improving weaknesses, and positive when they value improving strengths. I find slopes are negative on average, implying parents desire well-rounded children, and, through the lens of the model, that relative production costs are more uneven than returns. To test the model's predictions, I run a teacher-facing information experiment that randomizes access to a web portal displaying parent-reported levels and priority ranks for students in their class. In the framework, this acts as a production-side shock—teachers reduce the effective cost of inputs (e.g., materials) and tailor pedagogy toward parent-prioritized dimensions. Consistent with the model, treatment increases endline achievement and lowers priority ranks for the parent's top-ranked skill, shifting the rank-level slope upward. This framework provides a low-cost way to guide personalizing education beyond level-based grouping and helps choose policy levers when supporting weak skills: moving the production frontier versus shifting perceived returns.

Works in Progress

What Do People Want? with Daniel Benjamin, Kristen Cooper, Ori Heffetz, and Miles Kimball
Philosophical perspectives on human desires and values vary; economic theory-driven measurement techniques can provide relevant empirical evidence. We elicited over half a million stated preference choices over 126 dimensions or "aspects" of well-being from a sample of 896 online respondents. We also elicited, via self-reported well-being (SWB) questions, respondents' current levels of the aspects. From the stated preference data, we estimate for each aspect its relative marginal utility per point on our 0-100 response scale. We validate these estimates by comparing them to alternative methods for estimating preferences, and we offer a range of estimates between those that take self-reports at face value and those that (over-)correct for potential social-desirability reporting bias. Our findings suggest that our respondents want, first and foremost, three basic things: family, money, and health (not necessarily in this order). While commonly studied concepts such as happiness, life satisfaction, where life ranks on a ladder, and meaning, are all important, respondents place the highest marginal utilities on aspects related to family well-being and health, and financial freedom and security. We document substantial heterogeneity in preferences across respondents within (but not between) demographic groups, with current SWB levels accounting for a significant portion of the variation.

Explainable AI and Human Decision Making: Preferences, Beliefs, and Biases with Peter Bergman and Kadeem Noray
Increasingly, AI is being used as a gatekeeper to key areas that affect economic mobility. AI is screening applicants for jobs, loans, healthcare and housing. Generative AI has accelerated this trend; its pre-trained models can readily be deployed across a variety of contexts. However, there are concerns that these models discriminate against protected groups. We construct a model of applicant selection that distinguishes between different forms of discrimination -- taste-based discrimination, statistical discrimination, and biased beliefs -- at the employer or recruiter level. We collect data that allow us to record resume review and hiring outcomes for applicant profiles and overcome the selection issue of observing hiring outcomes only for interviewed applicants. We compare AI decision making to the distribution of human decision makers and use the model to simulate policies such as blinding resume characteristics ("ban the box") and to build non-discriminatory screening algorithms.

Signaling in Female Education with Akanksha Vardani
We study the role of labor market and marriage market considerations in motivating investment in female education. We replicate previous work contrasting the signaling and human capital accumulation models of educational attainment conducted on a US sample in the developing country contexts of India and Zambia. In India, we find that increased access to secondary school shifts the entire distribution of educational attainment upwards, with more ambiguous effects in Zambia. We extend the analysis to include marriage market considerations, and test to see if shifts in educational attainment depend on marriage payment norms. We find that shifts are attenuated among populations that practice marriage payments.

Community and Household Networks and Women's Workforce Entry with Akanksha Vardani

Seniority and the Gender Wage Gap with DongIk Kang

Empowering Youth with Digital Skills: A Large-Scale Clustered Randomized Intervention in Kenya with Palaash Bhargava, Daniel Chen, Tommaso Batistoni, Ken Maina