Job Market Paper

Program Evaluations and Policy Spending (Draft coming soon)

Despite the rise of program evaluation, there is limited rigorous evidence on the relationship between causal estimates of impact and policy spending. This paper examines the systematic relationship between program evaluations and policy spending, in the context of Conditional Cash Transfers in Latin America and the Caribbean. Using a novel dataset of 128 program evaluations mapped to spending, I consider patterns of policy spending consistent with responses to individual evaluations, and to cumulative bodies of evidence. I find a robust zero association between individual program evaluations and policy spending. More statistically significant, larger magnitude, or more surprising results do not correspond with larger changes in spending. Moreover, stronger cumulative evidence on impact, estimated using a Bayesian hierarchical model, does not correspond with higher policy spending. As these evaluations are often directly or indirectly facilitated by policymakers, this result seems unlikely to be driven by lack of policy relevance, and instead suggests the presence of either inefficiencies or constraints. To make sense of the average zero relationship, I explore patterns of policy spending for subsets of evidence that are more credible, actionable, or generalizable. I find that when evaluations are time-actionable and political constraints are low, there is a positive association between causal estimates of impact and spending. This suggests that the timeliness of publication is an overlooked channel in the evidence to policy pipeline.

Works in Progress

Targeting Barriers to Gender Evidence Use, with Michael O'Sullivan and Léa Rouanet

Peer learning and Productivity across NGOs, with Jack Thiemel

Working Papers

Women and Cash transfers: how program design and local conditions relate to causal estimates of impact, with Gabriela Diaz-Pardo.  (link to paper)

In this paper, we conduct a systematic review of the literature on the impact of cash transfers on women’s employment and empowerment. We construct a dataset of 265 impacts of cash transfers on adult women across 56 studies and 30 programs in Lower and Middle-income countries. Our dataset is the first that matches estimated treatment effects with harmonised information on the design of cash transfer programs, including the transfer size, payment methods and frequencies, and program conditionalities. Across studies, we find that cash transfers have a positive and insignificant impact on women’s employment and empowerment. We use our data to explore how the impact of cash transfers differs by program design features and baseline country conditions, including local labor market structures and gender social norms. We find that cash transfers have a larger impact on women’s labor force participation when they are larger in size, and when there is a higher proportion of women who work in formal employment before the program evaluation. Overall, our results suggest that cash transfers have more meaningful impacts on women’s employment and empowerment when pre-existing gender constraints are low. Our findings highlight the importance of interpreting estimated program impacts in the context of country-level conditions and program design.

Gender differences in altruism: a Bayesian hierarchical analysis of dictator games  (link to paper)

I aggregate evidence on gender differences in dictator game giving from experiments published in all working papers and peer-reviewed journals since 1990. Using a two-stage Bayesian hierarchical model, I find that on average women give around 3 percentage points more than men in studies of dictator games. I show that while this estimate is smaller than that found in previous studies, it is likely to be an upper bound estimate due to publication bias. Using a truncated selectivity model, I estimate the conditional probability of publication as a function of experiment results. My findings suggest that experiments that find positive results (i.e. women contribute more than men) and are statistically different from zero at the 5% level, are around 13 times more likely to be published than statistically significant and negative results.

Published Papers

Men are from Mars and Women too: a Bayesian meta-analysis of overconfidence experiments, with Oriana Bandiera, Barbara Petrongolo and Nidhi Parekh.  Economica, 2022  (link to paper)

Gender differences in self-confidence could explain women's under-representation in high-income occupations and glass-ceiling effects. We draw lessons from the economic literature via a survey of experts and a Bayesian hierarchical model that aggregates experimental findings over the last 20 years. The experts’ survey indicates beliefs that men are overconfident and women underconfident. Yet the literature reveals that both men and women are typically overconfident. Moreover, the model cannot reject the hypothesis that gender differences in self-confidence are equal to zero. In addition, the estimated pooling factor is low, implying that each study contains little information over a common phenomenon. The discordance can be reconciled if the experts overestimate the pooling factor or have priors that are biased and precise. 

Pre-doctoral work

Gender-oriented languages and Female Labour Force participation: Evidence from sub-Saharan Africa, CSAE working paper (link to paper)