Essays on generalizability and evidence-use in policy

To increase evidence-use in policy, it is important to understand both the generalizability of evidence, and the existing use of evidence for policy decisions. This thesis comprises three papers on this topic. Chapter 1 studies evidence-use in policy, focusing on one of the most heavily evaluated anti-poverty programs — Conditional Cash Transfers (CCTs). Using a novel dataset of 128 program evaluations of CCTs in Latin America and the Caribbean mapped to policy spending on the evaluated programs, I find a robust zero relationship between research results and spending. The only exception is when evaluations are timely and politically aligned. When evaluations are released within four years of the effect year and can be attributed to the political party in power, there is a positive and significant relationship between evaluation outcomes and spending. Chapters 2 and 3 examine the generalizability of evidence, using Bayesian hierarchical models to aggregate the evidence-base on gender differences in altruism and overconfidence. In chapter 2, I find that women give three percentage points more than men in dictator games, but this estimate is likely to be an upper bound due to publication bias. In chapter 3, joint with Oriana Bandiera, Barbara Petrongolo, and Nidhi Parkeh, we find that while experts believe that men are overconfident and women are underconfident, the literature suggests that both men and women are overconfident.