Evidence aggregation in development economics via Bayesian hierarchical models

Abstract

It is increasingly recognized that translating research into policy requires aggregating evidence from multiple studies of the same economic phenomenon. This translation requires not only an estimate of the impact of an intervention across different contexts, but also an assessment of the generalizability of the evidence and hence its applicability to policy decisions in other settings. This thesis performs evidence aggregation using Bayesian hierarchical models, which both aggregate evidence and assess the true underlying heterogeneity across settings, for applications in development economics. Where necessary, the thesis develops new methods to aggregate evidence on certain measures of evidence currently neglected in the aggregation literature such as distributional treatment effects or risk ratios. The applications considered are randomized controlled trials of expanding access to microcredit and randomized access to vitamin A supplementation in developing nations.