A1520
Title: Evidence aggregation with ignorance in mind: Learning what is not known for archetypes discovery
Authors: Davide Viviano - Harvard University (United States) [presenting]
Emily Breza - Harvard (United States)
Arun Chandrasekhar - Stanford (United States)
Abstract: When evaluating policy interventions, researchers often pursue two related goals: identifying which individuals or contexts benefit most, and determining whether patterns of treatment effect heterogeneity can be used to aggregate evidence across environments. A framework is developed that partitions observations, defined by individual and environmental characteristics, into groups within which treatment effects can be stably aggregated, while setting aside contexts in which extrapolation is unreliable and further evidence is needed. The procedure therefore learns both how to summarize heterogeneity and when researchers should admit ignorance. Finite-sample regret guarantees are derived for estimation and model selection, and inference procedures that quantify the value of follow-up data collection. The approach is illustrated by reanalyzing a multifaceted anti-poverty program implemented in six countries.
