Identification of Treatment Effect Heterogeneity as a Variable Selection Problem

• Activity Number: 547
• Type: Invited
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Social Statistics Section
• Author(s): Marc Thomas Ratkovic*+ and Kosuke Imai
• Companies: Princeton University and Princeton University
• Keywords: causal inference ; LASSO ; optimal treatment regimes ; randomized experiments ; Support Vector Machines

Abstract

Identification of treatment effect heterogeneity plays an essential role in a number of situations that are frequently encountered by applied researchers. They include (1) selecting the most effective treatment from a large number of available treatments, (2) ascertaining subpopulations for which a treatment is most effective, (3) designing optimal treatment regimes for individuals, (4) testing the existence of heterogeneous treatment effects, and (5) generalizing causal effect estimates obtained from an experimental sample to a target population. In this paper, we propose a method that combines optimal classification with variable selection to identify heterogeneous treatment effects when the outcome is binary. Specifically, we adapt the Support Vector Machine classifier by placing a separate sparsity constraint over the causal heterogeneity parameters of interest. As confirmed in simulation studies, the proposed method tends to yield lower false discovery rate than some commonly used alternatives. For empirical illustrations, we apply the proposed method to randomized field experiments from political science and economics.