Bayesian Nonparametric Modeling for Causal Inference

Researchers have long struggled to identify causal effects in non-experimental settings. Many recently-proposed strategies assume ignorability of the treatment assignment mechanism and require fitting two models – one for the assignment mechanism and one for the response surface. We propose a strategy that instead focuses on very flexibly modeling just the response surface using a Bayesian nonparametric modeling procedure, Bayesian Additive Regression Trees (BART). BART has several advantages: it is far simpler to use than many recent competitors, requires less guess-work in model fitting, handles a large number of predictors, yields coherent uncertainty intervals, fluidly handles continuous treatment variables and missing data for the outcome variable. BART produces more efficient estimates in the non-linear situations tested in our simulations compared to propensity score matching, propensity-weighted estimators, and regression adjustment. Further, it is highly competitive in linear settings with the “correct” model, linear regression. Correspondence should be addressed to the first author at Columbia University, School of International and Public Affairs, 420 West 118th St., 1425 IAB, New York NY 10027