A Robust Method for Estimating Individualized Treatment Effect
share
We consider an additive model with a main effect and effects from multiple treatments. Our goal is to estimate the heterogeneous treatment effects among patients. Traditionally, one can fit standard regressions in which models for the main effect and the treatment effects are specified. However, mis-specification of either the main or the treatment effects could severely undermine the estimation. A set of recent proposals directly estimate the treatment effect, avoiding the potential mis-specification issue for the main effect. However, performance of these methods rely on either a known or accurate estimator of the propensity score. In this paper, we propose a doubly robust direct learning method (RD-Learning) to estimate the treatment effect. The double robustness comes form the fact that it is robust to the two issues of (1) main effect mis-specification and (2) inaccurate propensity score estimates. As long as these two do not occur at the same time, our estimate is consistent and has a smaller variance than competing methods. It can be used in both the binary and the multi-arm settings. As a by-product, we develop a competitive statistical inference tool for the treatment effect, assuming the propensity score is known. We provide theoretical insights to the proposed method using risk bounds under both linear and non-linear settings. Our method is further demonstrated by simulation studies and a real data example.
READ FULL TEXT