Augmented direct learning for conditional average treatment effect estimation with double robustness

Inferring the heterogeneous treatment effect is a fundamental problem in many applications. In this paper, we focus on estimating Conditional Average Treatment Effect (CATE), that is, difference conditional mean outcome between treatments given covariates. Traditionally, Q-Learning based approaches estimate each outcome. However, they are subject to model misspecification. Recently, flexible one-step methods directly learn (D-Learning) CATE without specifications have been proposed. require specification of propensity score. We propose robust direct learning (RD-Learning), augment D-learning, leading doubly estimators effect. The consistency for our estimator guaranteed if either main or score correctly specified. framework can be used both binary and multi-arm settings general enough allow different function spaces incorporate generic algorithms. conduct thorough theoretical analysis prediction error using statistical theory under linear non-linear settings. effectiveness proposed method demonstrated by simulation studies real data example about an AIDS Clinical Trials study.