[3G2-GS-2h] 機械学習：推論

[3G2-GS-2h-04] A Study on a Method for Estimating Conditional Average Treatment Effects Taking Account of Selection Bias Based on Causal Tree

〇Yuki Tsuboi¹, Yuta Sakai¹, Satoshi Suzuki¹, Masayuki Goto¹ (1. Waseda University)

Keywords:Causal Inference, Treatment Effect, Causal Tree, Selection Bias, Effect Verification

It is important for companies to verify the effects of their marketing measures and to make right decisions. To verify the effects from observational data correctly, they make use of causal inference. In recent causal inference, after allocating subjects to two groups and treating them differently, they often seek to estimate the Conditional Average Treatment Effect (CATE) to better understand causal mechanisms. CATE makes it possible to identify the group of users for whom it is effective to take measures. As a CATE estimation method, Causal Tree which has high interpretability and usefulness for analyzing the factors that affect measures, has been proposed. However, this method cannot be used when they allocate subjects to two groups on purpose due to selection bias. Therefore, we propose a method for estimating CATE taking account of selection bias based on Causal Tree. Finally, we evaluate the precision of CATE estimates by simulation experiments. In addition, we apply the proposed method to an actual data and show the usefulness of the proposed method.