統計学輪講 第19回

日時 2023年11月07日(火)
14時55分 ~ 15時45分
場所 経済学部新棟3階第3教室
講演者 武石 将大 (統計数理研究所)
演題 Hypothesis Testing for Treatment Effect Heterogeneity via an Adaptively Augmented Debiased Estimator
概要

This study develops a novel method for testing the existence of treatment effect heterogeneity without a parametric assumption.
Many existing nonparametric methods have either of the following drawbacks: poor performance when the dimension of personal attributes, possibly characterizing heterogeneity, is moderate and inadaptability to data with right-censored outcome.
Against this background, we first construct Wald-type test statistics based on the debiased estimator of variance of conditional average treatment effect function. Furthermore, to deal with complication caused by degeneracy of the aforementioned debiased estimator, we propose adaptively augmenting the original Wald statistics locally around the null. We show that the proposed test has the correct asymptotic size and and consistency property under the fixed alternative. The extension to data with right-censored outcome is also discussed.