A1501
Title: Double/debiased machine learning estimation of the conditional tail average treatment effect in mediation analysis
Authors: Yu-Min Yen - National Chengchi University (Taiwan) [presenting]
Abstract: A double/debiased machine learning (DML) estimator for the conditional tail average treatment effect (CTATE) in mediation analysis is studied. The CTATE captures heterogeneity and provides aggregated local information on treatment effects across different quantile levels. It is closely related to the concepts of second-order stochastic dominance and the Lorenz curve, making it a valuable tool for policy evaluation. At a given quantile level, the CTATE is decomposed into indirect and direct effects, which facilitates analysis of how the treatment and mediator jointly contribute to the causal impact. Identification results are established and a DML estimator based on the joint influence function of the quantile and the conditional tail expectation of potential outcomes is developed. Monte Carlo simulations are conducted to evaluate the performance of the proposed approach. Its usefulness is further illustrated through an empirical application that examines the effects of a job training program using data from the National Job Corps Study.
