A0613
Title: Robust estimation of the quantile mediation treatment effect with machine learning
Authors: Martin Huber - University of Fribourg (Switzerland)
Yu-Chin Hsu - Academia Sinica (Taiwan)
Yu-Min Yen - National Chengchi University (Taiwan) [presenting]
Abstract: The quantile mediation treatment effect estimation is studied using a double/debias estimator. The nuisance parameters of the proposed estimator are estimated with a machine-learning method, and cross-fitting is used to reduce estimation bias from overfitting and/or regularization of the machine learner. A multiplier bootstrap procedure is then used to conduct statistical inferences. Relevant uniform consistency of the proposed estimator and uniform validity of the multiplier bootstrap procedure is established. The performance of the proposed estimator is illustrated by conducting a simulation. Then a sensitivity analysis for the proposed estimator and a possible extension to allow for sequential mediators are discussed.
