A1408
Title: Denoised IPW-lasso for heterogeneous treatment effect estimation in randomized experiments
Authors: Mingqian Guan - The University of Osaka (Japan) [presenting]
Komei Fujita - CyberAgent (Japan)
Naoya Sueishi - Kobe University (Japan)
Shota Yasui - CyberAgent (Japan)
Abstract: A new method is proposed for estimating conditional average treatment effects (CATE) in randomized experiments. Inverse probability weighting (IPW) is adopted for identification; however, IPW-transformed outcomes are known to be noisy, even when true propensity scores are used. To address this issue, a noise reduction procedure is introduced and a linear CATE model is estimated using Lasso, achieving both accuracy and interpretability. It is theoretically shown that denoising reduces the prediction error of the Lasso. The method is particularly effective when treatment effects are small relative to the variability of outcomes, which is often the case in empirical applications. Applications to the get-out-the-vote dataset and the Criteo Uplift Modeling dataset demonstrate that the method outperforms fully nonparametric machine learning methods in identifying individuals with higher treatment effects. Moreover, the method uncovers informative heterogeneity patterns that are consistent with previous empirical findings.
