Robust Estimation of Causal Effects in Multilevel Observational Studies Under Cluster-Level Unmeasured Confounding

• In Event: Poster Session for Quantitative Methods and Statistical Theory

Abstract

Recently, there has been a growing interest in machine learning (ML) methods for causal inference. Almost all the ML methods have been studied under the assumption that all the confounders are measured. However, there is little research on handling omitted/unmeasured confounding bias in ML methods. This study provides three ML methods to estimate the average treatment effect and the conditional average treatment effect in the presence of cluster-level unmeasured confounding. We conduct a simulation study with two different types of multilevel data: two-level and cross-classified data. Our simulation study finds that our proposed ML methods not only eliminate the impact of cluster-level unmeasured confounders but also guarantee doubly robustness. An empirical example using ECLS-K data is provided.

Authors

• Youmi Suk, University of Virginia

• Hyunseung Kang, University of Wisconsin - Madison