A0782
Title: Bayesian ensemble learning for principal causal effects
Authors: Chanmin Kim - SungKyunKwan University (Korea, South) [presenting]
Corwin Zigler - Brown University (United States)
Abstract: Principal stratification analysis assesses how the causal effects of a treatment on a primary outcome vary across different strata of units, which are defined by their treatment effect on an intermediate variable. This task is significantly complicated when the intermediate variable is continuous, resulting in an infinite number of basic principal strata. To address this, a Bayesian nonparametric approach is used to flexibly evaluate treatment effects across these strata. The approach employs Bayesian causal forests (BCF), which simultaneously specify two Bayesian additive regression tree models: one for principal stratum membership and one for the outcome conditional on principal strata. BCF's ability to capture treatment effect heterogeneity is particularly useful for assessing treatment effects across the continuum of continuously scaled principal strata. Additionally, BCF offers benefits in targeted selection and regularization-induced confounding. The effectiveness of the method is shown through a simulation study, and this methodology is applied to investigate how the causal effects of power plant emissions control technologies on ambient particulate pollution vary with the technologies' impact on sulfur dioxide emissions.
