A0212
Title: Bayesian nonparametric methods for causal effects with intermediate variables
Authors: Chanmin Kim - SungKyunKwan University (Korea, South) [presenting]
Abstract: Principal stratification analysis is a method used to estimate causal effects by examining the relationship between treatment and an intermediate variable (such as post-treatment outcomes). However, when the intermediate variable is continuous, parametric modeling methods struggle to capture the complex relationship between the variables. Moreover, separately estimating the outcome and intermediate models leads to uncertainty in the final causal effect estimation. To address these challenges, we propose a fully Bayesian method that uses Bayesian nonparametric models to flexibly estimate all intermediate, outcome, and propensity score models. This method is applicable in both specific and broad confounding situations. We demonstrate the proposed method's performance through a series of simulation studies and apply it to examine the impact of the NOx abatement device (scrubber) installed in US coal-fired power plants on surrounding Ozone concentrations, considering the relationship between the scrubber and NOx emissions from various perspectives.
