B0755
Title: Confounder selection with Bayesian decision tree ensembles
Authors: Chanmin Kim - SungKyunKwan University (Korea, South) [presenting]
Abstract: An approach is presented to address the growing challenge of analyzing observational studies. The approach focuses on identifying the necessary covariates to establish the assumption of ignorable treatment assignment for estimating causal effects. It employs a Bayesian nonparametric method that tackles this challenge through three key aspects. Firstly, it gives priority to including adjustment variables based on established principles for selecting confounders. Secondly, it allows for the estimation of causal effects by considering the intricate relationships among confounders, exposures, and outcomes. Lastly, it produces causal estimates that account for the uncertainty surrounding the confounding nature. The method utilizes multiple Bayesian additive regression tree models that share a common prior distribution. It accumulates posterior selection probability for covariates associated with both the exposure and the outcome of interest. Various simulation studies demonstrate that the proposed method performs favourably compared to other similar methods across different scenarios. The approach is applied to examine the causal effect of SO2 emissions from coal-fired power plants on ambient air pollution concentrations. The findings provide compelling evidence of a causal relationship between SO2 emissions and ambient particulate pollution over consecutive years.
