Estimating Heterogeneous Causal Mediation Effects with Bayesian Decision Tree Ensembles

The causal inference literature has increasingly recognized that targeting treatment effect heterogeneity can lead to improved scientific understanding and policy recommendations. Similarly, studying the causal pathway connecting the treatment to the outcome can be useful. We address these problems in the context of causal mediation analysis. We introduce a varying coefficient model based on Bayesian additive regression trees to estimate and regularize heterogeneous causal mediation effects. Even on large datasets with few covariates, we show LSEMs can produce highly unstable estimates of the conditional average direct and indirect effects, while our Bayesian causal mediation forests model produces stable estimates. We find that our approach is conservative, with effect estimates "shrunk towards homogeneity." Using data from the Medical Expenditure Panel Survey and empirically-grounded simulated data, we examine the salient properties of our method. Finally, we show how our model can be combined with posterior summarization strategies to identify interesting subgroups and interpret the model fit. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.