Bayesian decision trees and their ensembles
Michael J. Daniels, Antonio R. Linero, Jason A. Roy · 2023
Bayesian additive regression tree (BART) models are powerful tools for both estimating the outcome regression and propensity score. They are also widely accessible to practitioners due to the availability of easy-to-use software packages, several of which focus on causal inference. We review BART (from prior specification to posterior computation) and show how BART can be applied to causal inference. We also discuss the phenomenon of regularization induced confounding (RIC) and discuss the use of Bayesian causal forests (BCFs) to mitigate RIC.