A Bayesian Additive Regression Tree Model for Learning Conditional Average Treatment Effects in Regression Discontinuity Designs
This paper introduces a new Bayesian model called BARDDT to estimate conditional average treatment effects (CATE) in regression discontinuity designs (RDD). It uses a variant of Bayesian additive regression trees, which adaptively partitions data based on covariate differences. This method avoids rigid assumptions found in previous frequentist approaches and provides flexible, data-driven CATE discovery.