We extend Bayesian Additive Regression Trees to the setting of density regression, resulting in an accurate and efficient sampler with strong theoretical guarantees.

We present the FLAME package for performing fast, accurate matching for observational causal inference.

We use Bayesian Additive Regression Trees to analyze data from the United Network for Organ Sharing database. We find that ABO incompatible infant heart transplant does not affect post-transplant survival, incidence of rejection, or postoperative length of stay, and that it therefore remains a viable and important strategy for increasing the size of the infant donor pool.

We find that receiving multiple transplants for pediatric patients is associated with increased post-transplant mortality, but not increased acute rejection episodes.

We propose a method for matching in the presence of network interference. To reduce bias, we match units with similar neighborhood graphs, prioritizing similarity in subgraphs learned to be more important.

We match units via axis-aligned hyper-boxes that encompass space where the treatment effect is nearly constant. The resulting treatment effect estimates are therefore both case-based and interpretable.

We propose a latent-variable extension to BART that is capable of performing highly-flexible density regression. We target estimation of quantile treatment effects and apply our methodology to a real-world example estimating returns to education.

An R package implementing interpretable matching algorithms for causal inference. These algorithms match units via a learned, weighted Hamming distance that determines which covariates are more important to match on.