Biostatistics short course: Bayesian Regression Trees
Jason Roy, PhD, Rutgers School of Public Health, will present an introduction to Bayesian regression trees, especially focusing on Bayesian Additive Regression Trees (BART). Bayesian regression trees can be thought of as priors for unknown regression functions. They are more flexible than parametric models and are quite easy to use in practice. This short course will cover several key topics, including an introduction to the basics of the models, in-depth coverage of how the priors balance flexibility and regularization, examples of the use of BART in missing data and causal inference problems, and implementation in R. It is assumed that participants have familiarity with the basics of Bayesian modeling.
Jason Roy, PhD, is professor of biostatistics and chair of the Department of Biostatistics and Epidemiology at Rutgers School of Public Health. He is interested in methodological research in developing flexible Bayesian methods for large, observational studies, especially data from EHR and mobile health. He is particularly interested in causal inference problems, where Bayesian nonparametric methods can be used in conjunction with g-computation. Much of his collaborative research is in pharmacoepidemiology. Roy was recipient of the 2020 Causality in Statistics Education Award from the American Statistical Association.