Bayesian Regression Trees for Modeling the Health Effects of Environmental Stressors

• Activity Number: 601
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Gregory Watson* and Donatello Telesca
• Companies: UCLA and UCLA
• Keywords: Bayesian ; Spatial ; CART ; Nonparametric ; Tree

Abstract

Predicting and understanding the effect of environmental stressors on health outcomes such as hospital admissions and mortality present a number challenges including excess zeroes, nonlinear effects, interactions, variable selection and spatiotemporal dependence. This paper employs Bayesian regression trees with a zero-inflated sampling model, balancing prediction with regression. The trees recursively partition the covariate space, providing variable selection and incorporating nonlinear effects and interactions, while maintaining an interpretable inferential framework. The zero-inflated regression models at the tree leaves (terminal nodes) account for the excess zeroes and spatiotemporal dependence. Posterior inference is discussed in detail with a particular focus on scalability.