Fall Seminar Series - October 13, 2005
University of Minnesota
School of Statistics
College of Liberal Arts

"Robustifying" Parametric Models via Mixtures of Polya Tree Priors

Tim Hanson
Division of Biostatistics
University of Minnesota

Thursday, October 13, 2005
3:30 PM, 115 Ford Hall
Minneapolis, East Bank Campus
Social at 3:00 PM, 300 Ford Hall

Abstract

Mixtures of Polya trees models are straightforward to code and provide a highly flexible alternative when a parametric model may only hold approximately. In this talk, I provide computational strategies for obtaining semiparametric inference for mixtures of finite Polya trees models given a standard parameterization, including models that would be difficult to fit using Dirichlet process mixtures. Recommendations are put forth on choosing the level of a finite Polya tree and model comparison is discussed. Several examples demonstrate the utility of Polya tree modeling including data on bivariate (CD4,CD8) counts fit to a semiparametric linear mixed model; the classic V.A. lung cancer study data fit to proportional hazards, proportional odds, and accelerated failure time models; and serology scores modeled with a stochastic order constraint.