Spring Seminar Series - April 20, 2006
University of Minnesota
School of Statistics
College of Liberal Arts

Nonparametric Transformation Regression Models for Skewed Data with Heteroscedastic Variance

Xiao-Hua Andrew Zhou
Department of Biostatistics
University of Washington

Thursday, April 20, 2006
3:30 PM, 115 Ford Hall
Minneapolis, East Bank Campus
Social at 3:00 PM, 300 Ford Hall

Abstract

In this talk we introduce a new non-parametric heteroscedastic Transformation regression model that allows us to predict the expected value of an outcome of a patient with given covariates when the distribution of the outcome is highly skewed with a heteroscedastic variance. In our new model, we allow both the transformation function and the error distribution function to be unknown. We show that estimators for regression parameters, the expected value of the original outcome, and the transformation function converge to their true values at the same rate as that one can expect for a parametric model. In a simulation study, we demonstrate that our proposed nonparametric method is robust with little loss of efficiency. Finally, we apply our new model to a study on health care costs. This is a joint work with Huazhen Lin.