Fall 2002 Seminar Series - November 21, 2002
University of Minnesota
School of Statistics
College of Liberal Arts

Semiparametric lack-of-fit tests for the Aalen hazard regression model

Birgit Grund
School of Statistics
University of Minnesota

Thursday, November 21, 2002
4:00 PM, 115 Ford Hall
Minneapolis, East Bank Campus
Social at 3:30 PM, 300 Ford Hall

Abstract

     The goal is to investigate the influence of covariates on the hazard function for time-to-event data; covariates are measured only once, at baseline. In the Aalen model, contributions of covariates to the hazard function are additive, and may change over time. Regression coefficients are curves over time. Usually, these regression coefficient curves are estimated nonparametrically; there are no model assumption on their shape. Breaking free of parametric model assumptions comes at a price; nonparametric estimates for the regression coefficient curves are difficult to interpret, and they require a large sample size.

In this talk, I will discuss bootstrap tests for the hypothesis that some of the regression coefficients are constant versus a nonparametric alternative. This is very different from classical model choice, where we test whether certain coefficients are zero (corresponding covariates could be dropped from the model). In essence, we are testing whether a certain curve can be adequately estimated under a given parametric model, or whether we should go to the expense of nonparametric curve estimation. The talk will be short on theory and long on bootstrap (censored data) and simulations.