Model building in Semiparametric Models

 Thomas Scheike
 Department of Mathematical Sciences
 University of Aalborg
 Fredrik Bajers Vej 7 G
 9220 Aalborg
 Denmark 


The additive time-varying regression  model and its semiparametric
submodel has recently received much attention in the longitudinal
setting. In survival analysis models such as Aalen's additive risk
model and the Cox model are examples of such models in a counting
process setting. In this talk I will discuss how to make model
building in the context of these models. I will show how one may
decide if  covariate should be included with or without a
time-varying effect. Other more general questions may be answered
by similar techniques. The theoretical developments are based on a
study of the cumulative regression functions for which the
asymptotic distribution can be derived and simmulated.