Estimation and Inference in Nonparametric Cox-models: Time Transformation Methods

Abstract

In this paper generalization of the Cox proportional hazards regression model to a completely nonparametric model with an unspecified smooth covariate function is studied. A class of methods for Cox-regression called time transformation methods are defined, and a new method for nonparametric Cox-regression in this class is in particular studied. It turns out that this method enjoys a number of useful properties. Ways of doing inference and model checking in nonparametric Cox-models are also discussed, and a brief overview and comparison of methods for nonparametric Cox-regression is given.