Assessing goodness-of-fit of parametric regression models for lifetime data-graphical methods

Graphical methods are often used to check goodness-of-fit of models to data. It is common to plot residuals against a reference distribution so that when the model fits the data, the configuration should be close to a straight line. Since the resemblance to a straight line is often unclear, it has been suggested to add simulated envelopes within which the configuration is expected to lie. The implementation of this method for survival data analysis is not straightforward. In this paper we point out the difficulties which arise in constructing envelopes on residual plots for randomly censored data. Methods are suggested to deal with the problems and evaluated; they are illustrated by simulated data and on a follow-up study of myeloma.