ABSTRACT
Methods for model checking are as important for parametric models as they are in the Cox regression model. In this chapter, standardised, Cox-Snell, deviance, martingale and score residuals, are defined for parametric accelerated failure time models. Ways in which these residuals can then be used to identify observations that are not well fitted by the model are summarised, again with an emphasis on graphical procedures. Methods for identifying observations that have an undue influence on a particular parameter estimate or the set of estimates of the coefficients in a risk score are described. This chapter also includes a description of how an observed survivor function can be compared with that for a parametric model, and a method for testing the assumption of proportional hazards in a Weibull model.
