Cox-Snell generalized residuals in the evaluation of model fitting

Several kinds of residuals have been proposed for survival models, the most suitable for this purpose are Cox and Snell (1968) generalized residuals. The objective of this work was to evaluate the adequacy of models by graphical diagnostics using Cox-Snell generalized residuals. To illustrate the theory three applications were considered. The first application sought to illustrate the heuristics by plotting ordered residuals from three distributions: normal (0,1), logistics (0,1) and extreme value (0,1), versus the expected order statistics of these residuals in consonance with the assumed distributions. The second application consisted of lifetime data of electric insulating, obtained by Nelson (1990). Starting from graphical diagnostics using Cox-Snell generalized residuals, it was found that the model appropriate for lifetime of electric insulating was log-normal. The third application referred to censored data of lifetime of patients, obtained by Collett (1994). The adequacy of several models was evaluated by Cox-Snell generalized residuals adapted for survival data. The results show that Weibull model was the most appropriate.