Cleber Lack
FCUL / Ministério da Defesa-Brasil

Residual Analysis for generalized linear mixed model with binary responses is more difficult than for linear regression models. The ordinary residuals will not be normally distributed and, indeed, their distribution under the assumption that the fitted model is correct is unknown. Plots of ordinary residuals against fitted values or predictor variables will generally be uninformative.
A half-normal probability plot of the deviance residuals with a simulated envelope can be  useful both for examining the adequacy of model and  plot helps to highlight outlying residuals even though the residuals are not normally distributed.
In this study, we propose that the $kth$ ordered absolute residual is plotted against the value other than those cited in the literature and we propose use a different number of simulation to generate the envelope.
An application of the model to resignation from the Brazilian Army is given. It aims at identifying the variables intrinsically related to early exit. From 2011 to 2014, 16540 militaries were analysed. The sampled data was clustered by military's background (IME, AMAN, EsSEx and EsFCEx), military ranks (Lieutenant, Captain, Major, Lieutenant Colonel and Colonel). Also, each military had periodic evaluations mainly on semester basis (21 variables).

This is joint work with Helena Mouriño.