Some diagnostics for binary logistic regression via smoothing

SUMMARY The paper discusses some diagnostic tools for binary logistic regression which use smoothing techniques. Smoothed binary data are employed to devise a battery of diagnostic plots, from simple procedures which indicate only the adequacy or inadequacy of a maximum likelihood logistic fit, to partial residual plots and delta local deviance plots which reveal the type of model inadequacy, suggesting perhaps the addition of nonlinear terms or interactions. Properties of the methods are studied by simulation. The diagnostic tools are illustrated on a set of data in which the probability of having telephone service is related to the demographic variables income and education.