ABSTRACT
Model t has traditionally been assessed by means of summary test statistics or by residual analysis. Since the analysis of t is so important to the modeling process, we shall discuss four categories of t assessment:
Traditional t tests• Hosmer-Lemeshow GOF test• Information criteria tests• Residual analysis•
Fit statistics were rst developed for the generalized linear models (GLM) parameterization of logistic regression. They were rst integrated into the GLIM statistical package, which was written in the early 1970s. Recall from earlier discussion that the GLM approach uses an iteratively re-weighted least squares (IRLS) algorithm to estimate parameters, standard errors, and other model statistics. The results are identical to the full maximum likelihood method of estimation. GLIM was the rst statistical software to incorporate a full set of t statistics and residual analysis for the various GLM family models.
