The GENMOD and GLIMMIX procedures can fit generalized linear models and estimate the parameters by maximum likelihood. For multinomial data, the GENMOD procedure fits cumulative link models for ordinal data. The GLIMMIX procedure fits these models and generalized logit models for nominal data.
When data are correlated, you can use the REPEATED statement in the GENMOD procedure to fit marginal models via generalized estimating equations. A working covariance structure is assumed, and the standard errors of the parameter estimates are computed according to an empirical ("sandwich") estimator that is robust to the misspecification of the covariance structure. Marginal generalized linear models for correlated data can also be fit with the GLIMMIX procedure by specifying the random effects as R-side effects. The empirical covariance estimators are available through the EMPIRICAL= option in the PROC GLIMMIX statement. The essential difference between the estimation approaches taken by the GLIMMIX procedure and generalized estimating equations is that the latter approach estimates the covariance parameters by the method of moments, whereas the GLIMMIX procedure uses likelihood-based techniques.
The GENMOD procedure supports nonsingular parameterizations of classification variables through its CLASS statement. The GLIMMIX procedure supports only the standard, GLM-type singular parameterization of CLASS variables. For the differences between these parameterizations, see the section Parameterization of Model Effects, in Chapter 19: Shared Concepts and Topics.