When a generalized linear model is formed with distributions other than the binary, binomial, or multinomial, you can use the GENMOD and GLIMMIX procedures for parameter estimation and inference.
Both procedures can accommodate correlated observations, but they use different techniques to accomplish this goal. The GENMOD procedure can fit correlated data models via generalized estimating equations that rely on a first- and second-moment specification for the response data and a working correlation assumption. With the GLIMMIX procedure, you can model correlations between the observations by (1) specifying random effects in the conditional distribution that induce a marginal correlation structure or (2) direct modeling of the marginal dependence. The GLIMMIX procedure employs likelihood-based techniques in parameter estimation.
The GENMOD procedure supports a Bayesian analysis through its BAYES statement.
With the GLIMMIX procedure you can vary the distribution or link function on an observation-by-observation basis.
To fit a generalized linear model with a distribution that is not available in the GENMOD or GLIMMIX procedure, you can use the NLMIXED procedure and code the log-likelihood function of an observation with SAS programming statements.