GLIMMIX: Software for estimating mixtures and mixtures of generalized linear models

GLIMMIX is a commercial WINDOWS-based computer program that implements the EM algorithm (Dempster, Laird and Rubin 1977) for the estimation of finite mixtures and mixtures of generalized linear models. The program allows for the specification of a number of distributions in the exponential family, including the normal, gamma, binomial, Poisson, and multinomial distributions. For each of those distributions, a variety of link functions can be specified to relate the expectation of the dependent variable to a linear predictor. Several statistics, including AIC, CAIC and BIC are computed to aid in model selection (cf. Akaike 1974; Bozdogan 1987), missing values are accommodated, and posterior membership probabilities are computed For cases, included or not included in the analysis. Simple discriminant type models dealing with concomitant variables to describe the classes are supported, and a random responder class can be added to the model. Various graphs are provided. A demonstration version of the program can be obtained from http://www/gamma.rug.nl. Before providing some details on the GLIMMIX software, a brief review of a few relevant issues in Mixture modelling are provided