Glimmix software description

GLIMMIX is a powerful approach to segmentation based on latent class models. Mixture regression methods for simultaneous segmentation and estimation of regression models have recently received a great deal of attention in marketing research. In the approaches, consumers are grouped into a number of segments and ,at the same time, a regression model which relates a dependent variable to a set of predictors, is estimated within each segment. The approach has been shown to provide a very powerful approach to segmentation problems.

GLIMMIX implements this mixture of regressions methodology in a user-friendly way. It enables the estimation of such models for a great variety of data, such as brand choice and purchase frequency, pick any n and paired comparisons data, (inter)purchase times, and conjoint data, allowing for the specification of normal, Poisson, gamma, binominal distributions, and in version 2 also for multinominal distributions. The use of GLIMMIX is not limited to marketing research, but extends to business and economic research. But also to areas like psychology, sociology, anthropology, political science, etc.

The major 32-bits release offers different model specifications for different segments, and the possibility to get the posteriors for the whole dataset even when analysis is based on a subset of the data. In addition, expanded input data formats, binominal rating scale analysis, class profiling with descriptor variables and analysis adding random responders class have been included. The user interface and output representation have been improved, as well as the manual and online help facility.

New features in Glimmix 3.0

GLIMMIX 3 is upward compatible with GLIMMIX2.0. All files previously analysed with GLIMMIX 2.0 can be read in by GLIMMIX 3.0. But: since GLIMMIX 1.0 allowed for different input formats, the old definition files *.GL1, *.GL2, *.GL4 from GLIMMIX 1.0 cannot be used. Definitions thus have to be entered again;

GLIMMIX 3 allows you to analyze up to 150 independent variables and up to 50.000 subjects/records;

GLIMMIX 3 has a new feature for large datasets. If you analyze a large dataset, GLIMMIX takes a sample of the data, and computes posterior membership probabilities for the entire dataset, automatically uses the Generalized EM algorithm with a less strict convergence criterion, which speeds up the computations;

GLIMMIX has an improved computation of the standard errors of the estimates through a cross product of the first derivatives of the likelihood function, which is more accurate and more stable that the approximation to the Hessian used previously.

Ordering

For price information and ordering, please visit the Prices and Ordering Page