Models for Categorical Dependent Variables
Part 2. Marketing Research
Published Online: 15 DEC 2010
Copyright © 2011 John Wiley & Sons, Ltd. All rights reserved.
Wiley International Encyclopedia of Marketing
How to Cite
Russell, G. J. 2010. Models for Categorical Dependent Variables. Wiley International Encyclopedia of Marketing.
- Published Online: 15 DEC 2010
The analysis of categorical dependent variables requires special care in both model specification and parameter interpretation. Researchers must understand that models for categorical data are designed to forecast the discrete probability distribution of categorical outcomes, not the actual outcome itself. By far, the most prominent analysis tools used in marketing research are variants of the multinomial logit model (random utility theory (RUT) logit, common logit). These models can be viewed as special types of generalized linear models (GLM). More complex models can be developed through the use of latent variables. In particular, latent variables permit the construction of models that incorporate ordinal relationships among outcomes (ordinal regression), differing patterns of similarity among alternatives (multinomial probit), and heterogeneity in response parameters (latent class and random coefficient treatments of logit and probit). Recent research has begun the development of multivariate models (multivariate probit, autologistic regression) that allow for correlations among categorical dependent variables.
- generalized linear model;
- multinomial logit;
- multinomial probit;
- ordinal regression;
- multivariate probit;
- autologistic regression