Multinomial Logistic Regression

This chapter addresses multinomial logistic regression, used when a nominal response measure contains more than two categories. It reviews multinomial regression applications in published communication research and discusses the fundamental components of multinomial logistic regression. In the multinomial model, maximum likelihood establishes parameter estimates, and a generalized logit serves as the link function. In addition to likelihood values, multinomial logistic regression reports three types of pseudo R-square measures, McFadden as well as the Hosmer and Lemeshow goodness-of-fit test. Closely related to multinomial logistic regression is the conditional logit, or discrete-choice, model. Developed by McFadden, conditional logit analysis considers as explanatory measures the characteristics of choice options as opposed to, or in addition to the characteristics of individuals making a choice. The chapter uses data gathered in the 2012 Monitoring the Future study to demonstrate multinomial logistic regression analysis in SPSS.