Trinomial Modeling in One Binary Logit Regression

• Activity Number: 537
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Marketing
• Author(s): Stan Lipovetsky*
• Companies: GfK North America
• Keywords: trinomial model ; MaxDiff ; binary logit regression.

Abstract

A new approach of constructing a triple-category outcome model in one binary logistic regression is presented. Various applied problems are formulated with a dependent variable of three categorical levels, for instance, positive-neutral-negative segments of meaning. It is commonly considered in a multinomial model for a categorical variable of three possible outcomes. This work shows that the problem can be reduced to a much more simple and convenient binomial logit model. It can be done in the approach developed in the area of marketing research and known in terms of Best-Worst scaling or MaxDiff modeling. In this approach the positive-neutral data subset is stacked with the negative-neutral subset. In the latter one the predictor signs are changed to opposite. The binary dependent variable is kept equal one for both positive-negative outcomes and equals zero for neutral outcomes, respectively. In the constructed logit regression the positive category predictions are close to 1, negative close to 0, and neutral are in the middle of its continuous 0-1 scale. Theoretical features and practical applications of the model are discussed.