Date of Award

Summer 2008

Degree Type

Thesis - Restricted

Degree Name

Master of Science (MS)

Department

Mathematics, Statistics and Computer Science

First Advisor

Bansal, Naveen

Second Advisor

Hamedani, Hossein, G.

Third Advisor

Ghose, Sanjoy

Abstract

Multinomial Probit or Multinomial Logit models are frequently used in marketing research for modeling brand choices of consumers. The main idea is that consumers choose brands by maximizing brands' utilities. This leads to multinomial probit or multinomial logit model depending upon the assumption on the distribution of errors. Recently, it has been pointed out by many authors that different consumers behave differently, and the brand utilities for different consumers are different. We show through an empirical study that this heterogeneity leads to a significant improvement in the model in a Bayesian framework. We also develop a goodness of fitness measure based on the predicted probabilities.

Share

COinS