Conjoint Analysis: a Tool for Preference Analysis Submitted by Acknowledgments

Conjoint analysis is a family of techniques that originated in psychology and later become popular in market research. The main objective in conjoint analysis is to measure individual’s or population’s preferences on a set of options that can be described by parameters and their levels. Here we consider choice based conjoint analysis, where the preferences of the people are obtained from choice tasks in which from a small subset of options the most preferred one has to be chosen. There are many ways to analyze the data from the population’s preferences. In this thesis we work with two main classes of analysis techniques: based on statistical assumptions and based on direct regression. We contribute by generalizing the least squares method for computing scale values for single attribute products and by applying it to the Bradley-Terry model. We also show that the Thurstone’s and the Bradley-Terry models can be considered as special cases of the family of discrete choice methods. We then develop a compositional approach for estimating scale values for multiattribute products and a new method based on regression analysis that tries to fit a largest inscribed ball. Finally we compare well known existing methods and our newly developed methods for solving the problem of assigning scale values to products. We conclude that both classes of analysis techniques behave comparably on both real world and synthetic data. We also conclude that no method outperforms the others in all situations.