Is What You Choose What You Want? Uncertainty in Choice-Based Conjoint Analysis

• Activity Number: 192
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Marketing
• Author(s): Yu-Cheng Ku*+
• Companies: Fannie Mae
• Keywords: Bayesian ; Conjoint ; Hierarchical Bayes ; Probit ; Random Effects ; Discrete Choice

Abstract

With the creative use of hierarchical Bayes (HB) methods, choice-based conjoint (CBC) analysis, a stated preference research methodology, has become the most popular statistical technique used in marketing research. In the HB-CBC framework, logit models are most considered, where the individual goodness-of-fit is often evaluated with the 'root likelihood' (RLH) advocated by some commercial software packages. While RLH is a widely accepted measure, it fails to answer a natural question in CBC survey: How uncertain a respondent's answer is? In other words, how far away a respondent's stated preference is from the "truth"? To study the question, this paper proposes a novel uncertainty measure embedded in random-effects probit models. A Markov chain Monte Carlo (MCMC) algorithm is developed for model estimation. In the spirit of Bayesian residuals, the measure is defined in terms of the difference between the "true" and the representative utilities and hence directly gauges the respondents' uncertainty in taking CBC tasks. The proposed measure is compared with RLH using real CBC survey data.