Bayesian Statistics and Marketing, 2nd Edition

10Case Study 1: Analysis of Choice‐Based Conjoint Data Using A Hierarchical Logit Model

Abstract

R

In this case study, we consider a workhouse model in marketing and economics, the random coefficient logit model. One popular application of this model is for the analysis of conjoint survey data.¹ Conjoint survey data has a panel format in which respondents choose among a small number of alternative products specified by specific product features (attributes) and specific values of these features (levels). Each respondent in the conjoint survey makes a relatively small number of choices (usually between 8 and 16) and these responses are analyzed using a random coefficient logit model. In bayesm, we implement Bayesian inference for a random coefficient logit model with both a mixture of normal components as well as the ability to impose sign constraints on the random coefficients in the routine, rhierMnlMixture.

10.1 CHOICE‐BASED CONJOINT

The most popular form of conjoint survey analysis is called choiced‐based conjoint (CBC). At least 10,000 CBC surveys are conducted per year in a wide variety of commercial and academic contexts. A frequent commercial application is to forecast demand (either in sales or market share) for new products or new product features. Here the hypothetical nature of the survey context is useful to make forecasts of the demand for products, features, or configurations which are not currently in the marketplace. Conjoint surveys can provide one input ...