Modeling the latent preference structure implied by discrete choice experiments

Applications of discrete choice experiments to study consumer choice behavior have grown significantly over the past decade. Typically, discrete choice experiment data are analyzed using McFadden’s Multinomial Logit (MNL) model or a more sophisticated extension that relaxes some of MNL’s restrictive assumptions. Advantages of MNL for analyzing discrete responses are well-known to marketing academics and practitioners, but arguably the latent preference structure that explains discrete responses has yet to be clearly articulated. For example, integration of latent variable systems and discrete choice experiments has several advantages. First, it is possible to model the variances associated with consumers’ discrete choices by introducing latent preference variables, which can provide useful information on the consistency of consumers’ choices. Second, one can specify fully generalized models whereby brand characteristics explain variation in consumers’ latent brand preferences to provide insight into processes that drive consumers’ preferences. Third, the resulting latent variable systems generalize to all of the characteristics of brands one chooses to study. We illustrate the integration in an application to consumer preferences for brands in the “last minute” hotel accommodation market. For convenience, we use Joreskog’s LISREL model to fit the latent structure, although other latent variable systems are considered. The study has immediate implications for researchers analyzing stated preference data, with additional implications for potential applications to revealed preference data