Latent variable modeling of consumer brand preferences using best-worst experiments

Latent variable models are well suited to the analysis of choice experiment data because of their ability to provide estimates of effects as well as variances. Typically, best-worst choice experiment data are analyzed with traditional choice models. For simple brand name only experiments, this analysis yields estimates of effects or measures of the strength of preference for the brands studied. Yet, the application of latent variable models to best-worst experiments provides estimates of effects as well as variances (and covariances), making it possible to estimate (measure) the strength of consumers’ preferences for the brands studied and estimate (measure) the consistency of their preferences. First, using data from best-worst experiments, latent mean analysis provides measures of effects or preferences for each brand studied. These models are based on repeated choices and the have the advantage of controlling for measurement error in the estimation of effects. Furthermore, the analysis shows relationships among scales of best, estimates of effects from traditional choice models and estimates of latent means. Second, application of confirmatory factor analysis models yields measures of variance or consistency of preference for each brand studied. Finally, one can specify and estimate covariances among brand preferences, yielding further insights into consumers’ choices. Methods and models are illustrated with data from a consumer setting and extensions to more complex best-worst experiments are discussed.