Probabilistic feature analysis of product perception based on pick any/n data

Studies of product positioning often use perceptual mapping techniques to study the cognitive dimensions that lie at the basis of product perception. As the metric assumptions at the basis of dimensional geometric presentations may be doubtful, perceptual analysis models based on qualitative features rather than quantitive dimensions are a useful alternative to consider. In the present paper we describe a latent class extension of a non-compensatory probabilistic feature-based model which can be used to model binary product-attribute associations on the basis of binary latent features, as well as to model differences in product perception among consumer segments. A fully Bayesian approach involving the computation of a sample of the posterior distribution is used for parameter estimation. Moreover, specific model checks are proposed to evaluate the fit of the model with posterior predictive checks. As an illustration, the models are applied to explain associations between sandwich fillings and filling characteristics on the basis of latent features. The results of the analysis indicate that models with consumer differences clearly fitted the data better than models without consumer differences. Moreover, the extracted features are meaningful and the identified consumer differences yield interesting insights with respect to differences in product perception. In particular, segments turn out to differ in the way they discriminate between products on the basis of a certain latent feature, such as whether fillings are 'healthy' and 'pure', or whether they are 'caloric', 'salty' and 'fat'.