Predicting paired preferences from sensory data ¤

Paired-preference studies for market research tend to focus on highly replicating the few comparisons of most interest to the client. By establishing a predictive model between sensory panel means scores for products and the relative proportions of consumers preferring them when presented in pairs, it is possible to get reliable predictions of all possible comparisons among a whole group of products. The binomial discrete choice logistic model is appropriate for building this relationship between sensory and consumer preference and can be easily fitted using the commercially available statistical package SAS. Modifications to the model are described to account for consumers who score products either as "equally liked" or "equally disliked". Rather than treat these as ties, the argument is made that no-preference is best modelled as a function of the sensory data itself. An application of the new model is given to data from a home-test of 12 personal care products that were evaluated by more than 500 consumers, each receiving six pairs of products.