Segmentation of subjects in multivariate paired comparisons. Application to the preference for concert halls acoustics

Multivariate paired comparisons allow one to compare stimuli by pairs on more than one situation (different contexts, several types of preferences or various sensory attributes). This paper focuses on the segmentation of the subjects after such a task. An adaptation of the latent class segmentation, based on the Bradley-Terry-Luce model, to the multivariate nature of the data is developed, allowing the study of the dependencies between attributes. This technique is applied to a comparison between concert halls’ acoustics. The influence of context (type of music and position of the seat in the hall) on the preference for the concert halls is presented.