Two Methods for Clustering Products in a Sensory Study: STATIS and ClusMB

Sensory experiments are essential for product characterization and development. One of the main objectives of these tasks is to highlight similarities and differences among the target products. For this purpose, cluster analysis of the products can be helpful in investigating the relationships among the products. To perform such an analysis, we propose two methods: a new original method named ClusMB, and one based on clustering on the STATIS axes. The methods are demonstrated in two case studies involving Check-All-That-Apply (CATA) and projective mapping, but can easily be extended to other methods producing multi-block data structures. A paradox is highlighted to show the importance of these new methods in overcoming the theoretical limitations of average or contingency tables. To aid the interpretation of the outcomes, insightful indices are provided to quantify subjects' agreement and contribution to the chosen clustering solution.