Partitioning methods outperform hierarchical methods for clustering consumers in preference mapping

In preference mapping, mean liking scores of the products often appear to be the resultant of diverging opinions that need to be summarized with a small number of homogeneous consumer clusters. For doing so, the common practice of most consumer research agencies is to use hierarchical clustering methods. The objective of this work is to compare this common practice with partitioning methods (e.g. k-means clustering) that we generally apply. These comparisons are made on actual Nestlé studies and on simulated datasets. For actual studies, methods are compared according to four indexes measuring the compactness and the separation of the clustering. For simulated data, the methods are compared on their ability to recover the a priori defined clusters. Results show that common partitioning methods like k-means outperform hierarchical methods for clustering consumers in preference mapping.