Eigenschaften und Erweiterungen der Methode CLV zum Clustern von Variablen : Anwendungen in der Sensometrie

Abstract

In this work, the properties of the method of clustering of variables around latent components (CLV) are investigated. A statistical model is postulated. This model is especially appropriate for sensory profiling data. It sheds more light on the method CLV. The clustering criterion can be expressed in terms of the parameters of the model. It is shown that, under weak conditions, the hierarchical algorithm of CLV finds the correct partition while the partitioning algorithm depends on the partition used as a starting point. Furthermore, the performance of CLV on the basis of a sample is investigated by means of a simulation study. It is shown that this performance is comparable to the performance of known methods such as the procedure Varclus of the software SAS. Finally, two methods for determining the number of groups are proposed and compared.