Technical Report, SFB 475: Komplexitätsreduktion in Multivariaten Datenstrukturen, Universität Dortmund 2002,58
The analysis of sensory profiling demands skillful statistical methods to account for different variations that are unknown in other statistical appliances. Besides others, these are the different use of the descriptors by the assessors and the different use of the scales. The two most important approaches to cope with such data are given by Generalized Procrustes Analysis (GPA) and STATIS. Recently, for the latter one several variants have been proposed in order to either simplify the calculation or to improve the results. The aim of this paper is to compare these methods with respect to their performance. For this purpose, a model will be stated to describe the outcomes of a sensory profiling study. On the basis of this model, we give a short insight into the ideas of the methods under consideration, and simulations to compare these methods are realised. From those, systematical differences between the methods occur. Finally, a comparison between the methods with respect to the interpretation of the estimated consensuses is given by means of graphically displaying the outcomes. It will be found that the choice of the method is accidental and can be made according to the simplicity of use for each operator.