DATA COMPRESSION AND STATISTICAL INFERENCE
H. STRASSER · 2012
Data driven methods for clustering and vector quantization can be used to construct nonparametric tests for multivariate testing problems. We combine the old concept of permutation tests with recent clustering algorithms to obtain multivariate inferential methods which in the univariate case are very similar to rank tests. At hand of computer experiments we illustrate some typical properties of those methods. For theoretical results we give references to the literature. Keywords: Permutation test; multivariate classification; vector quantization; information in a partition. 1 INTRODUCTION The subject of this paper originated from a joint project which has been carried out together with marketing scientists. As starting points in the applied area we refer to Mazanec, [18], [19] and [20]. Using business terminology the topics are market segmentation and product positioning. From the statistical point of view we are dealing with classical statistical problems like the k-sample p...