The role of permutation tests in exploratory multivariate data analysis

In sensory and consumer science the type of research and the type of data are such that using formal statistical testing is not always the best approach. A disadvantage of such 'strict' statistical methods is that assumptions should be fulfilled to be able to perform statistical testing. In practice the assumptions are seldom checked. The exploratory use of methods of multivariate data analysis may be a good alternative. The sometimes heard criticism that these methods suffer from a lack of experimental design, could be countered by realising that there are certain, implicit, design-choices associated with the use of these methods. Furthermore the paper presents the use of permutation tests as a means to test the results from an exploratory MVA method (PCA). It is concluded that permutation tests can almost always be performed for determining significant deviations from an alternative random explanation of the effects in the data.