Title: Two-sample dispersion tests for interval-valued data
Authors: Przemyslaw Grzegorzewski - Warsaw University of Technology (Poland) [presenting]
Abstract: Two-sample tests for dispersion belong to a basic toolbox of statistical procedures. The goal is to generalize a suitable test for interval-valued data perceived both from the epistemic and ontic perspective. Usually to verify whether two populations differ in scale one compares two population variances or standard deviations. Unfortunately, a generalization of such statistical procedures into the interval-valued framework may cause considerable computational problems, especially if a sample is large enough. Indeed, a sample variance computation for the epistemic intervals is NP-hard task. Moreover, most of the tests for comparing variances assume that the underlying population is normally distributed. Therefore, to avoid problems in verifying assumptions on the underlying distribution we consider nonparametric tests for dispersion which are not based on variances. They are not only distribution-free, but fortunately, unlike typical generalizations of statistical procedures into the interval-valued framework, our generalized tests entail very low computational costs.