B0557
Title: Anytime-valid permutation tests and general tests of symmetry
Authors: Muriel Perez - Eindhoven University of Technology (Netherlands) [presenting]
Tyron Lardy - Leiden University (Netherlands)
Abstract: A general framework is presented to perform anytime-valid tests of distributional symmetry under general families of transformations. The framework includes anytime-valid versions of permutation tests, rank tests and tests of rotational symmetry, among others. Just as their classic fixed-sample counterparts, the resulting anytime-valid tests are distribution-free; they retain their type-I error guarantees under large, nonparametric families of null distributions. As an important application, two-sample testing is addressed. There, univariate observations from two populations are monitored continuously and equality in distribution is tested for. To this end, tests of exchangeability between the samples, rank tests and other existing anytime-valid tests are compared. It discusses how to design experiments with continuous monitoring, the computational challenges that are faced, and the trade-offs between the methods presented.
