The Building Blocks of Classical Nonparametric Two-Sample Testing Procedures: Statistically Equivalent Blocks (2501.10844v3)

Abstract: Statistically equivalent blocks are not frequently considered in the context of nonparametric two-sample hypothesis testing. Despite the limited exposure, this paper shows that a number of classical nonparametric hypothesis tests can be derived on the basis of statistically equivalent blocks and their frequencies. Far from being a moot historical point, this allows for a more unified approach in considering the many two-sample nonparametric tests based on ranks, signs, placements, order statistics, and runs. Perhaps more importantly, this approach also allows for the easy extension of many univariate nonparametric tests into arbitrarily high dimensions that retain all null properties regardless of dimensionality and are invariant to the scaling of the observations. These generalizations do not require depth functions or the explicit use of spatial signs or ranks and may be of use in various areas such as life-testing and quality control. In the manuscript, an overview of statistically equivalent blocks and tests based on these blocks are provided. This is followed by reformulations of some popular univariate tests and generalizations to higher dimensions. A brief simulation study and comments comparing the proposed methods to existing testing procedures are offered along with some conclusions.