Contents Online
Statistics and Its Interface
Volume 15 (2022)
Number 3
Paired-sample tests for homogeneity with/without confounding variables
Pages: 335 – 348
DOI: https://dx.doi.org/10.4310/21-SII695
Authors
Abstract
In this article, we are concerned about testing the homogeneity on paired samples with or without confounding variables. These problems usually arise in clinical trials, psychological or sociological studies. We introduce new nonparametric tests for equality of two distributions or two conditional distributions of random vectors on paired samples. We show that their test statistics are consistent but have different asymptotic distributions under the null hypothesis, depending on whether confounding variables exist. The limit distribution of the test statistic is a mixed $\chi^2$ distribution when testing the equality of two paired distributions, while it is a normal distribution when testing the equality of two conditional distributions of paired samples. We conduct several simulation studies to evaluate the finite-sample performance of our tests. Finally, we apply our tests on real data to illustrate their usefulness in the applications.
Keywords
homogeneity, conditional distribution, paired samples, energy distance
Received 13 October 2020
Accepted 4 August 2021
Published 14 February 2022