Testing against uniform stochastic ordering with paired observations (Q1983624)

A random variable \(X\) is less than \(Y\) wrt stochastic ordering (SO) if for all real \(t\) holds \(P(X>t)\le P(Y>t)\). \(X\) is less than \(Y\) wrt uniform stochastic orderung (USO) if for all real \(t\) and \(s\) the conditional distribution of \(X\) given that \(X>s\) is smaller than the corresponding conditional distribution of \(Y\) wrt SO. In the absolute continuous case, USO is also known as hazard rate ordering which is important e.g. in reliability and actuarial sciences. In this paper the authors present a two-sample nonparametric goodness-of-fit test of \(H_0\): \(X\) is smaller than \(Y\) wrt USO against \(H_1\): not \(H_0\). They extend their own results [\textit{C.-F. Tang} et al., Ann. Stat. 45, No. 6, 2565--2589 (2017; Zbl 1388.62134)] to the case that \(X\) and \(Y\) are not necessarily independent. Compared with their first paper, they use a different way to find a data-driven critical value. They show that their critical value controls the type-I-error asymptotically and yields a consistent test regardless of the correlation between \(X\) and \(Y\). When \(X\) and \(Y\) are independent, simulations show that the new test is more powerful than the one in the former article.