SummaryA new testing procedure is derived which enables to assess the equivalence of two arbitrary noncontinuous distribution functions from which unrelated samples are taken as the data to be analyzed. The equivalence region is defined to consist of all pairs FY G of distribution functions such that for independent X $ F, Y $ G the conditional probability of fX b Yg given fX T Yg lies in some short interval around 1a2. The test rejects the null hypothesis of nonequivalence if and only if the standardized distance between the U-statistics estimator of PX b Y j X T Y and the center of the equivalence interval 1a2 À e 1 Y 1a2 e 2 does not exceed a critical upper bound which has to be computed as the a-quantile of a c 2 -distribution with one degree of freedom and a random noncentrality parameter proportional to the squared length of that interval. The test is shown to maintain the asymptotic significance level under very weak regularity conditions. Results of an extensive simulation study suggest that its level properties are very satisfactory in small samples as well. The power turns out to be inversely related to the rate PX Y of ties between observations from different samples.

A new testing procedure is derived which enables to assess the equivalence of two arbitrary noncontinuous distribution functions from which unrelated samples are taken as the data to be analyzed. The equivalence region is defined to consist of all pairs (F,G) of distribution functions such that for independent X ∼ F, Y ∼ G the conditional probability of {X > Y} given {X ≠ Y} lies in some short interval around 1/2. The test rejects the null hypothesis of nonequivalence if and only if the standardized distance between the U‐statistics estimator of P[X > Y ∣ X ≠ Y] and the center of the equivalence interval (1/2 — ε1, 1/2 + ε2) does not exceed a critical upper bound which has to be computed as the α‐quantile of a χ2‐distribution with one degree of freedom and a random noncentrality parameter proportional to the squared length of that interval. The test is shown to maintain the asymptotic significance level under very weak regularity conditions. Results of an extensive simulation study suggest that its level properties are very satisfactory in small samples as well. The power turns out to be inversely related to the rate P[X = Y] of ties between observations from different samples.