Linear rank tests for independence in bivariate distributions—power comparisons by simulation

Abstract

The powers of some usual linear rank tests of independence and of some adaptive tests are compared by simulation. The study shows that the Spearman test has, in comparison to the other linear rank tests considered, a relatively good power. However, for some distributions the exponential scores test or the van der Waerden test are better. Over all distributions considered the van der Waerden test exceeds the other tests, very densely followed by the Spearman test. Among the adaptive tests the test based on isotonized Alternative Conditional Expectation scores has the best power properties.