The randles-hogg test and an alternative proposal

Abstract

In the two-sample case we compare the adaptive distribution-free test of location proposed by Randias and Hogg (1973) witha (nonlinear) rank test proposed in the second port of this Paper While the performance of both tests is Comparable under exact shift alternatives,it is shown in the more realistic situation of stochastically larger observation of the first sample that the Randles-Hogg, test say perfonm poorly whereas the latter test behaves quite well. The test statistic of the (nonadaptive) rank test proposed here is the supremem of all standardised and centered simple linear rank statistics (cf. Hajek ana Sidak 1967,p.41) having nondecreasing scores.Finaiiy, critical values of the test are tabulated for(pooled) sample sizes from 12 to 20, and some results on the asymptotic behavior are presented.