Abstract
We analyze the behaviour of the selector measures used in the two-sample adaptive, distribution-free test of Hogg et al. (1975). Based on this analysis we propose new selection measures which do not depend on the location alternatives and which therefore improve the power of the adaptive test.