Optimality robustness of tests in two population problems (Q1819488)

The problems of testing the equality of two location parameters (without the presence of scale parameters) and the equality of two scale parameters (with or without the presence of location parameters) based on two independent samples are considered. It is shown that, under some mild conditions on the underlying distributions, the standard optimum invariant tests derived under normality or exponentiality continue to be optimum for a wide class of distributions. However, the tests are not null robust. As a technical tool, \textit{R. A. Wijsman}'s representation theorem [Proc. 5th Berkeley Sympos. math. Statist. Probab., Univ. Calif. 1965/1966, 1, 389-400 (1967; Zbl 0189.186)] is used.