Asymptotic powers of one- and two-sample rank tests against location- alternatives including contamination (Q1083149)

For the nonparametric hypotheses of symmetry about zero and equality of distribution functions in one- and two-sample problems, we often consider only location alternatives. But after treatments are received, we cannot predict precisely enough if observations give rise to only variation of location of a distribution. So we consider the alternative distribution of the form \((1-\epsilon)F(x-\theta)+\epsilon H(x-\theta)\) for the null distribution of the form F(x) and discuss asymptotic powers of one- and two-sample rank tests under contiguous sequences of the above alternatives. Then in the one-sample case, we show that asymptotic relative efficiencies (ARE's) of signed rank tests with respect to the t-test are equivalent to the classical ARE-results against shift alternatives for the distribution H(x) that is symmetric about zero, and the ARE of the signed rank test based on normal scores with respect to the t-test is one for \(F(x)=normal\) irrespective of H(x) and the Wilcoxon signed rank test is asymptotically the most powerful for \(F(x)=\log istic\) and \(H(x)=\{F(x)\}^ 2.\) Also we give results for the two-sample case similar to some results obtained in one sample case and discuss asymptotic powers of k-sample rank tests additionally.