Asymptotic optimality of data-driven Neyman's tests for uniformity (Q1354511)

Data-driven Neyman's tests resulting from a combination of Neyman's smooth tests for uniformity and Schwarz's selection procedure are investigated. Asymptotic intermediate efficiency of those tests with respect to the Neyman-Pearson test is shown to be 1 for a large set of converging alternatives. The result shows that data-driven Neyman's tests, contrary to classical goodness-of-fit tests, are indeed omnibus tests adapting well to the data at hand. Further, the paper provides some theoretical results supporting the nice simulation results for the data-driven test proposed by \textit{T. Ledwina} [J. Am. Stat. Assoc. 89, No. 427, 1000-1005 (1994; Zbl 0805.62022)]. Namely, for the class of data-driven smooth tests described by \textit{W. C. M. Kallenberg} and \textit{T. Ledwina} [Ann. Stat. 23, No. 5, 1594-1608 (1995; Zbl 0847.62035)], asymptotic power is investigated via the intermediate efficiency approach. More precisely, for a given member of the class of tests, a large set of alternatives, say \(P_n\)'s, converging to the null distribution (as the sample size \(n\) increases) is found for which the data-driven Neyman's test is asymptotically as efficient as the Neyman-Pearson test for testing uniformity against \(P_n\). It is shown that in this set the actual direction of approach to the null hypothesis is immaterial but the rate of convergence plays a role. So, the data-driven Neyman's tests considered in this paper have qualitatively different asymptotic properties from those of the Kolmogorov-Smirnov and Cramér-von Mises tests.