scientific article; zbMATH DE number 3974054

From MaRDI portal
Publication:3740035





[For the entire collection see Zbl 0579.00015.] The theory of two-sided tests is much less developed than that of their one-sided counterparts. This paper comprises several aspects of two-sided testing problems that are largely independent of each other but are linked by the idea to make them amenable to optimization techniques by linearizations around the null-hypothesis. More precisely we strive to prove the existence of locally most powerful (LMP) unbiased two-sided tests and to exhibit their form. Section 1 derives an extension of the Neyman-Pearson fundamental lemma for certain nonlinear objective functions under linear side conditions. In particular necessary and sufficient conditions for local optimality are given. Here we heavily rely on the methods of convex programming. In section 2 the simplest two-sided sequential testing problem is discussed. Section 3 shows that for the most important two-sided nonparametric problems the LMP unbiased rank test is asymptotically equivalent to the intuitively obvious rank test, namely to that one based on the squared one-sided rank statistic. To this end several results on one-sided linear rank tests [cf. \textit{J. Hájek}, Ann. Math. Stat. 33, 1124-1147 (1962; Zbl 0133.420) and \textit{J. Hájek} and \textit{Z. Šidák}, Theory of rank tests. (1967; Zbl 0161.381)] are extended to the two-sided case. In section 0 basic facts of the theory of LMP two-sided tests, are reviewed. This justifies the introduction of \({\mathbb{L}}_ 1\)- differentiable classes of distributions, which seem for these kinds of problems more adequate than the \({\mathbb{L}}_ 2\)-differentiable classes.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3740035)