Tests based on equivariant estimators

From MaRDI portal





For absolutely continuous distribution functions F and G, whose densities have interval supports contained in (0,\(\infty)\), F is said to be star- shaped with respect to G if \(G^{-1}(F(x))/x\) increases in (0,\(\infty)\). If the densities of F and G have interval supports in \((-\infty,+\infty)\) then F is more dispersed than G if \(F^{-1}(x)-G^{-1}(x)\) is increasing in (0,1). The author proves the following: Let \(X_ 1,X_ 2,...,X_ n\) be an iid sample from F(x-\(\mu)\) where \(\mu\) is an unknown location parameter and \(T^{(1)}\) and \(T^{(2)}\) be two location equivariant estimators of \(\mu\). The test for \(H_ 0: \mu \leq \mu_ 0\) versus \(H_ A: \mu >\mu_ 0\) that rejects \(H_ 0\) when \(T^{(1)}\) is large, is uniformly more powerful than the test that rejects \(H_ 0\) when \(T^{(2)}\) is large if and only if the distribution of \(T^{(2)}\) is more dispersed than that of \(T^{(1)}\). A similar result is proved for tests on a scale parameter using the star- shaped ordering.











This page was built for publication: Tests based on equivariant estimators

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3723476)