Würtz, A. H., A Universal Upper Bound on Power Functions

Abstract

The finite sample power function is shown to be bounded by a power upper bound. The same power upper bound is valid for any test of a hypothesis. In normal regression models the power upper bound is simple to calculate including cases where the hypotheses are composite and where the regression function is nonlinear.

Keywords: Composite hypothesis, finite sample power, Lagrange multiplier test, likelihood ratio test, nonlinear regression models, nonmonotonic power, uniformly most powerful test.