Frequentist Hypothesis Testing with Background Uncertainty

Abstract

We consider the standard Neyman-Pearson hypothesis test of a signal-plus-background hypothesis and background-only hypothesis in the presence of uncertainty on the background-only prediction. Surprisingly, this problem has not been addressed in the recent conferences on statistical techniques in high-energy physics -- although the its confidence-interval equivalent has been. We discuss the issues of power, similar tests, coverage, and ordering rules. The method presented is compared to the Cousins-Highland technique, the ratio of Poisson means, and ``profile'' method.