A0488
Title: A statistical test for the overlap of normal distributions based on generalized p-values
Authors: Gerhard Goessler - University of Graz (Austria)
Vera Hofer - University of Graz (Austria) [presenting]
Hans Manner - University of Graz (Austria)
Walter Goessler - University of Graz (Austria)
Abstract: In the context of the development of generic drugs, it is often necessary to test whether the distribution of a quality attribute (QA) of a test product (generic drug) is suitably covered by a reference product. Based on k-cover, a statistical test for simultaneously comparing the k- and the (1-k)-quantiles of the test and the reference distribution is proposed, assuming both distributions are normal. It constitutes a multiple testing problem that demands, depending on the distributions at hand, a proper correction (similar to Bonferroni correction) to guarantee that the significance level chosen is maintained. The proposed test is an extension of an existing test for comparing a quantile of two distributions, which utilizes the theory of generalized inference. It results in a two-step test procedure, which implicitly performs a pre-test that decides the correction to apply to avoid error inflation due to multiple testing. Simulations show that the test proposed keeps the type-I error under control and has the desired asymptotic properties.
