A0354
Title: Multivariate finite-sample adjustments for equivalence testing
Authors: Luca Insolia - University of Geneva (Switzerland) [presenting]
Stephane Guerrier - University of Geneva (Switzerland)
Maria-Pia Victoria-Feser - University of Geneva (Switzerland)
Yanyuan Ma - Pennsylvania State University (United States)
Younes Boulaguiem - University of Geneva (Switzerland)
Dominique-Laurent Couturier - University of Cambridge (United Kingdom)
Abstract: Average equivalence testing procedures aim at assessing whether two or more effects are comparable. They rely on the definition of a tolerance region inside which the compared effects (e.g., differences in means) could be considered negligible. This is in striking contrast to the traditional hypothesis testing framework, where the null and alternative hypotheses being tested are switched, as it reverses the burden of proof to demonstrate that the compared effects are indeed similar. The Two One-Sided Tests (TOST) procedure is widely used across different domains, but it is known for being too conservative. This leads to lower power to detect equivalence, especially in the presence of effects with higher variability. We propose a finite-sample adjustment of the TOST to guarantee that the resulting test is exactly of size alpha and, at the same time, is uniformly more powerful than existing methods based on the TOST. The proposed approach is defined at the population level, but it maintains good property when it is estimated from the data. Computationally lean algorithms, approximations, and extensions to multivariate equivalence testing problems are also discussed. Our results are supported by extensive Monte Carlo simulations and a real-world application related to pharmacokinetics bioequivalence.
