📝 SummarXiv

Abstract

MaxT is a highly popular resampling-based multiple testing procedure, which controls the familywise error rate (FWER) and is powerful under dependence. This paper generalizes maxT to what we term ``multi-resolution'' False Discovery eXceedance (FDX) control. FDX control means ensuring that the FDP -- the proportion of false discoveries among all rejections -- is at most $\gamma$ with probability at least $1-\alpha$. Here $\gamma$ and $\alpha$ are prespecified, small values between 0 and 1. If we take $\gamma=0$, FDX control is the same as FWER control. When $\gamma=0$, the new procedure reduces to maxT. For $\gamma>0$, our method has much higher power. Our method is then strongly connected to maxT as well. For example, if our method rejects fewer than $\gamma^{-1}$ hypotheses, then maxT rejects the same. Further, the implied global tests of the two methods are the same, although the implied closed testing procedures differ. Finally, our method provides an easy-to-use simultaneous, multi-resolution guarantee: the procedure outputs a single rejection threshold $q$, but ensures that with probability $1-\alpha$, simultaneously over all stricter thresholds, the corresponding FDPs are also below $\gamma$.