A0344
Title: Powerful multiple test with a fixed sample size
Authors: Masaki Toyoda - Hitotsubashi University (Japan) [presenting]
Yoshimasa Uematsu - Hitotsubashi University (Japan)
Abstract: Modern data analyses often encounter challenges due to very small sample sizes despite high dimensionality. The focus is on statistical inference in such a situation, where the number of hypotheses is very large relative to a limited sample size, and a novel method of FDR-controlled multiple test is proposed. The key idea is to use an accumulation test and data fission, which have recently been developed in the literature. Remarkably, it is shown that the power can tend to unity as the number of hypotheses increases, even though the sample size is fixed. The validity of the method is also confirmed by numerical experiments and a real data analysis.
