pipette
ENEnglish

Uniformly Improved Adaptive Filtering Procedures for Replicability Testing with FWER Control

Ninh Tran

Preprint

En palabras de los autores

Identifying signals that replicate across multiple studies is a fundamental challenge in modern data analysis. Partial conjunction (PC) testing addresses this problem by assessing whether a signal is present in at least a specified number of studies. The AdaFilter framework of Wang et al. (2022, Annals of Statistics 50, 1890-1909) improves the power of PC testing through adaptive filtering, but its family-wise error rate (FWER)-controlling procedures rely on conservative Bonferroni corrections at two stages: (i) the construction of PC -values and filtering statistics, and (ii) the multiplicity adjustment used to control the FWER. We show that both stages can be improved using the \v{S}id\'ak correction. Under the same independence assumptions as the original AdaFilter framework, our improvements preserve finite-sample FWER control and guarantee at least as many rejections as their Bonferroni counterparts for every realisation of the data. Simulations and a real data analysis demonstrate the practical benefits of these improvements.

Resultado principalEl resumen no menciona limitaciones.

Apareció: jueves, 22 de octubre. arXiv. Preprint, todavía sin revisión por pares.

Comentario de los autores: 56 pages, 7 figures