[2311.10900] A robust rank-based correction to a number of testing underneath constructive dependency

0
35


Obtain a PDF of the paper titled A robust rank-based correction to a number of testing underneath constructive dependency, by Alexander Timans and three different authors

Obtain PDF

Summary:We develop a novel a number of speculation testing correction with family-wise error fee (FWER) management that effectively exploits constructive dependencies between doubtlessly correlated statistical speculation assessments. Our proposed algorithm $texttt{max-rank}$ is conceptually straight-forward, counting on the usage of a $max$-operator within the rank area of computed take a look at statistics. We examine our strategy to the regularly employed Bonferroni correction, theoretically and empirically demonstrating its superiority over Bonferroni within the case of current constructive dependency, and its equivalence in any other case. Our benefit over Bonferroni will increase because the variety of assessments rises, and we keep excessive statistical energy while guaranteeing FWER management. We particularly body our algorithm within the context of parallel permutation testing, a situation that arises in our major utility of conformal prediction, a not too long ago popularized strategy for quantifying uncertainty in complicated predictive settings.

Submission historical past

From: Alexander Timans [view email]
[v1]
Fri, 17 Nov 2023 22:44:22 UTC (3,263 KB)
[v2]
Thu, 25 Jan 2024 15:43:15 UTC (3,263 KB)



Supply hyperlink

LEAVE A REPLY

Please enter your comment!
Please enter your name here