Statistics Papers

Document Type

Journal Article

Date of this Version

2011

Publication Source

The American Statistician

Volume

65

Issue

1

Start Page

11

Last Page

15

DOI

10.1198/tast.2011.10130

Abstract

In a randomized 2×2 factorial trial, more than one hypothesis is to be tested, so some method must be used to control the probability of falsely rejecting at least one true hypothesis. We contrast familiar elementary methods of controlling the family-wise error rate based on the Bonferroni–Holm procedure with a less familiar but equally elementary form of structured testing associated with the large class of procedures that descend from the closed testing approach of Marcus, Peritz, and Gabriel. In a range of plausible situations, giving priority to main effects in structured testing typically yielded greater power to detect main effects for a given sample size or reduced sample size for a given power; it also permitted testing for interaction when main effects are found.

Copyright/Permission Statement

This is an Accepted Manuscript of an article published by Taylor & Francis in The American Statistician on 01 Jan 2012, available online: http://wwww.tandfonline.com/10.1198/tast.2011.10130.

Keywords

factorial experiment, familywise error rate, multiple testing, ordered family of hypotheses

Share

COinS
 

Date Posted: 27 November 2017