To preliminary test or not to preliminary test, that is the question

Derrick, Ben

Abstract

Despite the consequences of violations to the assumptions of statistical tests, there is often neglect in published works to report on the assumptions of the tests performed. Preliminary testing the assumptions prior to preforming the test is done routinely by some researchers, and some do in an ad-hoc manner. Literature and opinions on these contrasting approaches are discussed.

Complications with preliminary testing arise due to each assumption having many possible preliminary tests that could be performed. In addition the arbitrary 5% significance level is often used but this need not be standard. In the comparison of two independent samples, the significance level for the preliminary tests is considered, along with two preliminary tests for equality of variances, and two preliminary tests for normality. The results indicate that preliminary testing may not be unreasonable, if the analysis strategy is determined at the study design stage.

• Presentation Conference Type: Conference Paper (unpublished)
• Start Date: Jul 24, 2018
• End Date: Jul 27, 2018
• Acceptance Date: May 26, 2018
• Publicly Available Date: Jun 7, 2019
• Peer Reviewed: Not Peer Reviewed
• Keywords: preliminary test, conditional test, independent samples t-test, robustness
• Public URL: https://uwe-repository.worktribe.com/output/867726
• Additional Information: Title of Conference or Conference Proceedings : Research Students Conference in Probability and Statistics, 2018