How to Set Delta in the Two-One-Sided T-tests Procedure (TOST). (29th August 2019)

• Record Type: Journal Article
• Authors: Juzek, Tom; Kizach, Johannes
• Is Part Of: Journal of research design and statistics in linguistics and communication science. Volume 5:Number 1/2(2018)
• Page Start: 153
• Page End: 169
• Publication Date: 2019-08-29
• Subjects: twoone- sided t-tests -- test validation -- data simulation -- equivalence testing -- similarity testing -- t-test -- tost -- statistical methods; Linguistics -- Research -- Periodicals; Linguistics -- Statistical methods -- Periodicals; Mathematical linguistics -- Periodicals; 410.72
• DOI: 10.1558/jrds.39002
• Languages: English
• ISSNs: 2052-417X

Abstract

Abstract : The Two-One-Sided T-test procedure (TOST) is used to show that two samples are equivalent or similar, in contrast to classical statistical tests which check for dissimilarity. The TOST relies on a parameter called delta, which has to be set by the researcher using their intuition. Doing so can be difficult, because of complex interactions of relevant parameters. In this article we present a method to set delta, which is established and validated through extensive simulations based on real data sets from linguistics and other sciences. The presented method is shown to be sound and reliable, but we cannot exclude deviant early model behaviour (N≤10) and deviant late model behaviour (N>100, 000).