Least equivalent allowable differences in equivalence testing

This paper aims at an extension and refinement of some recent work on equivalence testing in sensory science. The performance of the two-one-sided-tests (TOST) approach using the t-test statistic for normally distributed data is briefly compared from a practical point of view to alternatives found in the literature. Based on the TOST and on several other methods like, e.g., the Binomial test for 2- and 3-AFC methods, it is shown how a least equivalent allowable difference (LEAD) can be determined. This is related to an appropriate confidence interval, while typically only the absolutely larger bound is taken into account. The LEAD gives valuable information to the investigator and is particularly independent from the analyst’s choice of a maximum allowable difference. Re-visiting some examples from the literature illustrates this concept.