Best practices in sensory equivalence testing
Sensory professionals seeking guidance in best practices often turn to publications from standards organizations such as ISO and ASTM. A review of guides related to sensory equivalence testing will be presented. In several cases the power approach is prescribed for determining equivalency, but this approach is problematic. It attempts to control beta risk in the difference test and declare samples equivalent when the null hypothesis is retained. Its ineffectiveness in practice can be demonstrated through simulations following the approach used by Bi (2005). Code was implemented independently in R and simulations obtained. For example, triangle and duo-trio test results were simulated where the true proportion of discriminators was set to 0.1. The power approach confirmed similarity in the triangle test with probability 0.6005 when the sample size was 54 and with probability 0.0278 when the sample size was 540. The power approach confirmed similarity in the duo-trio test with probability 0.6273 when the sample size was 96 and with probability 0.0323 when the sample size was 960. As precision of measurement increased (with increasing sample size) the probability of concluding that samples were equivalent was diminished, which further underscores the failure of the power approach. Until standards and guidelines can be updated, practitioners must look elsewhere for direction when conducting equivalence tests. Results are contrasted with other statistical approaches. For special cases investigated the sensR package (Christensen & Brockhoff, 2010) implements methods related to equivalence testing where power calculations correspond more closely to simulated results.
Castura, J. C. (2010). Best practices in sensory equivalence Testing. 9th Sensometric Meeting. 25-28 July. Rotterdam, The Netherlands.