Confirmation of odd sample bias in triadic design preference tests with a no-preference option, using confusable stimuli, and a solution

Abstract

Two preference test methods utilizing triadic designs (i.e., presenting two identical products and a different product in a test) have been proposed to improve the validity of the paired preference test using a no-preference response option, by considering responses generated from preference bias (i.e., consumers’ responses are tilted toward a preference response rather than a no-preference response, resulting in the overestimation of preference responses). One of the test methods was the replicated duo-trio difference-preference (DT-Preference) test method, which had a balanced-reference design, while the other was the triadic preference test method. These two methods were designed to provide more conservative preference results in different ways. For the DT-Preference test method, it was assumed that the preference responses that were simply generated from the preference bias would also be prone to an odd sample bias (i.e., consumers tilted the preference against the stimulus that was most different from the other two) in the triadic design. For the triadic preference test method, the responses were screened out if the subjects reported a preference between the identical stimuli, assuming no bias in the triadic design. In the present study, the odd sample bias assumed in the triadic design was tested by the triadic preference test method and the DT-Preference test method by using two pairs of confusable orange juice stimuli assumed to have negligible preference by 256 consumers. For the DT-Preference test method, two variants of the duo-trio test designs were used. The odd sample bias was confirmed for all test methods using triadic designs, invalidating the application of the triadic preference test method without replication. © 2019 Elsevier Ltd