Abstract:

 

Average liking ratings of test products or paired preference proportions are often used to guide product development in consumer product companies. When liking ratings are used, the performance of the products are often tested using an analysis of variance and mean comparisons to select one or more products for further consideration. While a statistically significant difference may provide insight on product superiority, quantification of the effect size itself is also often of interest. Paired preference results are particularly intuitive to quantify such an effect size. Some companies use a preference action standard that corresponds to a meaningful measure of superiority, for instance a 55/45 or 60/40 preference split. Preference tests are valuable to compare test results to these thresholds.

 

A greater number of variables come into play when considering a liking threshold to set an action standard. For instance, the exact structure of a rating instrument, such as a 9-point hedonic scale or a 7-point numerical liking scale, will produce different measures of hedonic difference. A difference of 0.5 on a 9-point word category scale will be different from the same difference on a 7-point numerical scale and may even be different from a 9-point endanchored scale. Paired preferences are not subject to these types of effects. However, it is not always possible or costeffective to use paired preferences. In these situations, a sequential monadic presentation may be used and average liking ratings calculated. Converting these ratings into expected preference proportions would provide effect size information that can be referred to a preference action standard. This report will provide an approach to making that conversion based on Thurstonian models of different types of hedonic data.

​

​

​