Models for Replicated Ratings
Bi, J. and Ennis, D. M. (1998). IFPress, 1(4) 2-3.
A competitor has introduced a new fragrance for young females into a market that your company dominates with an existing fragrance. Since these fragrances are quite different according to your perfumers, you decide to test them among separate consumer samples to avoid effects introduced by testing the different fragrances on the same individuals. Your main interest is in comparing the degree of liking for your fragrance and the new rival. However, you also suspect that there may be differences in individual hedonic response to the fragrances and in how individuals use rating scales. For these reasons you decide to use replicated tests in which each consumer evaluates only one type of fragrance. Replicated ratings data are collected from two young female groups with 50 and 54 consumers, respectively. The smaller sample is for your product. Six replications are obtained from each consumer. A 5-point liking scale is used where “1” means “dislike very much” and “5” means “like very much”.
Figure 1. Relationships among categorical data models. All of the models are special cases of the generalized Dirichletmultinomial model.