GENERALIZATION OF PREFERENCE MAPPING TO I NCOMPLETE BLOCK-DESIGNS

Classical techniques of Preference Mapping require that each consumer gives a preference score to each product. In many situations, it is di fficult to ask a consumer to taste more than, say, four or five sample s, while the total number of products in a Preference Mapping study is usually greater than seven. Mutually Orthogonal Latin Squares have be en recommended in these situations for designing incomplete experiment s, balanced for the order and carry-over effects. The paper proposes a technique to derive homogeneous clusters of consumers from incomplete data sets. Each cluster can then be summarized by its vector of produ ct mean scores, making it possible to apply any technique of preferenc e mapping on these smaller, but complete, data sets. A Monte-Carlo sim ulation was undertaken in order to study the accuracy of this method w ith respect to several parameters, The most important parameters were the number of products assessed by each consumer and the complexity of the preference structure. Technically speaking, complete data sets we re simulated according to a multi-cluster model and then made incomple te by removing a given number of products for each consumer without di sturbing the balance of the design. Clustering from these simulated in complete data sets, based on the proposed method, was then compared to that obtained from the complete data sets.