Livia Dancelli

When cluster analysis is performed on ranking or rating data, methods requiring quantitative variables commonly used to characterize the obtained groups, such as the cluster profile plots, are not appropriate. Instead, the bipolar mean, originally introduced in the literature in 2005 to deal with such kind of data, can be useful to interpret the resulting clusters, possibly in association with other available information. An application on real data coming from an extensive survey carried out in 2011 in the Italian McDonald’s restaurants is presented. A selection of ranking data, regarding some features of products and service, was analysed by a hierarchical cluster algorithm. In order to emphasize the concordance between the most important ranks, a weighted rank correlation coefficient was employed to measure the dissimilarity between respondents. Five groups were finally obtained, which show interesting differences on the given rankings.