A statistical procedure for clustering ordinal data

This article presents a testing procedure for comparing ordinal data distributions which helps the interpretation of results in presence of complex surveys involving the evaluation of several items or the evaluations expressed by different groups of respondents about a given item. For this purpose, a mixture model (denoted CUB) for ordinal data is considered. Specifically, Kullback-Liebler divergence is used in order to measure the dissimilarity among ratings distributions and a clustering technique is applied for grouping distributions. A case study on university teaching evaluation is finally illustrated.