Clustering upper level units in multilevel models for ordinal data

We consider the issue of clustering upper level units in hierarchical data with an ordinal response. An example is the clustering of university courses (level 2 units) on the basis of the ratings expressed by students (level 1 units) using an ordinal scale, with the aim of identifying ‘good’ and ‘bad’ courses. Another example is the clustering of European regions on the basis of individual satisfaction levels on a Likert scale. In this framework, model-based clustering can be achieved by fitting multilevel models for ordinal data and then exploiting the predicted random effects.