Using latent variables in model based clustering: an e-government application

Besides continuous variables, binary indicators on ICT (Informationand Communication Technologies) infrastructures and utilities are usually collected in order to evaluate the quality of a public company and to define the policy priorities. In this paper, we confront the problem of clustering public organizations with model based clustering and we assume each observed binary indicator to be generated from a latent continuous variable. The estimates of the scores of these variables allow us to use a fully Gaussian mixture model for classification.