Model-based clustering with parsimonious covariance structure

Complex multidimensional concepts are often explained by a tree-shape structure by considering nested partitions of variables, where each variable group is associated with a specific concept. Recalling that relations among variables can be detected by their covariance matrix, this paper introduces a covariance structure that reconstructs hierarchical relationships among variables highlighting three features of the variable groups. We finally present an application of the latter covariance structure to the model-based clustering.