GVARCLUS Procedure

The GVARCLUS procedure analyzes the relationships among variables in a multivariate analysis. It estimates the inverse covariance matrix which encodes partial correlations between pairs of variables given the other variables. At each step, a regularization parameter is used to control the sparsity of the estimated inverse covariance matrix. Zero elements imply conditional independence between variables and the variables are divided into disjoint clusters. Tuning the regularization parameter from low to high increases the number of disconnected components and splits larger clusters into smaller one. The divided clusters from sequential steps naturally form a hierarchical structure during the process.