GVARCLUS Procedure

Elements that have the value 0 in a precision matrix imply conditional independence between variables given the other variables in the network. So the estimate of a precision matrix can recover the conditional independence structure, which can be displayed in an undirected graph. If node is connected to node by an arc, then is called a neighbor of . If is connected to though some chain of arcs, then is called a connected component of .

The regularization parameter in the penalized log likelihood can be seen as a quasi-measure for the strength of connection between and , because as increases from small to large, some connections drop out, and the connections that drop out should be weaker than those that stay in.