GVARCLUS Procedure

This section compares the GVARCLUS procedure with the VARCLUS procedure in SAS/STAT software.

PROC GVARCLUS interprets the conditional dependency among variables by estimating the inverse covariance matrix. The off-diagonal elements of an inverse covariance matrix correspond to partial correlations, so the zero elements imply conditional independence between the pair of variables. The conditional independence provides a better model for understanding the direct link between variables than does simple correlation analysis, which models each pair of variables without considering other variables.

PROC VARCLUS performs variable clustering based on latent components. It assigns a variable to its most correlated latent component, so eventually the variables within a cluster have high correlation.

The primary difference between the two procedures is that PROC GVARCLUS considers the association between variables after the other effects have been factored out.