[R-pkgs] perturb package for evaluating collinearity

John Hendrickx john_hendrickx at yahoo.com
Fri Apr 1 00:16:39 CEST 2005


I've uploaded the R package "perturb" to CRAN. Perturb contains two
programs for evaluating collinearity. "Colldiag" calculates condition
indexes and variance decomposition proportions to detect and track
down collinear sets of variables. 

"Perturb" takes a different approach. It re-estimates the model a
specified number of times, adding random noise ("perturbations") to
selected variables at each iteration. Categorical variables are
randomly reclassified. A summary then displays the impact of these
perturbations on the stability of the coefficients. Perturb is not
limited to linear regression and can handle interaction effects and
transformations of independent variables.

Version 1 of perturb was available from my website. This new version
makes it much easier to reclassify categorical variables. In
addition, colldiag now has a "fuzz" option to suppress printing of
small variance decomposition values.

Comments and criticisms invited,
John Hendrickx




More information about the R-packages mailing list