Hi Tim,
Variance proportions (and condition indices) are exactly the tools
described in Belsley, Kuh & Welsch, "Regression Diagnostics" - see my
previous post. Good to see I'm not the only one to use them! BKW also
describe in detail how to calculate all this using SVD, so you don't
need to use SAS...
And I certainly agree that a problematic system means that you need to
do more work - probably either collect more data or refine your research
agenda, as collinearity may just be inherent in the independent
variables you have been collecting.
Best,
Stephan
Tim Paysen schrieb:
Actually, the CI index and VIF are just a start. It is best to look
at what they call a matrix of "variance proportions" (found in SAS
and a few other places...)--which hardly anyone understands
(including the SAS folks). It is a matrix of estimates of what the
variences of the regression coefficients would be if you could figure
them out in the first place. It shows which factors dominate over
others IN THE PARTICULAR SETUP you are analyzing. The matrix is
often calculated using eigenvalues, but is best done with Singular
Value Decomposition techniques (you don't have to have a square
matrix, and you maintain better precision). Analysts will say that
it can display an unstable system -- which is correct, but they
generally say that, if its true, you have bad data and should throw
it out--or collect more. I suggest care, because it may be
illustrating the nature of the system you are studying.
The only decent reference that I know of is a little book (hard to
read) that I can't remember off the top of my head. Have to look it
up.
Timothy E. Paysen, Phd Research Forester (ret.)
________________________________ From: John Sorkin
<jsor...@grecc.umaryland.edu> To: Alex Roy <alexroy2...@gmail.com>;
r-help@r-project.org Sent: Tuesday, July 21, 2009 4:19:11 AM Subject:
Re: [R] Collinearity in Linear Multiple Regression
I suggest you start by doing some reading about Condition index (CI)
and variation inflation factor (VIF). Once you have reviewed the
theory, a search of search.r-project.org (under the help menu in a
windows-based R installation) for VIF will help you obtain values for
VIF, c.f. http://finzi.psych.upenn.edu/R/library/HH/html/vif.html
John
John David Sorkin M.D., Ph.D. Chief, Biostatistics and Informatics
University of Maryland School of Medicine Division of Gerontology
Baltimore VA Medical Center 10 North Greene Street GRECC (BT/18/GR)
Baltimore, MD 21201-1524 (Phone) 410-605-7119 (Fax) 410-605-7913
(Please call phone number above prior to faxing)
Alex Roy <alexroy2...@gmail.com> 7/21/2009 7:01 AM >>>
Dear all, How can I test for collinearity in the predictor data set
for multiple linear regression.
Thanks
Alex
[[alternative HTML version deleted]]
______________________________________________ R-help@r-project.org
mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do
read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
Confidentiality Statement: This email message, including any
attachments, is for th...{{dropped:11}}
------------------------------------------------------------------------
______________________________________________ R-help@r-project.org
mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do
read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
