Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Multicollinearity: VIF vs. variance decomposition


From   "Lucas Bremer" <bremer@bwl.uni-kiel.de>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: Multicollinearity: VIF vs. variance decomposition
Date   Wed, 18 Nov 2009 08:51:26 +0100

Dear all,

I have some problems in detecting harmful multicollinearity in my random
effects panel regression.

To use the VIF as an indicator I made a linear regression and had a look at
the VIFs (estat vif)

Here are 6 (out of 20) variables with VIFs ranging from 8 to 15. Now I'm
interested between which of my independent variables problems due to
multicollinearity arise.

Therefore I used the variance decomposition (coldiag2). Now I'm a little bit
confused about the results because two of my variables have a very high
VIF>10, but in the variance decomposition there is no other variable with a
variance proportion > 50 for the same eigenvalue. This indicates that there
is no quasi-linear relationship for this variable.

Which measure can I trust if they have from my point of view contradicting
results?

By the way, someone told me that it is quite normal to have higher VIFs in a
panel regression and that the thresholds could be higher? Is this right?
What intuition is behind that statement?

Thanks in advance for your help,
Lucas 



*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index