So far I have checked the tolerance value, VIF and condition indexes. But checking the variance of the regression coefficients I have to wonder: how little variance of the regression coefficient should be associated with the smallest eigenvalue and what is too much (indicating multicollinearity)?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
The rule of thumb I've often read is VIF > 5 indicates a level of multicollinearity worth investigating (and Tolerance is just the reciprocal of VIF).
-
$\begingroup$ Yes, I've read similar information about the VIF. So if the VIF value doesn't give a reason for concern I don't have to check the other values like condition index? $\endgroup$– JenniferCommented Feb 20, 2014 at 6:59
-
$\begingroup$ No I'd look at both VIF and condition index - also see this similar question: stats.stackexchange.com/questions/4099/…. Keep in mind if you're building a predicitve model, a little multicollinearity isn't necessarily bad since coefficient estimates are unbiased even if SEs are inflated. $\endgroup$– RobertFCommented Feb 20, 2014 at 17:52