Re: st: Variable Inflation Factor (VIF) for non-linear models

[David Hoaglin <dchoaglin@gmail.com>]

Hi, James.

VIF stands for Variance Inflation Factor, the factor by which the
variance of the estimated coefficient of a predictor is inflated by
the relation between that predictor and the other predictors.  A
predictor's VIF equals the reciprocal of 1 - R^2, where the R^2 comes
from the multiple regression of that predictor on the other
predictors.  It may be helpful to think of that 1 - R^2 as the "usable
fraction" of the predictor.

In types of regression other than ordinary linear regression, the VIF
from the corresponding linear regression will not be the whole story
(as Nick pointed out), but it should be a helpful start.

A "large" VIF indicates the presence of a collinearity relation
involving the predictor, but it provides no further information.  To
understand the nature of the collinearity (or collinearities), I use
the diagnostics discussed in the book by Belsley, Kuh, and Welsch
(1980) and in a later book by Belsley.  For Stata they're implemented
in the user-written command -coldiag2-, available from SSC.

David Hoaglin

Belsley DA, Kuh E, Welsch RE (1980).  Regression Diagnostics. Wiley.

On Sun, Oct 27, 2013 at 6:42 AM, James Bernard <jamesstatalist@gmail.com> wrote:
> Hi all,
>
> I would like to know if after a count model (xtpoisson command) Stata
> can give me VIF of the predictors?
>
> I looked for it in the post-estimation section of -xtpoisson-, but
> could not find any thing. Is it a theoretical issues that VIF can not
> be calculated for count models? Or, is it a software issues?
>
> Is there any substitute for it in log-linear models? I don't want to
> use correlations to justify absence of multicollinearity.
>
> Thanks,
> James
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/