Hi Kristen,
VIF(betai) = 1 / (1 - R-squaredi), where R-squared i is the squared multiple correlation of predictor i with all other predictors. As you can see, no standard errors actually enter the VIF formula, so complex survey design adjustments (at least stratification and clustering) have no effect on the computed VIF - these only correct the standard errors post-estimation.
Cam
----------------------------------------
> From: [email protected]
> To: [email protected]
> Date: Thu, 13 Aug 2009 14:45:12 -0400
> Subject: st: VIFs under SVY command and -collin- command
>
> Hello Statalisters,
>
> I followed the UCLA help page instructions for checking collinearity (i.e. calculating VIFs) by doing the following while using the svy command :
>
> quietly svy: regress age5 raceth union ieduc povcat3 ireligion timeinUS
> display "tolerance = " 1-e(r2) " VIF = " 1/(1-e(r2))
>
> and then I also ran the following code using the -collin- command:
>
> collin age5 raceth union ieduc povcat3 ireligion timeinUS
>
> and I got almost the exact same (if not identical) VIF and tolerance values... It doesn't seem like it should be the same since the -collin- command doesn't account for the complex survey design. Can anyone make a suggestion as to why this might be? Also, I just noticed that the R-squared for the regressions (with and without the -svy- command) are similar which I guess explains why the VIFs are the same.
>
> Any thoughts are appeciated?
>
> Thanks!
>
> Kristen
>
> *
> * 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/
_________________________________________________________________
Attention all humans. We are your photos. Free us.
http://go.microsoft.com/?linkid=9666047
*
* 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/
