Re: st: collinearity adjustment

[Michael Norman Mitchell <Michael.Norman.Mitchell@gmail.com>]

Dear Fabio

  I wonder if you have collinearity arising from another source, other 
than including interaction terms. Maybe including output showing your 
model without the interaction terms and the VIF, then the model with the 
interaction terms and the VIF, and the model with the centered terms and 
the VIF would help provide the information to be able to diagnose the 
problem.

  I suggest this because, from what I am imagining, centering around 
the mean should have solved collinearity resulting from squared terms, 
regardless of the preponderance of zeros.

Best regards,

Michael N. Mitchell
See the Stata tidbit of the week at...
http://www.MichaelNormanMitchell.com

On 2010-03-15 1.27 AM, Fabio Zona wrote:
> Dear Statlists
>
> centering variables around their means is a standard procedure to solve the problem of multicollinearity which arises when dealing with interaction terms and linear/quadratic terms.
>
> I tried to center my variables, but, after centering, -collin- commands tell VIF values are still very high; maybe it depends on the fact that in my database there are thousands of zero-observations, and the mean is very close to zero: as a consequence, subtracting the mean from the variable does not change it enough to solve the problem of multicollinearity.
>
> Is my interpretation correct? Any idea on how I might solve this problem?
>
> Thanks a lot
>
>
> *
> *   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/
>    
*
*   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/