Re: st: Multicollinearity in panel data

[Maarten buis <maartenbuis@yahoo.co.uk>]

--- On Wed, 9/2/11, Jing Zhou wrote:
> > I am using FE model to test x and x^2 in a panel data set.
> > the VIFs of x and x^2 are more than 10 (about 11 and 12),
> > suggesting that there exists some multicollinearity between
> > the two variables. Should I consider about this problem in
> > the FE estimation and if so, how can i deal with it? 

--- On Wed, 9/2/11, Johan Hellstrom wrote:
> As x^2 are an interaction term of it self; x and x^2 will be
> somewhat collinear. Nothing to worry about.

In general I agree with Johan. The one exception I would make
is when all values of x are large and well away from 0, e.g.
year of birth. In those cases x^2 can become very close to a 
linear function of x, and you would loose a lot of precision.
The good news is that there is an easy solution: just substract
a reasonable value from x before computing the square.

*-------- begin example -----------
sysuse nlsw88, clear

gen byr = 1988-age
gen byr2 = byr^2
corr byr*

gen cbyr = byr - 1950
// Why did I choose 1950?
// It is a nice round number somewhere in
// the middle of the range of byr.
gen cbyr2 = cbyr^2
corr cbyr*
*-------- end example --------------

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------



      

*
*   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/