As noted in the thread referred to here, collinearity is a problem
among the independent variables. An often used approach is to run a
regression with any dependent variable and then run the collinearity
diagnostics.
I proposed an alternative approach in my posting to that thread.
Collinearity is a problem if independent variables are so strongly
correlated that estimates become unstable, i.e. if a small change to
one of the independent variables causes strongly different parameter
estimates. I've written an ado-program that will do this for you. It
estimates your model 100 times, adding a random value to specified
variables at each iteration. It's not limited to regression models
and it can handle interactions, transformed variables, and
categorical variables.
A priliminary version is available at
http://www.xs4all.nl/~jhckx/stata/perturb/
Take a look and let me know what you think.
John Hendrickx
--- john emmet <[email protected]> wrote:
> Dear Scott,
> Thanks very much for replying.
>
> I'm sorry, I must not have explained myself very clearly in my
> original
> posting.
>
> I did mention, though that if I were using a cross-sectional
> dataset I
> would probably use -collin- but as I am not I can't. The dataset
> is a
> panel set with 6 panels and I am using xtprobit (but could consider
> xtlogit)
>
> I suppose I am trying to evaluate the situation relating to
> multicollinearity with time series cross-sectional analysis.
>
> Any help or advice from will be very gratefully received.
>
> John
>
> > One possibility would to be to use -collin-
> >
> > I would also suggest you take a look at the thread started by
> Matt Barreto
> > on
> > December 10th, 2003 ("multicollinearity test for probit?"), and
> the
> > replies by
> > Richard Williams, Nick Cox, Joao Pedro W. de Azevedo, and John
> Hendrickx.
> >
> > Scott
> >
> > ----- Original Message -----
> > From: "john emmet" <[email protected]>
> > To: <[email protected]>
> > Sent: Monday, January 12, 2004 10:04 AM
> > Subject: st: collinearity and xtprobit, xtlogit
> >
> >
> >> Hi
> >>
> >> I am new to panel data analysis, and I don't seem to be able to
> forget
> >> my
> >> experience when using cross-sectional analysis in terms of
> avoiding
> >> multicollinearity amongst explanatory variables. All the
> analysis is
> >> using binary dependent variables.
> >>
> >> If I were doing logit I would probably use the collin command
> and
> >> examine
> >> the VIF, tolerance, etc.
> >>
> >> If I look at the correlation coefficients for the panel dataset
> (6
> >> panels)
> >> they are very high for some explanatory variables, but then.
> But there
> >> is obviously some collinearity between the RHS variables because
> when I
> >> put one variable in particular in the model, it affects the prob
> values,
> >> SEs and coefficients.
> >>
> >> Am I completely missing a point somewhere????
> >>
> >> Is there a test for using with xt models?
> >>
> >> *
> >> * For searches and help try:
> >> * http://www.stata.com/support/faqs/res/findit.html
> >> * http://www.stata.com/support/statalist/faq
> >> * http://www.ats.ucla.edu/stat/stata/
> >
> > *
> > * For searches and help try:
> > * http://www.stata.com/support/faqs/res/findit.html
> > * http://www.stata.com/support/statalist/faq
> > * http://www.ats.ucla.edu/stat/stata/
> >
>
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
__________________________________
Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
http://hotjobs.sweepstakes.yahoo.com/signingbonus
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
