Stas,
Thanks very much for your help. You're totally right about the choice of a
model, but since this is for my PhD I need to show all my results even if
the former invalidate the latter), and go step by step.
The main reason why I'm estimating OLS is that my approach has not been done
much in panel, and I need OLS results to compare with previous literature.
Thanks again, I'll have a look on -xtdata-
Andrea
----- Original Message -----
From: "Stas Kolenikov" <[email protected]>
To: <[email protected]>
Sent: Tuesday, August 03, 2004 9:24 PM
Subject: Re: st: Testing OLS assumptions in panel data
> Andrea,
>
> this does not look methodologically consistent to me. If you know that OLS
> is wrong, and there is a panel effect (which you test with -xttest0-),
> then you probably don't need all other tests for OLS, as you are ready to
> give up with it. Once you admit the panel structure, the auxiliary
> regressions such as in -hettest- or -ovtest- won't get the correct
> statistics as they assume i.i.d. errors, which you already know is wrong.
> If you don't reject the OLS, then all other estimators are inefficient, so
> why bother with them?
>
> Also, most of the normality tests are superfluous, anyway, as long as the
> residuals even in the OLS case are bound to be heteroskedastic for
> homoskedastic errors: the variance of e_i is sigma^2 (1-h_i), where h_i is
> the hat-value, the diagonal entry of X'(X'X)^{-1}X'... so their
> distributions are not the same for different i, the observation number.
> I've never tried to figure that out for the panel case, but I don't think
> it is going to get any simpler.
>
> Having said that, may I suggest that you look at -xtdata- that I mentioned
> on the list a few days ago to somebody in a different thread. It does
> various panel data transformations (fixed effects, random effects, between
> transformation), but Stata Corp. warns that is sole value is in trying out
> different model specifications which tend to be somewhat computationally
> faster with -regress- rather than with -xtreg-.
>
> In statalist, Andrea Molinari wrote:
>
> > I'm trying to get some of the typical OLS tests (heteroscedasticity,
> > normality and missing observations) after a pooled OLS, a between and a
> > within a between estimations. I also need to look at the estimations
> > with robust std. errors. For pooled OLS, I ran (respectively) the
> > commands hettest, jb (on the estimated residuals), and ovtest. My
> > problem started when running panel data commands. After searching on the
> > web and manuals, the only command that I could find to do (part of) this
> > is running xttest3 after estimating a fixed effects model, which gives a
> > heteroscedasticity test after a fixed effects estimation. I could also
> > correct for this with areg depvar varlist, abs(group) robust. pantest2
> > gave me a normality test for the residuals.
> >
> > I couldn't, however, run any of the three tests, nor getting the robust
> > std. errors, for the between estimates. And I couldn't get a test for
> > omitted variables for the fixed effects model.
> >
> > If anyone has any suggestions on how to get (some of) these tests,
> > they'll be more than appreciated!
>
> --- Stas Kolenikov
> -- Ph.D. student in Statistics at UNC-Chapel Hill
> - http://www.komkon.org/~tacik/ -- [email protected]
>
> * This e-mail and all attachments to it are not intended to provide any
> * reasonable point of view and was transmitted to you in error. It
> * should be immediately deleted by all recipients unless they really
> * enjoy communicating with the author :). Other restrictions apply.
>
> *
> * 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/
