Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: xtpcse vs. xtgls / polsci vs. econ


From   "Clive Nicholas" <Clive.Nicholas@newcastle.ac.uk>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: xtpcse vs. xtgls / polsci vs. econ
Date   Mon, 26 Apr 2004 23:40:02 +0100 (BST)

Moritz Schularick wrote:

> I've been running both -xtpcse and -xtgls on my
> serially correlated and heteroskedastic, unbalanced
> panel (N=50, T=30).

[...]

> However, I was not prepared to see (for some
> variables) markedly different coefficient estimates
> (including switching signs) between the two, but
> that's exactly what I get.

Take a look at Table 1 in Beck's (2001) paper: you should find that
opposite signs ought not to be _that_ common when running -xtpcse- versus
-xtgls- (how many have you found in your model?), but they are _not_
impossible to find with datasets of your size.

> Given heteroskedasticity, the coefficient estimates
> via GLS (xtgls) could be more efficient than OLS
> (xtpcse), but I am not sure how the small T, large N
> problem affects the reliability of coefficient
> estimates in -xtgls? Does anybody know?

As far as my reading of the classic papers of pooled analysis make out, it
is the reliability of the _standard errors_ of FGLS parameters (where N is
small(ish)) that is the problem, _not_ that of the coefficients. Monte
Carlo tests alone appear bear this out.

In an indirect apology to David Greenberg, I must confess that I
misunderstood the problems that arise with fitting FGLS models. As Brian
Poi informed me privately, FGLS estimators are indeed correct when N is
_large_ (though how large is large is a moot point: it is not addressed
directly in the pooled-analysis literature when one leaves the realm of
OLS-PCSE and enters the kingdom of FGLS).

As a practical tip, Moritz, try running a different model (comparing
-xtpcse- with -xtgls-) with a different dataset and see what happens.
Whatever does happen, since N > T in your dataset, you will not be able to
correct for cross-panel correlation, which may well affect the accuracy of
your OLS-PCSE estimators (another tip for which I'm grateful to Brian).

I'm afraid I cannot answer your final question, since I'm not an economist!

CLIVE NICHOLAS        |t: 0(044)191 222 5969
Politics              |e: clive.nicholas@ncl.ac.uk
Newcastle University  |http://www.ncl.ac.uk/geps

Beck N (2001) "Time-Series-Cross-Section Data: What Have We Learned in the
Past Few Years?" ANNUAL REV POLIT SCI 4: 271-93.
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index