----- Original Message -----
From: "tom blade" <[email protected]>
To: <[email protected]>
Sent: Friday, July 04, 2003 11:54 AM
Subject: st: which command is better?
> Dear friends,
>
> I have a doubt. I am working with panel data and have a problem of
heteroskedasticity. So I could choose among these alternatives:
>
> 1) xtgls, p(h)
> 2) areg, robust
> 3) regress with id dummies and option robust
>
> 2) and 3) give the same coefficients than xtreg, fe but different standar
errors. xtgls gives different coefficients and standar errors.
>
> Which is the best solution?
>
> Any help will be much appreciated.
>
> Sincerely,
> Tom
>
If you are comparing the alternatives to -xtreg, fe- then should not the
first alternative include id dummies - a fixed effects GLS(?).
Example
. use http://www.stata-press.com/data/r8/grunfeld.dta
. xtreg invest mva kst time, fe
. xi i.c
. xtgls invest mva kst time _*
These two approaches will produce the same parameter estimates.
Wooldridge in "Econometric Analysis of Cross Section and Panel Data" page
276 has the following to say about the Fixed Effects GLS:
"Rather than compute a robust variance matrix for the FE estimator, we can
instead relax Assumption FE.3 to allow for an unrestricted, albeit constant,
conditional covariance matrix. This is a natural route to follow if the
robust standard errors of the fixed effects estimator are too large to be
useful and if there is evidence of serial dependence or a time varying
variance in the u(it)."
However, given that Wooldridge goes on to discuss how the E(u'u) matrix is
of deficient rank and therefore should be estimated with one of the time
periods dropped, make me wonder if -xtgls- with id dummies is the correct
procedure. Perhaps the wizards at Stata or on this list could shed some
light on this topic.
Scott
*
* 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/
