I asked
> I am wondering if -xtpcse- really performs just ordinary OLS, as Stata's
> help says. If this was so, a regression with -xtpcse- of mean-differenced
> data (within transformation) should give me the same coefficients as one
> with -xtreg, fe- on the original data, only with standard errors adapted
for
> heteroskedasticity and contemporaneous cross section correlation (assuming
> no autocorrelation). However, I get both different coefficients and an
even
> more different R-square. Does anyone have an idea of what could be the
> reason?
and Allen McDowell ([email protected]) answered:
When you removed the group means before fitting your model with -xtpcse- did
you also add the overall mean back in? That is what -xtreg, fe- does. An
easy
way to insure that you are using the same data is to use -xtdata, fe- to
perform the transformation and then use the resulting data with -xtpcse-.
Thank you Allen,
I corrected this, but this actually is not the problem since it changes the
intercept only, in which I am not interested anyway.
What I am wondering about is that although OLS on mean-differenced data
(LSDV - what I am doing with -xtpcse-) should give the same coefficients as
-areg, absorb(i)-, it does not. The coefficients differ quite a lot, and the
Rsq with -xtpcse- is only half that with -areg,absorb(i)-. (I know that
there is a difference in the calculation of Rsq, just would not have thought
it would be so large. Thus I would already be happy if somebody could say
something about why the coefficients might differ.)
I am comparing the two to have the choice between their different
adjustments for the SEs, thinking that that should be the only difference
between them.
Regards,
Markus Poschke
*
* 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/
