Dear Hui Wang "Jackie"--
The observations you exclude *do* offer information about the effect
of x2 on y, even allowing for an id-specific intercept. If you left
out x2 as an explanatory variable, you would get exactly the same
coefficient in both specifications.
input id y x1 x2
1 4 1 2
1 13 0 6
1 2 0 5
2 5 0 7
2 7 0 9
3 9 1 5
3 10 1 3
3 12 0 1
4 12 1 4
4 15 1 7
end
iis id
xtreg y x1 x2, fe
xtreg y x1 x2 if id!=2 & id!=4, fe
xtreg y x1, fe
xtreg y x1 if id!=2 & id!=4, fe
On 11/3/06, Hui Wang <[email protected]> wrote:

Here is a question about the fixed-effect regression.
I want to estimate the effect of x1 on y (controlling for x2) using fixed
effect regression.
But notice that for id=2 and id=4, x1 does not vary within the same id. If
the fixed effect realy make use of the variation within the same id, may be I should exclude these observations.
The coefficient on x1 in these two regressions are closed but not exactly
the same, so are their standard errors.
My question is: which one make more sense?

*
* 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/