All but Sarah:
Just to close this thread (for now): I surmised offlist that the FE
computed by -predict- were not the same as the FE computed by
demeaning coefs on dummy variables because the regression dropped some
collinear dummy vars, and the mean computation included zeros as
estimated FEs (when those FEs were not in fact
estimated). Given a section of code by Sarah, I suggested replacing
forv i=1/600{
qui replace fe=_b[d`i'] if school==`i'
qui replace se=_se[d`i'] if school==`i'
}
with
forv i=1/600{
qui replace fe=_b[d`i'] if school==`i' & _se[d`i']>0 & _se[d`i']<.
qui replace se=_se[d`i'] if school==`i' & _se[d`i']>0 & _se[d`i']<.
}
(before computing the mean of fe) and she reports that the -predict-
method now matches the coefs method.
On Feb 11, 2008 6:29 PM, Sarah Cohodes <[email protected]> wrote:
> I implemented your first solution on the sample data, and
> unsurprisingly, it worked perfectly. Interestingly, when implemented
> in my "real" data, the FE estimated by -predict- is not the same as
> the FE estimated by taking the demeaned betas from the dummy variables
> (it is the same in bpwide.dta). I find this slightly disturbing.
*
* 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/
