On Jun 3, 2004, at 7:33 AM, Clive wrote:
Stupid question, really! I forgot that, of course, FE models models the
time series whilst 'averaging out' the cross-sectional variation, which
- -areg- does not. I knew that: honest!
Still, the quandary of whether or not robust standard errors under FE
is
possible remains. I notice that FE is the only model (FGLS and -areg-
included) that reports sensible constant values for my data, which is a
plus for me.
Wrong, wrong. The model underlying areg, "LSDV" (least squares dummy
variable), aka the within estimator (because it will only have
explanatory power if the variation WITHIN each individual's timeseries
is systematically related to the variation in the regressors WITHIN
that TS) is algebraically identical to that estimated by xtreg, fe. The
difference is that, in Intercooled Stata, areg will not work if the
number of panel units > 2000, and xtreg,fe will, since it does not
actually have to create all of those dummies.
The reason why xtreg does not have a robust option and areg does, I
imagine, relates to the date of each of those routines. xtreg also does
not allow for timeseries ops (D. F. L.) which is a real nuisance. But
when the TS ops were added to Stata, no one went back and rewrote
xtreg; they wrote a bunch of groovy new XT commands. So there is no
technical reason why xtreg, fe could not be robustified.
*
* 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/
