Re: st: fixed effects with clustering when the number of levels of variable to be absorbed exceeds number of clusters

 From "Austin Nichols" To statalist@hsphsun2.harvard.edu Subject Re: st: fixed effects with clustering when the number of levels of variable to be absorbed exceeds number of clusters Date Thu, 2 Mar 2006 14:39:38 -0500

Mark Schaffer and I had a short conversation off-list, and it seems
that neither of us is sure what the consequences are when the number
of levels of the absorb() variable used in -areg-, or the number of
panel fixed effects more generally, exceeds the number of clusters. By
"consequences" I am thinking specifically of interpreting the standard
errors on multiple coefficients (which is really the point of using
the -cluster- option, isn't it), or using the less-than-full-rank
\hat{Var(\beta)} to test hypotheses.

Does anyone else want to weigh in?

Here is an interesting response from Gustavo Sanchez, Stata employee,
to a question about the text I copied from -help areg- into my first
_______________________________________________
Dear Austin,

Thanks for taking the time to bring this to our attention.

You wrote: Apparently, the help file for -areg- contains an inaccurate
caution, quoted twice in the attached email thread, and beginning
"Note: Exercise caution..."

The note is actually out of date and it will be removed from the
manual.

There were some previous simulation results that supported the
explanation in the FAQ. Our development staff is continuously checking
for new theoretical and empirical developments. When new results
provide evidence that may affect the performance of the estimators
implemented in Stata, the corresponding command might be modified to
empirically tested.

In order to learn about the changes in our fixed effects estimator, type,
"help whatsnew", then, scroll down to "update 15sep2005" and read item
-areg- was also modified accordingly.

Finally, let me point out that -xtreg- is a newer command and it is part of
our panel data [xt] commands. In general, any improvements to our fixed
effects estimator for linear regression are going to be mainly developed on
-xtreg- (although we would ensure the same behavior for the common
options in -areg-). Thus I would suggest that you use -xtreg- instead
of -areg- unless you are interested in particular features of this
last command that are not present in the former.

Please, let me know if you have any further questions
Sincerely,

Gustavo

___  ____  ____  ____  ____     Gustavo Sanchez, Ph.D.
/__    /   ____/   /   ____/     Statistician
___/   /   /___/   /   /___/      tech-support@stata.com
StataCorp LP
4905 Lakeway Drive
College Station TX 77845
_______________________________________________

Final Note (from me, not Gustavo):
An important distinction is that -areg- allows all weights except iweights,
which is not true of -xtreg, fe- so I would dispute the generality of
the claim implied by this line in -help areg-:
"Note:  See the command xtreg, fe in xtreg for an improved version of areg."

*
*   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/