you could use the cluster estimator with the areg option if you have a full
set of dummy variables corresponding to the clusters'id. The command works,
but the FAQ, warns you about the rank of the covariance matrix, which is k,
the number of parameters estimated not including the constant or the fixed
effects. Does this mean that to increase the rank you should have more
parameters beside the dummies? This seems counterintuitive to me.
Other issues of concern are:
-Why, if I use the cluster option with reg, including manually a full set of
dummy variables, the F test on the coefficients is not reported on the basis
that the rank of the VCE is too small whilst the areg command reports it?
-Why the NL command with the cluster option behaves like areg and not like
reg?
-More generally, I found some papers Kezdi(2003),Bertrand et al(2002), cited
in Wooldridge: www.msu.edu/~ec/faculty/wooldridge/
current%20research/clus1aea.pdf
that seem to warrant the use of the cluster option even when N, is not
particularly big with respect to T, as far as it has reasonable size.
Kezdi(2003), performs his simulations demeaning the variables instead of
including dummies. I suspect (probably erroneously) that there is a
computational convenience behind his choice that may have to deal with the
problem of degree of freedoms in the VCE (not considering the algebraic
equivalence of the two methods). Thanks for any insight you may be able to
give me about the topic
Have a good day,
Mike
_________________________________________________________________
Invia allegati della dimensione di 10MB - Con MSN Hotmail puoi!
www.msn.it/hotmail/minisite_10
*
* 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/
