Re: st: regress with vce(robust) and hascons

[Steven Samuels <sjsamuels@gmail.com>]

I am asking for a note something like: "note: The F statistic tests  
the hypothesis that the mean of every observation is zero, because  
options -hascons- and -vce(cluster)- are both specified."

The F statistic with -hascons- and -vce(cluster)- tests a hypothesis  
different from that of any of the other three combinations of - 
hascons- and -vce(cluster)-.  Certainly the reasons for this are  
valid.  But both Michael and I were surprised, and, if so, I imagine  
that other users would be and would appreciate the note.


On Mon, Dec 13, 2010 at 1:00 PM, Steven Samuels <sjsamuels@gmail.com>  
wrote:
> Thanks, Jeff. I suggest that, when the vce(robust) and -hascons-  
> options are
> present, the results contain a remark about the change in the F test.

Which exactly change? You are using a different estimator of a
variance-covariance matrix for the Wald test, so naturally your
results are going to be different. If you specify -vce(robust)-, you
admit that your data are heteroskedastic, and the sums of squares do
not make sense, so F-test is only available as Wald test, not the
sums-of-squares-ratio test. In this sense, it is more akin to chi2
reported with -ml- models.

--
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/