st: Xtoverid, cluster(clusvar) noisily

[May Ster <mayfrank@gmail.com>]

Dear Statalist,

I'm trying to get the correct (robust) S.E. for the Hausman Taylor
using -xtoverid, cluster(clusvar) noisily - option (after using
xthtaylor).
For your information, i have a panel data set which comprises of  N=100 , T=38.

I simply do not know which variable to choose to be clustered ; as in
to put in the cluster(clustvar) option.
I've tried to put different variables just to try it out. As expected
, the results have shown to be different in terms of S.E and the
overidentification test given.

Basically, my question is ...In order to choose a variable to be
clustered, what underlying assumption one has to make ?. Please help.

Best Regards,

may
*
*   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/