I took a closer look at the -reg-, -xtreg-, -logit-, -clogit- and -xtlogit- commands, and Sam's explanation seems to make sense: the -xt- commands seem to imply cluster() -- which in turn implies robust -- via their action on the groups, i().
Still, the manuals (v7.0) are not explicit about this. I guess what I'm really after, then, is the more detailed statistical explanation -- i.e. how does -clogit- estimate e(V).
The "methods and formulas" for -reg- and -logit- (and [U] 23.11 obtaining robust variance ...) do a decent job of explaining what the estimated variance-covariance matrix looks like (with and without cluster() and robust). But not so well for the -xt- command, particularly the fixed-effects logit, which is described under -clogit-.
Soren
I think it is because in non-xt- commands the use of cluster() implies
estimation of robust standard errors. If you check other commands (e.g.,
-reg-) that is what it will say. However, cluster is not a required
option in most non -xt- commands. Hence, it is possible to use robust
without using cluster in most commands. Yet, for -xt- commands, i(),
which is the analog to cluster(), is required, so there is no need for the
robust option. That is my understanding.
Hope this helps.
Sam
On Tue, 23 Jul 2002, Anderson, Soren wrote:
> Hi. I'm looking for a quick (footnote) explanation for why there is no "robust" option for the fixed effects (conditional) logit -- i.e., why can't I estimate:
>
> xtlogit depvar indvars, fe i(id) robust ?
>
> I couldn't find anything in the manuals.
>
> Thanks.
> Soren Anderson
*
* 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/