Re: st: cluster issues: var in the regression and wald chi2 for probit

[Austin Nichols <austinnichols@gmail.com>]

Luca Fumarco <luca.fumarco@lnu.se>:
No, you do not have to omit the complete set of occupation dummies you
are including with i.occupation, but you should be aware that 11
clusters will tend to produce SEs biased toward zero, leading to
rejection rates well above the nominal level. The cluster-robust
estimator of variance is consistent, meaning that as the number of
clusters approaches infinity it gets arbitrarily close to the truth,
not that you can use it with a very small number of clusters and
expect good performance (though 50 seems a good rule of thumb
threshold for "close enough to infinity"). The model stat is missing
because you have too few df to test all coefs. Read:
http://repec.org/usug2007/crse.pdf
http://fmwww.bc.edu/repec/dsug2011/desug11_schaffer.pdf
and the references therein.

On Thu, Dec 8, 2011 at 10:22 AM, Luca Fumarco <luca.fumarco@lnu.se> wrote:
> Dear Statalister
>
> I have a couple of questions:
>
> ->when I cluster on a variable, this variable has to be omitted from the regression?
> ex.
> xi: hetprob callback experiment Immig i.occupation, het(Immig) nolrtest vce(cluster occupation)
> (where occupation is an ordered variable)
>
> ->when I do so, Stata warns me that "chi2 model statistic has been reported as missing" (see the dot after Wald chi2)
*
*   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/