and I am paraphrasing or paraplagiarizing below from a private email
from one of its authors.]
When you use the cluster-robust covariance matrix estimator, and the
number of clusters is less than the number of instruments, then the
estimated covariance matrix is not of full rank. Stata claimed at one
point that this made the estimated var-cov matrix unusable--the
clusters are like observations, and so it is a bit like running a
regression with more regressors than observations.
It is definitely the case that you can't test at least some sets of
hypotheses in these circumstances--e.g you can't do an F-test of the
joint significance of all the regressors. Try this with official
Stata and you will see that the F-statistic for the estimator is
missing and there is a hyperlink to -help j_robustsingular-. This
includes the following statement:
"There is no mechanical problem with your model, but you need to
consider carefully whether any of the reported standard errors mean
anything. The theory that justifies the standard error calculation is
asymptotic in the number of clusters, and we have just established that
you are estimating at least as many parameters as you have clusters.
Putting that concern aside, the model test statistic issue is that you
cannot simultaneously test that all coefficients are zero because there
is insufficient information. You could test a subset, but not all, and
so Stata refuses to report the overall model test statistic."
So it _may be_ that the reported var-cov matrix is OK for at least
some purposes, even though it isn't of full rank...
[end paraplagiarism]
If you think so, you can hack into the guts of -ivreg2- (comment out
line 1574 to start) and save a new copy called ivreg2b.ado or
somesuch, or you can run -ivreg- if you don't want the added
functionality of -ivreg2-.
ps. The record of official Stata positions on the dangers of the
less-than-full-rank cov matrix available at
http://web.archive.org/web/*/http://www.stata.com/support/faqs/stat/aregclust.html
is worth reading, IMHO.
On 3/14/06, Albrecht Glitz <[email protected]> wrote:
> I am trying to estimate an IV model with a large number of exogenous
> interaction terms and one endogenous variable that I instrument. Using
> ivreg2 (for the additional output it offers) I get the above error message,
> however, using ivreg works fine.
>
> Does anyone know what to do and whether I would be doing something
> wrong using the old ivreg command?
>
> Your advice would be greatly appreciated.
*
* 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/
