Re: st: RE: problem using xtivreg2 without instruments

[Samia Tavares <samia.tavares@rit.edu>]

Mark,

I tried without the weight as you suggested, and the standard errors  
from xtreg were slightly higher than those with xtivreg2. In fact, I  
tried dropping one of the interaction variables, which eliminated the  
error message, and still the standard errors were different (xtreg  
slightly higher than xtivreg2). Any ideas what could be wrong?

Thanks for the help!

Samia

On Apr 14, 2007, at 1:16 PM, Schaffer, Mark E wrote:


> Samia,
>
> > -----Original Message-----
> > From: owner-statalist@hsphsun2.harvard.edu
> > [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of
> > Samia Tavares
> > Sent: 14 April 2007 17:33
> > To: statalist@hsphsun2.harvard.edu
> > Subject: st: problem using xtivreg2 without instruments
> >
> > Hi everyone,
> >
> > I'm trying to estimate a fixed effects regression with
> > analytical weights. I understand that this is not possible
> > with xtreg (only if I were using random effects), so I am
> > using xtivreg2 instead, since the help file indicated that I
> > could use it without instruments to obtain the same results I
> > would get with xtreg. The equation I am estimating is of the form:
> >
> > xi:xtivreg2 y dummy control i.year*continent1 i.year*continent2
> > i.year*continent3 i.year [aweight=variance^-1], fe cluster(country)
> >
> > When I estimate this, I get the following:
> >
> > Error: estimated covariance matrix of moment conditions not
> > of full rank;
> >         overidentification statistic not reported, and
> > standard errors and
> >         model tests should be interpreted with caution.
> > Possible causes:
> >         singleton dummy variable (dummy with one 1 and N-1 0s or vice
> > versa)
> > fwl option may address problem.
> >
> > I read in the FAQs about this error, but in the context of
> > including instruments with xtivreg2. I have checked the
> > dataset and none of the dummy variables (dummy, continent,
> > continent2, continent3) contain a singleton. Obviously since
> > I have no instruments, I am not concerned about the
> > overidentification statistic, but I want to be sure about the
> > standard errors.
> >
> > When I estimate the model without the year-continent
> > interactions, I don't get this error. However, I do wish to
> > include them to check for the robustness of my results.
> >
> > Any help will be greatly appreciated!!
> The error message is correct, I think.  Most likely, all the
> interactions mean that your var-cov matrix is not of full rank because
> you have more regressors than clusters.  Since you're not doing any
> instrumenting or efficient GMM, this will probably cause you only  
> minor
> problems at most - you won't, for example, be able to test the joint
> significance of more variables than clusters.
>
> A useful cross-check would be to estimate using xtivreg2 without
> weighting and using xtreg, both with cluster.  The SEs should be the
> same, but xtivreg2 will again warn you.
>
> Cheers,
> Mark
>
>
> > Samia
> >
> >
> >
> > *
> > *   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/
> *
> *   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/
*
*   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/