Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: RE: a question about option orthog for command ivreg2


From   "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>
To   "Jian Zhang" <jzh@ucdavis.edu>, <statalist@hsphsun2.harvard.edu>
Subject   RE: st: RE: a question about option orthog for command ivreg2
Date   Mon, 21 Aug 2006 09:37:55 +0100

Jian Zhang, 

> -----Original Message-----
> From: Jian Zhang [mailto:jzh@ucdavis.edu] 
> Sent: Sunday, August 20, 2006 4:16 AM
> To: Schaffer, Mark E; statalist@hsphsun2.harvard.edu
> Subject: RE: st: RE: a question about option orthog for 
> command ivreg2 
> 
> 
> Thanks, Mark!  But when I run ivreg2 (using the updated 
> version), the results reported that (right after the first 
> stage regression results):
> 
> 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.  See help ivreg2.
> 
> It stated "overidentification statistics nor reported".  
> However at the end of the second stage regression, it showed 
> IN FACT the value of the Hansen J statistic 
> (overidentification test of all instruments).  Why did STATA 
> report overidentification statistics after it already stated 
> "overidentification statistics nor reported"?  This seems 
> inconsistent. 

The error message appears within the first-stage regression results,
right?  Then it refers to the first-stage regression results, not the
main results.  You are right to point out that this is inconsistent.  It
arises because ivreg2 is called recursively to do the first-stage
regression, in which (of course) there are never any overidentification
stats reported.

Probably we should fix this. 

> Also if estimated covariance matrix of moment conditions is 
> not of full rank, will we not be able to obtain estimates of 
> GMM?  It seems to me that if this matrix is not of full rank, 
> this means that we could not get the weighted matrix (since 
> the weighted matrix is the inverse of estimated covariance 
> matrix of moment conditions).  Is this right?

Right.  Again explained by this relating to the first stage regressions,
not the main eqn.

> In addition, I have been trying the examples provided at the 
> end of the documentation for ivreg2 using the Griliches data. 
>  It seems that there is a problem on the consistencies of the 
> results from different overidentifiying restriction testings.
> 
> First I run regression:
> 
>  ivreg2 lw s expr tenure rns smsa _I* (iq=med kww age mrt), 
> gmm orthog(age mrt)
> 
> The Hansen J statistic indicated that instruments med kww age 
> and mrt are not  valid (i.e., correlated with the error 
> term).  The C statatistic indicated that the subset of 
> instruments age and mrt are not valid.  However, when I run 
> the following regression:
> 
> ivreg2 lw s expr tenure rns smsa _I* (iq=age mrt), gmm 
> 
> the Hansen J statistic indicated that instruments age and mrt 
> are valid.  It seems that two sets of regressions gave 
> inconsistent results on the validity of instruments age and 
> mrt.  Why does this 
> happen? Which result should we trust?

Roughly speaking, med and kww identify a different parameter value for
the coefficient on iq vs. the one identified by age and mrt.  Which one
you believe depends on other information: your priors, or strength of
identification, or whatever.  For example, you might have a priori
doubts about whether a variable is a plausible instrument, or you might
put more weight on the med-kww results because age-mrt are weak
instruments, or....

Cheers,
Mark

> I found your emaill 
> response to someone else in 2004 in 
> the statalist but could not really find an answer to this 
> from your email response.  Many thanks!
> 
> Best regards,
> Jian Zhang
> 
> 
> 
> 
> 
> 
> 
> > Jian Zhang,
> > 
> > > -----Original Message-----
> > > From: owner-statalist@hsphsun2.harvard.edu
> > > [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Jian 
> > > Zhang
> > > Sent: 18 August 2006 21:23
> > > To: statalist@hsphsun2.harvard.edu
> > > Subject: st: a question about option orthog for command ivreg2
> > > 
> > > Dear Statalisters,
> > > 
> > > I am using ivreg2 command to implement GMM estimation on a panel 
> > > data set.  I specified option orthog( ) to examine if a 
> subset of my 
> > > instruments are really exogenous (by putting the list of 
> the subset 
> > > of instruments in the parentheses of orthog).  However, 
> the results 
> > > reported by STATA say
> > > that: 
> > > 
> > > Collinearity/identification problems in restricted equation:
> > >   C statistic not calculated for orthog option
> > 
> > First, you should check that you are using the latest version of 
> > -ivreg2-; I think the wording of this error message is from 
> an earlier 
> > version.  The problem with the wording is that when the subset of 
> > instruments is tested, the equation is, if anything, better 
> described 
> > as UNrestricted - because the variables listed are no 
> longer "restricted"
> > to being assumed exogenous.
> > 
> > Anyway... after you update, you can find out what is going on by 
> > estimating the 2nd equation by hand.  Put the instruments listed in
> > orthog() in either the list of endogenous regressors (if they were 
> > exogenous regressors in the original eqn), or omit them 
> entirely (if 
> > they were excluded instruments in the original eqn).  FWIW, 
> my guess 
> > is that the 2nd equation is either exactly identified or 
> unidentified.
> > 
> > Cheers,
> > Mark
> > 
> > Prof. Mark Schaffer
> > Director, CERT
> > Department of Economics
> > School of Management & Languages
> > Heriot-Watt University, Edinburgh EH14 4AS tel 
> +44-131-451-3494 / fax 
> > +44-131-451-3296
> > email: m.e.schaffer@hw.ac.uk
> > web: http://www.sml.hw.ac.uk/ecomes
> >  
> > > Is there anyone knowing what is going on?  Since STATA 
> has reported 
> > > the estimates of the restricted equation(i.e., the 
> regression using 
> > > the entire set of instruments (overidentifying 
> restrictions)), why 
> > > did STATA still state "collinearity/idenfification problems in 
> > > restricted equation"?  Many thanks!
> > > 
> > > Best regards,
> > > Jian Zhang
> > > 
> > > 
> > > *
> > > *   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index