<>
An earlier exchange discussed the behavior of overidentification test statistics via -overid- (SSC: Baum, Schaffer, Stillman & Wiggins) after -reg3- and -reg3, 2sls-. We have corresponded StataCorp developers, and they have pointed out that some of the differences we identified were arising because of degrees-of-freedom adjustments. If the -2sls- option on -reg3- is used, the -nodfk- option should also be invoked. Then reg3, applied to a single equation, will yield the same Sargan statistic as single-equation estimation via -ivreg2- or -ivregress-.
If -reg3- is used to estimate multiple equations, its estimated coefficients and covariance matrix will differ if the -2sls- option is specified (w/w/o -nodfk-). In that instance, the Sargan statistic after -reg3, 2sls- will not match that of -overid- after -reg3, 3sls-. If the full 3SLS estimator is used, the Sargan statistics of -ivreg2-, -ivregress- and -reg3- will match.
As far as we can see, both official -reg3- and user-written -overid- after -reg3- are behaving perfectly.
Kit and Mark
Kit Baum | Boston College Economics & DIW Berlin | http://ideas.repec.org/e/pba1.html
An Introduction to Stata Programming | http://www.stata-press.com/books/isp.html
An Introduction to Modern Econometrics Using Stata | http://www.stata-press.com/books/imeus.html
*
* 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/
