re: Re: st: Different coefficient magnitudes in ols and 2sls estimation

[Christopher Baum <baum@bc.edu>]

<>
> We get a Sargan Chi-sq (overidentification test of all instruments) of
> of 6.259, p-value 0.0124.

I would try estimating the equation with robust or possibly cluster() options, which will give you a Hansen test. The Sargan test assumes IID errors and is not reliable 
in their absence. But to take the stat above at face value, your IV estimates are not consistent.  That, by the way, invalidates the Hausman test, which assumes
they are always consistent. The "Hausman" test in -ivreg2- can be computed with endog(). 

It may be that you will get better results out of the Hansen J statistic when you allow for a more general VCE, but if not, you must seek better instruments to
make the case that this IV estimate is any improvement over OLS.

If OLS is soundly rejected by a Durbin-Wu-Hausman strategy, then of course the point estimates of the coefficients are significantly different, both statistically
and to the naked eye -- that is what the Hausman-type test is testing, in fact!

Kit Baum   |   Boston College Economics and 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/