Ivar,
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of
> Ivar Pettersen
> Sent: 18 April 2006 10:40
> To: statalist@hsphsun2.harvard.edu
> Subject: st: RE: RE: Interpretation of Smith-Blundell exog. test
>
> Thanks Mark,
>
> I have been googeling this subject a while and found mixed
> approaches to finding the best instruments.
> Some seem to experiment with instruments until the
> Smith-Blundell-test is rejected (i.e. regressor is exogenous).
> Then they conclude that the instruments are valid. This is
> maybe an inappropriate use of the test?
This is dubious, for the usual reasons. The danger is that
experimenting until you find what you want tells us ... what you where
looking for in the first place, and not what is actually there.
In this case, experimenting with instruments until you fail to reject
exogeneity can be driven simply by eventually finding instruments that
generate the same bias in the coefficient estimated by ivprobit as
appears in a straight probit. It's like the null of a Hausman test.
Under the null, the two estimates are both consistent, and you reject
the null if the efficient but possibly inconsistent estimator is
different from the assumed consistent but inefficient estimator. But if
in reality, the two estimates are inconsistent in the same way, i.e.,
they suffer from the same bias, the test will not reject the null.
> It seems the current literature recommends close inspection
> of the first-step regression to identify valid instruments
> (testing exlusion restrictions, Shea's partial r-squared etc.)
You can identify "relevant" instruments from the first-step regression,
but to identify "valid" - i.e., exogenous - instruments, you need an
overid test.
--Mark
> Sincerely,
> Ivar Pettersen
> Phd-student
> Norwegian University of Science and Technology
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of
> Schaffer, Mark E
> Sent: Tuesday, April 18, 2006 6:16 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: st: RE: Interpretation of Smith-Blundell exog. test
>
> Ivar,
>
> > -----Original Message-----
> > From: owner-statalist@hsphsun2.harvard.edu
> > [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Ivar
> > Pettersen
> > Sent: 18 April 2006 02:53
> > To: statalist@hsphsun2.harvard.edu
> > Subject: st: Interpretation of Smith-Blundell exog. test
> >
> > Hello.
> >
> > I've got a Q related to the interpretation of the
> Smith-Blundell test.
> > It is provided both by *ivprobit* and *probexog*.
> > The help file for probexog states "A rejection indicates that the
> > standard probit (tobit) estimator should not be employed."
> > So if rejected one should use an IV-technique.
> >
> > When the test is run it is necessary to provide some
> instruments. The
> > question is: would not a rejection also mean that the
> instruments you
> > have provided are not good enough in an IV-estimation?
>
> In fact, for the test to be valid, you have to assume that
> the instruments *are* good enough for an IV estimation. The
> test is a test of the exogeneity of the specified regressor.
> If you reject, it means you need to treat it as endogenous
> and therefore instrument; if you fail to reject, it means you
> can consider treating it as exogenous.
>
> Maybe you are thinking of a test of overidentifying restrictions.
> That's a test of the exogeneity of the instruments; a
> rejection there means the instruments are not valid.
>
> HTH.
>
> --Mark
>
> > Sincerely
> > Ivar Pettersen
> >
> > *
> > * 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/
>
>
*
* 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/