Hi Kit and Mark,
I have spent quite some time to double check the conditional moment
test and I believe I am doint it right. Here is what I get. The endog
option within xtivreg2 gives me a p-val of 0.3. Then I perform the
conditional moment test as follows:
Model: Lit = b0 + b1mi1 + b2mi2 + b3zit + eit, with m1 and m2 both
possibly endogenous and zit exogenous. Which I estimate with xtreg,
fe. Then, predict fe_resid, ue. Then,
gen var1= m1_it*w_hat_it
gen var2= m2_it*w_hat_it
gen var3= zit*w_hat_it
sureg (var1 var2 = w_hat_it var3 zit).
Finally,
test [var1]_cons [var2]_cons
The tes gives me a p-val=0.000, which I interpret as evidence that m1
and m2 are endogenous (while again the p-val from the endog is 0.3).
Can it be that this difference is due to the power of the two tests
and the cond. mom. test performs bettwer especially if ivs are weak?
In performing the cond. moment test I follow Greene, 2003, pp: 505-508
(especially last paragraph of sect. 17.6.4).
Any thoughts would be truly appreciated,
Erasmo
On 4/21/07, Kit Baum <[email protected]> wrote:
I would have to wonder whether the execution of the 'conditional
moment' test in the xtreg,fe setting was done correctly. If xtivreg2
cannot detect any evidence of endogeneity, then any asymptotically
equivalent test should yield qualitatively similar results, even if
one test is more powerful than the other in finite samples.
Kit Baum, Boston College Economics
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html
On Apr 21, 2007, at 2:33 AM, statalist-digest wrote:
> in a previous email you mentioned the possibility that the endogeneity
> test can be done within the FE context or the IV framework:
> Kit said:
> "I think what is being confused here is that one could do two Hausman
> tests in a FE context: one for FE vs RE (which does question the
> correlation between regressor and fixed effect) and another, in the
> pure FE context, for endogeneity of regressors. That is, the test of
> OLS vs IV for the LSDV model. The former Hausman test would look at
> the correlation between regressor (mangerial ownership) and the firm-
> specific effect, but if done purely in a OLS/LSDV framework, could be
> weakened if OLS is inappropriate in the first place. But I don't see
> how that critique would apply if the test was done in a IV framework,
> allowing for endogeneity of regressors."
>
> I obtained the conditional moment test within the FE suggested in the
> paper and found strong rejection for exogeneity (unlike what I find
> with the endog option in xtivreg2).
>
> If you have any thoughts on this, it would be truly appreciated.
*
* 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/
