Kelvin,
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of Kelvin Tan
> Sent: 03 November 2009 02:20
> To: [email protected]
> Subject: st: RE: re: overidentification test and endogeneity
> test with -ivreg2-
>
> Thanks Kit for your helpful insight about the endog test.
>
> One more question,
>
> Kit said: "You get different results for the overid test and
> endog test in (2) because you're using a different VCE. If
> you use the same VCE they agree exactly."
>
> I am sorry for the typo. I meant to have clustered standard
> errors "cl(idcode)" in both -ivregress- and -ivreg2- command
> lines. I get different results for the overid test and endog
> test when I used clustered standard errors in both
> -ivregress- and -ivreg2-.
Kit and I are looking at this now. We've worked out what's going on
with the overid test. Curiously, -estat- after -ivregress- seems to
report a standard (heteroskedasticity)-robust overid test even when the
equation is estimated as cluster-robust. Compare the following:
************************************************************
webuse nlswork, clear
qui ivregress 2sls wks_ue ( tenure = hours c_city union) grade, robust
estat overid
qui ivregress 2sls wks_ue ( tenure = hours c_city union) grade, robust
cl(idcode)
estat overid
qui ivreg2 wks_ue ( tenure = hours c_city union) grade, robust
di e(j)
qui ivreg2 wks_ue ( tenure = hours c_city union) grade, robust
cl(idcode)
di e(j)
************************************************************
You'll see that after -ivregress,robust-, -estat- reports the same
overid stat whether or not cluster(idcode) is also specified (I checked
Stata 10.1 and Stata 11.0). This is the same overid stat that -ivreg2-
reports with robust. -ivreg2-, however, reports a cluster-robust overid
stat that is (naturally) different from the
robust-but-not-cluster-robust stat.
Still looking at the endog option.
--Mark
>
> **************************************************************
> **********
> *************
> webuse nlswork, clear
> ivregress 2sls wks_ue ( tenure = hours c_city union) grade,
> cl(idcode) estat overid estat endog
> ivreg2 wks_ue ( tenure = hours c_city union) grade, endog(tenure)
> cl(idcode)
> **************************************************************
> **********
> **************
>
> ***************************************************************
> Results from -ivregress-
> . estat overid
> Test of overidentifying restrictions:
> Score chi2(2) = 4.32413 (p = 0.1151)
>
> . estat endog
> Tests of endogeneity
> Ho: variables are exogenous
> Robust regression F(1,4007) = .195661 (p = 0.6583)
> (Adjusted for 4008 clusters in idcode)
> ****************************************************************
>
> **************************************************************
> **********
> *****
> Results from -ivreg2-
> --------------------------------------------------------------
> ----------
> ------
> Hansen J statistic (overidentification test of all instruments):
> 3.519
> Chi-sq(2) P-val =
> 0.1722
> -endog- option:
> Endogeneity test of endogenous regressors:
> 0.567
> Chi-sq(1) P-val =
> 0.4513
> Regressors tested: tenure
> --------------------------------------------------------------
> ----------
> ------
> **************************************************************
> **********
> ******
>
> Thanks
>
> Regards,
> Kelvin
>
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of Kit Baum
> Sent: Tuesday, 3 November 2009 11:58 AM
> To: [email protected]
> Subject: st: re: overidentification test and endogeneity test with
> -ivreg2-
>
> <>
>
> You get different results for the overid test and endog test
> in (2) because you're using a different VCE. If you use the
> same VCE they agree exactly.
>
> webuse nlswork, clear
> ivregress 2sls wks_ue ( tenure = hours c_city union) grade
> estat overid estat endog
> ivreg2 wks_ue ( tenure = hours c_city union) grade, endog(tenure)
>
> Re (1), ivreg2 without the gmm2s option (that is, 2SLS) bases
> its overid test on the Hansen J statistic from the two-step
> GMM VCE. The endog test is the difference between two Hansen
> J statistics. Thus it does not matter, for the overid test
> and the endog test, whether you estimate the equation with
> 2SLS or GMM. The coefficient vector and its VCE change, of
> course, but the tests are based on the optimal two-step VCE
> in either case.
>
> 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/
>
> *
> * 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/
>
--
Heriot-Watt University is a Scottish charity
registered under charity number SC000278.
*
* 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/
