Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# st: R: RE: Quick doubt about the 2-way clustering option used in IV_GMM2s

 From Jordana Rodrigues Cunha <[email protected]> To "[email protected]" <[email protected]> Subject st: R: RE: Quick doubt about the 2-way clustering option used in IV_GMM2s Date Mon, 10 Jan 2011 13:41:56 +0000

```I didn't explain right, sorry me.  The problem is the second stage regression.
I am using -ivreg2-for the GMM estimation and in the first-stage F-stat is not missing like this:

. ivreg2 y t p c d (x w = h l m n o ) if sampleCM==1, gmm2s endog (x w) cluster (id deal )

2-Step GMM estimation
---------------------
Estimates efficient for arbitrary heteroskedasticity and clustering on id and deal
Statistics robust to heteroskedasticity and clustering on id and deal
Number of clusters (id) =         2903                Number of obs =     8194
Number of clusters (deal) =     6512                F( 51,  2902) =     1.90
Prob > F      =   0.0001
Total (centered) SS     =   84.6613152             Centered R2   =  -0.3689
Total (uncentered) SS   =  85.85567541           Uncentered R2 =  -0.3499
Residual SS             =  115.8950279                Root MSE      =    .1189

All the tests show that my regressors are not endogenous and that my instruments are valid.
And so, I back to my OLS regression results that had produced good estimations (and unbiased too).
Due to 2-way clustering I am running the -cgmreg routine, that doesn't show any F-statistic in its output, like this:

.cgmreg y x w  if sampleCM==1,  cluster (id deal)

Note: +/- means the corresponding matrix is added/subtracted
Calculating cov part for variables:  id (+)
Calculating cov part for variables:  id deal (-)
Calculating cov part for variables:  deal (+)
Number of obs     =    8199
Num clusvars      =    2
Num combinations  =    3
If condition      =    if sampleCM==1
G(id)             =    2906
G(deal)           =    6516
------------------------------------------------------------------------------
y   |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
x   |   .0120439   .0050404     2.39   0.017     .0021648     .021923
w  |   .0016485   .0007922     2.08   0.037     .0000958    .0032012

And finally, when I ask for the stored statistics there is no F-statistic, but the r2, for example. And that was my point, Professor.

I hope you could help me, any comment from yours would be really appreciated,
Jordana

Jordana Rodrigues Cunha
PhD. Candidate
University of Bologna
Department of Management
Via Capo di Lucca, 34, 1st floor
40126 – Bologna, ITALY
Fixed line:  0039 (051) 20 98 073
Fax: 0039 (051) 20 98 074
[email protected]
www.sa.unibo.it

I am on a visiting scholar leave of absence until December/2009 at LMU-INNOtec (University of Munich/Germany).
For urgent matters, here are my contact info:

Fixed line : 0049 (0) 89 383 777 18
Private cell phone: 0039 (348) 0420 121
________________________________________
Inizio: [email protected] [[email protected]] per conto di Schaffer, Mark E [[email protected]]
Inviato: lunedì 10 gennaio 2011 13.16
Fine: [email protected]
Oggetto: st: RE: Quick doubt about the 2-way clustering option used in IV_GMM2s

Jordana,

> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of
> Jordana Rodrigues Cunha
> Sent: Monday, January 10, 2011 9:28 AM
> To: [email protected]
> Subject: st: Quick doubt about the 2-way clustering option
> used in IV_GMM2s
>
> Dear professors,
>
> I used the great GMM estimator with the 2-way clustering
> option in testing the endogeneity of my regressors and all
> the tests were excellent for my conclusions.
> But , when I am running the OLS regressions of the first
> stage, using the cgmreg routine to indicate my 2-way
> clustering, the F-statistic (explicitly required in the
> estout, scalars option) doesn't come out.

I'm guessing that you're using -ivreg2- for the GMM estimation - is that
right?  If so, you can get your first-stage regressions from the -first-
option, or you can use -ivreg2- to do the first-stage regression by hand
using OLS.  Is the first-stage F-stat still missing in that case?

If that doesn't work, maybe post an example to the list of the output to
illustrate what the problem is.

Cheers,
Mark

> I would like to understand if it happens because there is no
> sense in using the F-statistic because I am assuming that the
> s.e are non independent once I am indicating the 2-way
> clustering? How could I test the regression significance?
>
> Jordana
>
>
> Jordana Rodrigues Cunha
> PhD. Candidate
> University of Bologna
> Department of Management
> Via Capo di Lucca, 34, 1st floor
> 40126 - Bologna, ITALY
> Fixed line:  0039 (051) 20 98 073
> Fax: 0039 (051) 20 98 074
> [email protected]
> www.sa.unibo.it
> *
> *   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/
*
*   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/
```