Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: st: RE: Display of Shea's partial R^2 in ivreg2 output


From   "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: RE: Display of Shea's partial R^2 in ivreg2 output
Date   Sun, 28 Feb 2010 19:31:08 -0000

Tom,

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Tom Palmer
> Sent: 28 February 2010 18:04
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: RE: Display of Shea's partial R^2 in ivreg2 output
> 
> Thank you very much Mark.
> 
> I only tend to use a single endogenous variable. I ran a few 
> examples and the AP R^2 seems to be the same as the Shea R^2 
> in this case.

That's right, they are numerically equivalent in the case of a single
endogenous variable.

> I would still be interested in seeing the AP R^2 in the 
> output with the first/ffirst options.

Well, there's only so much space in the output.  Of the various
possibilities - 1st stage F, 1st stage R-sq, Shea R-sq, AP F (=weak ID
test statistic), AP R-sq, AP chi-sq (=underidentification test
statistic) - we went for first-stage F and the two formal test stats.
Angrist and Pischke don't actually recommend (or even discuss) the
partial R-sq corresponding  to their F test stat.  Personally, I think
the F and chi-sq are more useful than the R-sq since they come with p-
or critical values.  But there's no accounting for taste.  If you really
want the AP R2, you can get it in the saved e(first) matrix.

Cheers,
Mark

> best regards,
>   Tom
> 
> 
> On Fri, February 26, 2010 8:04 pm, Schaffer, Mark E wrote:
> > Tom,
> >
> > The answer to this is interesting because it turns out that Shea's 
> > partial R-sq is very closely related to the R-sq that is a 
> byproduct 
> > of calculating the Angrist-Pischke first-stage F-stats.
> >
> > Specifically, here is how you calculate the AP partial R-sq vs. the 
> > Shea partial R-sq:
> >
> > A-P R-sq:
> >
> > 1.  Get the fitted values of the uninteresting endogenous 
> regressors.
> > 2.  Partial (1) out of the endogenous regressor of interest.
> > 3.  Partial (1) out of the excluded instruments.
> > 4.  Regress (2) on (3).
> > 5.  R-sq is AP R-sq.
> >
> > Shea partial R-sq
> >
> > 1.  Get the fitted values of the uninteresting endogenous 
> regressors.
> > 2.  Partial the ACTUAL uninteresting endogenous vars out of the 
> > endogenous regressor of interest.
> > 3.  Partial (1) out of the excluded instruments.
> > 4.  Regress (2) on (3).
> > 5.  R-sq is Shea R-sq.
> >
> > Below is a demonstration with the toy auto dataset.
> >
> > Because the A-P approach has a sounder theoretical 
> foundation than the 
> > Shea approach, we decided to drop the Shea partial R-sq 
> from the main 
> > output in favour of the A-P stats.
> >
> > Cheers,
> > Mark
> >
> > ******************************
> >
> > sysuse auto, clear
> >
> > * AP stat
> >
> > * 1.  Get the fitted values of the uninteresting endogenous 
> regressors.
> > * 2.  Partial (1) out of the endogenous regressor of interest.
> > * 3.  Partial (1) out of the excluded instruments.
> > * 4.  Regress (2) on (3).
> > * 5.  R-sq is AP R-sq.
> >
> > qui reg weight turn trunk length displacement capture 
> predict double 
> > weight_hat, xb
> >
> > qui reg mpg weight_hat
> > capture predict double mpg_what_resid, resid
> >
> > qui reg turn weight_hat
> > capture predict turn_what_resid, resid qui reg trunk weight_hat 
> > capture predict trunk_what_resid, resid qui reg length weight_hat 
> > capture predict length_what_resid, resid
> >
> > reg mpg_what_resid turn_what_resid trunk_what_resid 
> length_what_resid, 
> > nocons dof(69) di e(r2)
> >
> > * Shea partial R-sq
> >
> > * 1.  Get the fitted values of the uninteresting endogenous 
> regressors.
> > * 2.  Partial the ACTUAL uninteresting endogenous vars out of the 
> > endogenous regressor of interest.
> > * 3.  Partial (1) out of the excluded instruments.
> > * 4.  Regress (2) on (3).
> > * 5.  R-sq is Shea R-sq.
> >
> > qui reg mpg weight
> > capture predict double mpg_wactual_resid, resid
> >
> > reg mpg_wactual_resid turn_what_resid trunk_what_resid 
> > length_what_resid, nocons dof(69) di e(r2)
> >
> > * Check Shea vs AP partial R-sqs from ivreg2
> >
> > qui ivreg2 price (mpg weight = turn trunk length 
> displacement), ffirst 
> > mat first=e(first) mat APr2=first["APr2",1] mat 
> > sheapr2=first["sheapr2",1] mat list APr2 mat list sheapr2
> >
> > ******************************
> >
> >> -----Original Message-----
> >> From: owner-statalist@hsphsun2.harvard.edu
> >> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of 
> Tom Palmer
> >> Sent: 26 February 2010 19:48
> >> To: statalist@hsphsun2.harvard.edu
> >> Subject: st: Display of Shea's partial R^2 in ivreg2 output
> >>
> >> Dear Statalist,
> >>
> >> I put the following query to the authors of -ivreg2-. I'm 
> posting it 
> >> here so the answer can be shared on the list.
> >>
> >> In previous versions of -ivreg2- when specifying the -first- or 
> >> -ffirst- options Shea's partial R^2 for the instruments 
> was displayed 
> >> in the output like the following:
> >>
> >> Summary results for first-stage regressions
> >> -------------------------------------------
> >>
> >> Variable    | Shea Partial R2 |   Partial R2    |  F(  1,
> >> 4792)    P-value
> >> zlnfat9     |     0.0082      |     0.0082      |       39.83
> >>       0.0000
> >>
> >>
> >> I'm not sure when the change was made but in the latest version 
> >> (3.0.01) the partial R^2 is no longer shown with these options. (I 
> >> know it is still available in the returned
> >> -e(first)- matrix.)
> >>
> >> I find the partial R^2 very useful in comparing the strength of 
> >> different instruments so I was wondering if it would it be 
> possible 
> >> reintroduce the display the partial R^2 in the output of future 
> >> versions? (I accept typing "mat list e(first)" after 
> -ivreg2- is not 
> >> too onerous!)
> >>
> >> I also take this opportunity to thank you for an extraordinarily 
> >> comprehensive command.
> >>
> >> best regards,
> >>
> >>   Tom
> >>
> >>
> >> *
> >> *   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/
> 


-- 
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/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index