Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: xtivreg2, overidxt,


From   "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: xtivreg2, overidxt,
Date   Thu, 15 Mar 2007 15:12:20 -0000

Nicola et al., 

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of 
> nicola.baldini2@unibo.it
> Sent: 15 March 2007 11:17
> To: statalist@hsphsun2.harvard.edu
> Cc: ncdcta00@uniroma2.it
> Subject: Re: st: xtivreg2, overidxt,
> 
> May be you installed -ivreg2- from SJ rather than from ssc, 
> which usually provides the newest version. Solution:
> ssc install ivreg2, replace
> For the interpretation, it depends on what you think about 
> the p-value. If you usually reject the null hypothesis for 
> any p<5%, then everything is fine. I you reject the null 
> hypothesis also for any 10%<p<5%, then the test casts doubt 
> on the validity of the instruments. Please note that 
> -xtoverid- produces a similar test and may give slightly 
> different results (running the example at the end of the help 
> for -overidxt- yealds a p=.4138; re-running with -xtoverid- 
> yealds a p=.4411)

In case you're curious, the reason for the difference is that -xtoverid-
(and -xtivreg2-) correct for the loss of degrees of freedom because of
the fixed effects, which are nuisance parameters.  -overidxt- does not
make this correction, and as a result it is more likely (too likely) to
reject the null of valid instruments.

It's similar to the adjustment needed for the error variance when
constructing the var-cov matrix for the fixed effects estimator.  The
error variance is bigger after the adjustment (and hence p-values for
paramater tests are bigger); here, the overid statistic is smaller after
the adjustment (and hence the p-value is bigger).  Or, put another way,
the usual Sargan overid test is an N*R2 test.  In a balanced panel with
N fixed effects and T periods, it's an (NT-N)*R2 test, because in effect
N observations are "lost" because of the estimation of the fixed
effects.

Cheers,
Mark



> 
> Nicola
> 
> At 02.33 14/03/2007 -0400, ncdcta00@uniroma2.it wrote:
> >Dear statalist,
> >I have some problems to implement xtivreg2 because stata 
> give me this error:
> >Error - must have ivreg2 (or ivreg28, for Stata 8 users) version
> >2.1.15 or greater installed
> >r(601);
> >
> >I have update stata, I use stata 9 and I installed the 
> package iverg28 
> >and ivreg2.
> >
> >Other question is: how I interpret overidxt? when my p value 
> is 10% my 
> >instrument is correleted with my dipendent variable? or I refuse the 
> >null hypothesis  and so the instrument that I use is ok?
> >thanks so much for your help
> 
> *
> *   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/



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