Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: st: RE: Weak instrument tests in small panel data
From
"Schaffer, Mark E" <[email protected]>
To
"[email protected]" <[email protected]>
Subject
RE: st: RE: Weak instrument tests in small panel data
Date
Sat, 15 Feb 2014 19:21:21 +0000
Steven,
> -----Original Message-----
> From: [email protected] [mailto:owner-
> [email protected]] On Behalf Of Xiao, Chong
> Sent: 15 February 2014 15:43
> To: <[email protected]>
> Subject: Re: st: RE: Weak instrument tests in small panel data
>
> Mark,
>
> Thanks for your reply. Does it mean that for regressions with weak instruments,
> I should focus on those IV robust test results (Anderson-Rubin and other -
> weakiv- output) rather than the second stage estimate of standard errors in
> interpreting the effect of the endogenous variable?
It's up to you and your particular circumstances. Some people use one, some the other, some use both (i.e., you can report both traditional and weak-instrument-robust confidence intervals). But if you think the weakness of your instruments is a serious issue for your particular application, then reporting a weak-IV-robust confidence interval either as your main result or as a robustness check seems sensible to me.
> Also the Anderson-Rubin result from -weakiv- seems to differ from that
> generated in -ivreg2-. Why is that?
If they are testing the same null, and using the same VCE (in your case, cluster-robust), then they should be the same test statistic. For example, if I use the toy auto dataset, after
sysuse auto, clear
ivreg2 mpg (price=weight), ffirst
there is a block in the output that says
Weak-instrument-robust inference
Tests of joint significance of endogenous regressors B1 in main equation
Ho: B1=0 and orthogonality conditions are valid
Anderson-Rubin Wald test F(1,72)= 134.62 P-val=0.0000
Anderson-Rubin Wald test Chi-sq(1)= 138.36 P-val=0.0000
and if I follow this estimation with weakiv, there is a line in the output that says
Weak instrument robust tests and confidence sets for linear IV
H0: beta[mpg:price] = 0
-------------------------------------------------------------------------------------------------
Test | Statistic p-value | Conf. level Confidence Set
------+----------------------------------------------+-------------------------------------------
AR | chi2(1) = 138.36 Prob > chi2 = 0.0000 | 95% [ -.004628,-.0020795]
Same chi-sq test stat.
--Mark
>
> Steven
>
> On Feb 14, 2014, at 7:45 AM, "Schaffer, Mark E" <[email protected]>
> wrote:
>
> > Steven,
> >
> > My guess is that you have a model where the iid assumption fails, and so
> there is a big difference between the standard first-stage F statistic (Cragg-
> Donald) and a cluster-robust first-stage F statistic (Kleibergen-Paap). This is
> probably also why the partial R2 from the first stage also looks large.
> >
> > FYI the Anderson-Rubin test is not a test for weak instruments, but a weak-
> instrument-robust test of H0: beta=0 (where beta is the coefficient on the
> endogenous regressor). If you want to go down this route, you might have a
> look at -weakiv- by Magnusson-Finlay-Schaffer, available from SSC in the usual
> way.
> >
> > HTH,
> > Mark
> >
> >> -----Original Message-----
> >> From: [email protected] [mailto:owner-
> >> [email protected]] On Behalf Of Xiao, Chong
> >> Sent: 13 February 2014 20:40
> >> To: [email protected]
> >> Subject: st: Weak instrument tests in small panel data
> >>
> >> Dear statalist users:
> >>
> >> I have an unbalanced panel data that consists of about 40 groups in
> >> the cross section and 21 periods. I have two instruments and one
> >> endogenous variable that are all time-invariant.
> >> When I run the 2SLS regression, the Kleibergen-Paap F statistic is
> >> below the conventional critical value of 10, but the Shea Partial R2
> >> is more than 10% and the Anderson-Rubin tests also show up strong.
> >> The first stage estimates for the instruments are also significant. I
> >> suspect that it is the small sample that drives the low F statistic.
> >> But could anyone
> >> provide more detailed explanations to reconcile the discrepancy
> >> between the various weak instrument test results? Thanks!
> >>
> >> Number of clusters = 47 Number of obs = 876
> >> Anderson-Rubin Wald test F(2,46)= 6.27 P-val=0.0039
> >> Anderson-Rubin Wald test Chi-sq(2)= 12.98 P-val=0.0015
> >> Underidentification test (Kleibergen-Paap rk LM statistic): 10.191
> >> Chi-sq(2) P-val = 0.0061
> >> Weak identification test (Cragg-Donald Wald F statistic): 114.638
> >> (Kleibergen-Paap rk Wald F statistic): 8.952
> >>
> >> sheapr2 .20971431
> >>
> >> Best,
> >> Steven Xiao
> >>
> >>
> >> *
> >> * For searches and help try:
> >> * http://www.stata.com/help.cgi?search
> >> * http://www.stata.com/support/faqs/resources/statalist-faq/
> >> * http://www.ats.ucla.edu/stat/stata/
> >
> >
> > -----
> > Sunday Times Scottish University of the Year 2011-2013 Top in the UK
> > for student experience Fourth university in the UK and top in Scotland
> > (National Student Survey 2012)
> >
> >
> > We invite research leaders and ambitious early career researchers to
> > join us in leading and driving research in key inter-disciplinary themes.
> > Please see www.hw.ac.uk/researchleaders for further information and
> > how to apply.
> >
> > 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/faqs/resources/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/faqs/resources/statalist-faq/
> * http://www.ats.ucla.edu/stat/stata/
-----
Sunday Times Scottish University of the Year 2011-2013
Top in the UK for student experience
Fourth university in the UK and top in Scotland (National Student Survey 2012)
We invite research leaders and ambitious early career researchers to
join us in leading and driving research in key inter-disciplinary themes.
Please see www.hw.ac.uk/researchleaders for further information and how
to apply.
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/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/