|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
st: RE: IV testing with ivreg2
Not sure why this didn't get through the first time, so I'm trying
again....
-----Original Message-----
From: Schaffer, Mark E
Sent: Wednesday, January 07, 2009 10:49 PM
To: [email protected]
Subject: RE: st: RE: IV testing with ivreg2
Johanna,
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of
> Johanna Avato
> Sent: 07 January 2009 22:09
> To: [email protected]
> Subject: Re: st: RE: IV testing with ivreg2
>
> Thanks Mark! but how about my K-P value being above the critical
> values when I estimate using liml?
This suggests you can reject the null that the LIML estimates are biased
by weak instruments (at your chosen level of bias/significance). Good
news for you.
> What about the significant
> Anderson-
> Rubin stat? Doesn't that improve the situation and could at
> least say
> that I am at the edge of having a weak IV?
Not exactly. The A-R approach is different from the traditional IV/LIML
point estimation approach. If you want to go down the A-R route with 2
endogenous regressors, you're looking at constructing a confidence area
for the coefficients (not interval, area, because 2 coefficients are
involved). Can be done with a grid search but kind of laborious. See
Dufour (2003) in the ivreg2 help file.
Cheers,
Mark
> I read the paper, though it didn't really answer my particular
> combination of test results...
>
> On Jan 7, 2009, at 3:25 PM, Schaffer, Mark E wrote:
>
> > Johanna,
> >
> >> -----Original Message-----
> >> From: [email protected]
> >> [mailto:[email protected]] On Behalf Of
> >> Johanna Avato
> >> Sent: 07 January 2009 20:05
> >> To: [email protected]
> >> Subject: st: IV testing with ivreg2
> >>
> >> I am running an IV model with 2 endogenous variables and 4 IVs with
> >> present heteroskedasticity. In my tests the partial Fs are
> >> well above
> >> 10 but the Kleinbergen-Paap statistic is only 5 thus not above the
> >> critical values, still the Anderson-Rubin statistic is
> >> significant. Do
> >> I have a weak IV problem or not? If I use the liml option, which I
> >> read is recommended for weak instruments, the
> >> Kleinbergen-Paap of 5 is
> >> still above the critical values... does anyone have a comment?
> >
> > It looks like you have a weak IV problem, though I have to add the
> > caveat that the Kleibergen-Paap statistic is only a
> generalization of
> > the Cragg-Donald stat and so the critical values for the C-D stat
> > should
> > be taken as rough guide only for the K-P stat.
> >
> > The reason large partial Fs aren't enough is because your
> instruments
> > don't have enough explanatory power to explain both your endogenous
> > regressors at the same time. They can explain one or the
> other (hence
> > the two large partial Fs), but not both. If you check out
> the paper
> > by
> > Kit Baum, Steve Stillman and myself in the Stata Journal
> (2003) you'll
> > find a slightly longer discussion of this.
> >
> > Hope this helps.
> >
> > Cheers,
> > Mark
> >
> > Prof. Mark Schaffer
> > Director, CERT
> > Department of Economics
> > School of Management & Languages
> > Heriot-Watt University
> > Edinburgh EH14 4AS
> > tel +44-131-451-3494 / fax +44-131-451-3296
> > http://ideas.repec.org/e/psc51.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/
> >>
> >
> >
> > --
> > 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/