st: obtaining correct significance level for the endogenous variable in ivprobit with a potentially weak instrument

["Danny Cohen-Zada" <danoran@bgu.ac.il>]

Suppose that my model is

y1= ao + a1*y2 + a2*x1 (y1 - is a dichotomous variable)
y2=b0+b1*x1+b2*z1 (excluded instrument0

(one instrument, one endogenous regresssor).


1) Am i correct that  if i have a reason to think that  z1 is a weak 
instrument (for example, based on AR test in ivreg2) i must bootsrap in 
order to obtain the correct standard errors of  a1?
If yes, does anybody know how should i do it.

2) Suppose that i want to depict the predicted probability as a function of 
x1 (when all other variables are at the mean) and i want to include the ci 
for this probability. Am i correct that if i think that z1 may be a weak 
instrument, i must use the boots option in prvalue? If yes, how do i do this 
after an ivprobit estimation? Can i do the following:
a) obtain the predicted probablity from the first stage (manually)
b)  run the second stage probit estimation
c)  run the prvalue commend with the boots option.

Thanks

Danny

----- Original Message ----- 
From: "Austin Nichols" <austinnichols@gmail.com>
To: <statalist@hsphsun2.harvard.edu>
Sent: Thursday, February 28, 2008 2:39 PM
Subject: Re: st: Critical values for stock and yogo test when the 
clusteroption is used



> No--hence I said:
> "
> You will find discussion of constructing Anderson Rubin confidence
> regions in http://www.stata.com/meeting/5nasug/wiv.pdf (and its refs).
> "
> rather than recommending -condivreg-.
> Did you try augmenting excluded instruments as I suggested?  Did you
> do an overID test?
>
> All of this advice was for the linear model, before you revealed you
> have a probit; for your setting you may need original simulations.
>
> On Thu, Feb 28, 2008 at 2:56 AM, Danny Cohen-Zada <danoran@bgu.ac.il> 
> wrote:
>
> > As you recommended me i came to conclusion that my instruments are weak.
> >  Therefore, i used condivreg to perform the conditinal likelihood ratio 
> > test.
> >  According to this test the pvalue is 0.000. However, this procedure 
> > doesn't
> >  have a cluster option. Is the conditional ratio test valid (when one 
> > have to
> >  cluster the standard errors)?
> >
> >  Best,
> >
> >  Danny
> *
> *   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/