Re: st: RE: IV with incomplete data on endog var

[Roger Harbord <roger.harbord@bristol.ac.uk>]

Many thanks, Dann and Scott. I had a quick look at the Angrist paper in 
JBES and it's clearly related but i'd need to work through all their 
algebra to work out exactly how my case differs, so Dann's idea of 
bootstrapping sounds attractive at the moment...

I'd be a little surprised if no-one's come across my situation before 
though - in some situations it seems to me that as long as you have enough 
info on the endogeneous variable to avoid problems of weak instruments then 
its the instrument-outcome association that mainly determines the power 
(particularly for dichotomous outcomes) and hence collecting data on the 
instrument and outcome only for most individuals becomes attractive (esp. 
if the endogenous variable is the most difficult/expensive to collect...) 
Of course need to be sure those with data on the endogenous variable are 
representative of those without.

Roger.

--On 20 July 2005 15:14 -0500 smerryman@kc.rr.com wrote:


> I believe the paper is:
>
> Angrist, Joshua D. and Alan B. Krueger (1995). "Split-Sample
> Instrumental Variables
> Estimates of the Returns to Schooling." Journal of Business and
> Economic Statistics 13:2, 225-235.
>
> The working paper version can be obtained at:
>
> http://www.irs.princeton.edu/pubs/pdfs/320.pdf
>
>
> Scott
>
> ----- Original Message -----
> From: "Millimet, Daniel" <millimet@mail.smu.edu>
> Date: Wednesday, July 20, 2005 11:40 am
> Subject: st: RE: IV with incomplete data on endog var
>
> > it seems that a two-step procedure where you form out of sample
> > predictions of the endogenous var and then do the 2nd stage with
> > the predicted regressor would yield valid estimates and that a
> > bootstrap procedure would yield valid inference.  this sounds
> > related to some papers by angrist on a procedure called split-
> > sample IV.  i think the paper i am thinking of was published in
> > journal of applied econometrics.  if you can't track it down, feel
> > free to email me and i can look for it.
> >
> > dann
*
*   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/