st: IV with incomplete data on endog var

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

Calling all IV experts:

I have an application of instrumental variables where the endogenous 
regressor is only measured on a subsample of the data, but the excluded 
instruments and the outcome are measured on the full sample. If i use 
-ivreg- or -ivreg2- then (I assume) i will be using the subsample only. It 
appears to me that  in principle under two-stage least-squares i could do 
the first-stage regression on the subsample but generate predictions for 
the full sample and hence make use of the full sample for the second stage. 
But I'm less clear how to get a valid variance estimate, or if there's a 
way of doing this within -ivreg- or -ivreg2-.

In the simplest case I have no exogeneous regressors (included 
instruments), though I may well wish to add some at a later stage. I'd have 
(almost) complete data on these for the full sample.

Many thanks for any thoughts (or pointers to any existing literature on 
this)

Roger.

--
Roger Harbord
Department of Social Medicine, University of Bristol
http://www.epi.bris.ac.uk

*
*   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/