Prof. Chan:
If M is an instrument for X that doesn't appear in your second stage, I
would suggest using M*Y as an instrument for X*Y. Note that M*Y is NOT
a linear function of M and Y, which is why this is possible to do.
You would first generate the variables XY and MY and then do this in
ivreg:
ivreg Z (X XY = M MY) Y
All the std. errors would be automatically calculated properly.
If you still want to instrument for X with just M (excluding MY), and
then substitute Xhat*Y for XY in the second stage, you would have to
correct standard errors "manually" (i.e. ivreg wouldn't help you here).
One procedure for doing this is bootstrapping. See the stata manual.
Mushfiq
A. Mushfiq Mobarak
Assistant Professor of Economics
University of Colorado at Boulder
303-492-8872
Date: Tue, 26 Nov 2002 14:38:02 -0500
From: "Sewin Chan" <[email protected]>
Subject: st: ivreg with an interaction
Hello,
I am trying to estimate the following model using instrumental
variables. The final stage has two independent variables of interest,
plus an interaction between them:
Z = aX + bY + c(X*Y) + more variables
I want to instrument variable X in a first stage regression using a set
of instruments that contain Y and some other variables that are not in
the final stage. Of course the variable X*Y will have to reflect this
instrumenting as well.
How can I use ivreg to do this? I can't figure out how I can use the
first
stage to predict X, and then use it in the final stage both by itself
and as an interaction.
If ivreg can't do this, I know I can still get coefficients by
estimating 2 separate regressions in sequence using the predicted values
of X in the final stage, but how do I correct the standard errors? To
make things more complicated, I also want to use robust and cluster( ).
Thank you for any suggestions.
Sewin Chan
Assistant Professor of Public Policy
New York University
*
* 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/
