Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: -ivreg2- with an interaction again


From   Jen Zhen <[email protected]>
To   [email protected]
Subject   st: -ivreg2- with an interaction again
Date   Fri, 13 Dec 2013 14:10:45 +0100

Dear list members,

in an earlier post I described an issue I had with an -ivreg2-
Regression, where I wanted to use an interaction to test whether the
causal effect differed between 2 subsamples.

In following earlier posts on this list I did this as follows:

--ivreg2 outcome controls ind (endogreg endogreg_ind = z z_ind)--

where ind is the indicator for the subsample (0 for about half, 1 for
the other half of the sample).

Besides the desired 2nd stage output, this yielded 2 first stage
regressions for respectively endogreg and endogreg_ind, each of which
is in its first stage regressed on both instruments, the indicator and
the set of further controls.

The issue I had was that I got R2=0.99 on the second first stage
regression. There were ultimately no replies to my post, but meanwhile
I figured out that this is because for about half the sample the
outcome of the second FS regression is
endogreg_ind=endogreg*ind=endogreg*0=0 and for all those also  z_ind=
z*ind= z*0=0 and, more importantly, ind=0.

This results in a huge T Stat for ind and therefore in a huhe R2.

Now I'm wondering whether given the above reason this is legitimate or
whether it means I should omit the indicator from first and second
stage, or (probably not) do only the affected second stage separately?

Thanks so much and kind regards,
JZ
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index