Erin,

Maybe others have different thoughts but I am deeply skeptical that
you could ever have multiple valid instruments for any question in
political science.  Its a very specific set of assumptions.  Further,
when you have multiple IV's you have to assume they don't interact (I
think).  So the assumptions become further complicated.  In general it
seems like you would want to focus on in defending one instrument
really well rather than spreading yourself thin.

If you are in some crazy world where multiple instruments could work,
I think showing that its robust across instruments in separate
regressions is compelling.  These estimators are very sensitive to the
choice of instrument typically.

The Zelig option comes from the underlying package on which the
implementation is based (systemfit).  This is just by way of saying
that you can never be sure of the reason an option is there!  It might
only be a good idea for other fields.

That being said, this is just my take.  I am sure you can find people
with a less grumpy worldview on the topic :)

Brandon

On Tue, Apr 26, 2011 at 5:48 PM, Erin Baggott <ebaggott@fas.harvard.edu> wrote:
> Hi everyone,
>
> Here is a question about instrumental variables.  It might be useful
> for everyone.  Say you get lucky and come up with a bunch of valid
> instruments.  When running your IV regressions, should you use the
> instruments separately, or should you throw them all in one
> regression?
>
> I am not sure which approach is more theoretically justifiable.  I've
> seen the first reported in economics papers frequently, but the Zelig
> option where you can stick in a bunch of instruments at once calls out
> at me.
>
> Best,
> Erin
> _______________________________________________
> gov2001-l mailing list
> gov2001-l@lists.fas.harvard.edu
> http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
>
_______________________________________________
gov2001-l mailing list
gov2001-l@lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l