Statalist


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

Re: st: multiple imputation and instrumental variables and several tests


From   Austin Nichols <austinnichols@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: multiple imputation and instrumental variables and several tests
Date   Fri, 18 Dec 2009 08:17:38 -0500

Matthijs De Zwaan <m.dezwaan@gmail.com> :

Note the conspicuous absence of IV at http://stata.com/help.cgi?mi_estimation

It seems to me that one has to be more than usually careful imputing
in an IV setting.
Which variables are you imputing? Excluded instruments? Included
instruments? Both?
What variables are you conditioning on when you impute?
Is there literature on imputing in an IV setting that you are consulting?
Can you send the citations?

If the sample loss arises primarily from missingness in
excluded instruments and the endogenous variable in different obs,
you could also try a two-sample IV and compare results.

I don't think the tests and diagnostics are available under MI because
no one has written a paper characterizing their large-sample and
small-sample performance; this could be a useful by-product
of your search, if you ran simulations showing how well your
chosen approach works under various conditions.

On Fri, Dec 18, 2009 at 5:36 AM, Matthijs De Zwaan <m.dezwaan@gmail.com> wrote:
> Dear Statalist, Maarten,
> Thanks for your advice. Excuse me for not specifying where I got
> -miest- from. I had forgotten that it's not downloadable from SSC.
> Instead, I downloaded it from Gary King's Amelia website.  Thanks also
> for pointing me to -mim-. I was unaware of the command.
>
> As for the tests and diagnostics: I guess I'll have to make do with
> what is available, and hope reviewers won't bother too much.
>
> Matthijs

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index