re: st: Synth: Abadie, Diamond, and Hainmueller

["Ariel Linden, DrPH" <ariel.linden@gmail.com>]

First off, the primary rule of engagement on the statalist is that you
describe the program you are using and where you got it from. I happen to
know what you are referring to, but I?d bet that most others would not.

In this case, synth is a user-written program found at:
http://www.mit.edu/~jhainm/Synth

As for your question: You don?t need to do anything with the weights. The
program provides them for you to see how control groups were derived using
the list of covariates you supplied. It also shows how the covariates
balance.

As for the outcomes, the program provides that for you by generating
predicted values for the synthetic control group. 

One cannot do traditional statistical analysis when there is only one
treatment unit, so the authors suggest running the model iteratively,
switching out the treatment unit with a different control each iteration. If
the magnitude of the difference remains in favor of the original treatment
group, you can claim that there is a treatment effect.

I strongly suggest you read the reference paper:

Abadie, A., Diamond, A., and J. Hainmueller. 2010. Synthetic Control Methods
for Comparative Case Studies: Estimating the Effect of California's Tobacco
Control Program.  Journal of the American Statistical Association 105(490):
493-505.


Ariel

Date: Mon, 14 May 2012 18:45:48 -0400
From: Jessie C <jessiecoh@gmail.com>
Subject: st: Synth: Abadie, Diamond, and Hainmueller

After using the synth command to generate weights among potential
donors, how do you get an average treatment effect and do inference?

I understand that synth over a set of covariates or lagged dependent
variables indicates which potential controls are most appropriate and
the corresponding weight of how relevant.  I don't understand what to
do with these weights.  Do I proceed as I would have done otherwise
(e.g. OLS) with just the subsample of the treatment and the selected
donors?  (That doesn't seem right because that would ignore the
weights. . . )




*
*   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/