Yes, I agree with Gary. Whenever you are dropping some controls you are essentially
always estimating the ATE. (That's a bit of an oversimplification but not too far
off).
But you could estimate the ATE using subclassification. (But right, not using the
MatchIt-generated weights). You would estimate the effect separately within each
subclass and then weight the subclass-specific effects by the proportion of people in that
subclass (in the whole population). I think my 2010 Statistical Science paper talks more
about that. And I think Zelig could implement that but I'm not sure.
Liz
On 2/1/12 11:42 AM, "Gary King" <king(a)harvard.edu> wrote:
With matching, its easier to think about estimating an ATT. Then you only need to ensure
that you keep all the treated units, and whatever controls are useful as matches.
as a separate point, i'd drop treated units when there are no sufficiently close
controls. that changes the quantity of interest, which is fine so long as you clarify
what you're estimating.
Gary
--
Gary King - Albert J. Weatherhead III University Professor - Director, IQSS - Harvard
University
GKing.Harvard.edu <http://gking.harvard.edu/> - King(a)Harvard.edu
<mailto:King@Harvard.edu> - @kinggary <http://twitter.com/kinggary> -
617-500-7570 - Asst 495-9271 - Fax 812-8581
On Wed, Feb 1, 2012 at 11:38 AM, Francois Maurice <maurice.francois(a)ymail.com>
wrote:
Hi,
I'm using MatchIt. I'm trying to understand which method produce ATE estimate
and which one produce ATT estimate.
In the documentation, section 5: Frequently Asked Questions: How Exactly are the Weights
Created?, it is said :
"These weights are constructed to estimate the average treatment effect on the
treated, [...]".
Is there a way with MatchIt to estimate ATE ? To be concrete, I'm using experimental
data with a control group almost three times the treated group. I'm using the
following four methods with ratio=2 in matchit():
Nearest : Drop some controls
Subclassification : Keep all controls
Nearest with exact : Drop some controls
Genetic : Drop some contrls
Since subclassification keeps all controls, can that be an ATE estimate or do I need to
built my own weights to make sure it is ATE?
In general, is MatchIt produced only match set with weights that can only be use to
estimate ATT ?
And if I use Zelig after MatchIt, is there a way to produce ATE estimate ?
Thanks,
François Maurice
-
---
MatchIt mailing list served by HUIT
List Address: matchit(a)lists.gking.harvard.edu
Subscribe/Unsubscribe:
http://lists.gking.harvard.edu/mailman/listinfo/ei
MatchIt Software and Documentation:
http://gking.harvard.edu/matchit/
Browse/Search <http://gking.harvard.edu/matchit/Browse/Search> List Archive:
http://lists.gking.harvard.edu/mailman/private/matchit/
Matchit mailing list
Matchit(a)lists.gking.harvard.edu
https://lists.gking.harvard.edu/mailman/listinfo/matchit