see below.
On Tue, Apr 14, 2009 at 8:43 PM, John Polley
<jpolley at jd10.law.harvard.edu>wrote:
I understand the matchit() function (when using
"nearest") to already be
calculating a propensity score from the covariates, using the logit function
to do so by default, and then matching based on that propensity score.
yes.
If our call to matchit() (using "nearest")
were to plug in the propensity
scores themselves (rather than the 5 covariates), wouldn't we be asking
matchit() to calculate a propensity-score-of-propensity-score, and to match
based on that?
yes, if you leave everything else at the defaults. alternatively, you could
pass to the distance argument your vector of propensity scores.
As it turns out, it appears one reaches the same answer either way (at
least in the problem set example). But substantively, I'm not clear on why
we would want such a propensity-on-propensity approach. I have understood
the usefulness of propensity scores to be in finding close matches where,
with multiple covariates, we would have too few exact matches.
the point of the question is to make sure that everyone understands what
propensity score matching is all about.
Am I mistaking the terminology somewhere? There is material from Gary's
site that refers to the matchit() function (using the default, "nearest")
with a call including the covariates as "matching on the estimated
propensity score."
Or is it that you would like us to figure out a way to execute nearest
neighbor propensity score matching without using the "nearest" function in
MatchIt at all?
On Tue, Apr 14, 2009 at 7:29 PM, Miya Woolfalk <woolfalk at fas.harvard.edu>wrote:
You are on the right track in thinking that question 3 wants you to use
nearest neighbor propensity score matching rather than some other nearest
neighbor distance metric.
However, the formula in your call to matchit() should not contain the 5
covariates. We want you to match on the propensity scores and not the
covariates. In other words, we want you to estimate the propensity scores
using a model specification of your choice before using matchit() and use
these values to conduct the matching.
Miya
On Tue, Apr 14, 2009 at 6:01 PM, John Polley <
jpolley at jd10.law.harvard.edu> wrote:
In my understanding, the "nearest"
method in matchit() creates a
propensity score using the covariates, and matches on that propensity score
(the syntax of matchit() will look quite like what we were shown in Miya's
section notes). When we use matchit(), the default that it uses to determine
propensity scores is the logit model. The matchit() manual on gary's website
details some other options that one can use in matchit() in place of the
logit model, and are specified by <distance="nameofmodel">.
The italicized phrase in the problem set seems only to be telling us not
to use some other form of matching on the covariates without a propensity
score.
On Tue, Apr 14, 2009 at 5:10 PM, Schmidt, William <wschmidt at hbs.edu>wrote:
Despite an earlier email chain on a related
subject, the syntax to use
matchit() with propensity scores rather than covariates is still not clear
to me, if anyone can help. Do you set ?distance=? to the propensity score
vector? What do you put in the ?formula? argument? Thanks!
_______________________________________________
gov2001-l mailing list
gov2001-l at
mailman.fas.harvard.edu
http://mailman.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l at
mailman.fas.harvard.edu
http://mailman.fas.harvard.edu/mailman/listinfo/gov2001-l
--
Miya Woolfalk
Ph.D. Student
Harvard University
Government and Social Policy
_______________________________________________
gov2001-l mailing list
gov2001-l at
mailman.fas.harvard.edu
http://mailman.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l at
mailman.fas.harvard.edu
http://mailman.fas.harvard.edu/mailman/listinfo/gov2001-l
--
Miya Woolfalk
Ph.D. Student
Harvard University
Government and Social Policy