Hi all,
Thanks to everyone who shared their experiences with multiple imputation
and matching. Below you can find a summary of the responses I recieved.
Similar to matching with complex survey designs, there seems to be
relatively little consensus in the literature
about the
best practices.
Cyrus Samii and colleagues first created multiply imputed data sets and
then carried out matching on each completed data set. They also provide
detailed and efficient R code they used in their JCR paper:
http://m.jcr.sagepub.com/content/57/4/598.
Here are a couple of papers that systematically explore different
approaches using simulations:
Mitra, R., & Reiter, J. P. (2012). A comparison of two methods of
estimating propensity scores after multiple imputation.
*Statistical Methods in Medical Research : *
https://stat.duke.edu/~jerry/Papers/smimr13.pdf
.
Qu, Y., & Lipkovich, I. (2009). Propensity score estimation with missing
values using a multiple imputation missingness pattern (MIMP) approach.
*Statistics in Medicine*, *28*(9), 1402-1414.
Bernard and colleagues use principal stratification to recover missing
data in a randomized experiment. I am not sure, though, to what extent this
framework can be extended to an observational study:
http://biosun01.biostat.jhsph.edu/~cfrangak/papers/sc/vouchers.pdf
.
Here is some code I received that shows how to go about using MatchIt and
Amelia in combination:
http://www.forrestlane.com/uploads/ECLS_PSM_syntax.txt
.
Not
about
multiple imputation, but
the
approach Paul Rosenbaum suggests
in his 2002 book
is to impute m
issing data on
the covariates
with their
mean
value
,
create
a dummy variable
indicating
missing cases
on each imputed covariate, and use the two variables, along with other
predictors, in estimating the propensity score.
Here's another paper that does not use multiple imputation but that is
useful in dealing with missing data in propensity score matching: D'Agostino
Jr, R. B., & Rubin, D. B. (2000). Estimating and using propensity scores
with partially missing data. *Journal of the American Statistical
Association*, *95*(451), 749-759.
Thanks once again to all who shared these resources.
Best,
Valerio
**********************************************************
Political Methodology E-Mail List
Editors: Ethan Porter <[log in to unmask]>
Gregory Whitfield <[log in to unmask]>
**********************************************************
Send messages to [log in to unmask]
To join the list, cancel your subscription, or modify
your subscription settings visit:
http://polmeth.wustl.edu/polmeth.php
**********************************************************
|