which Stata weight to use? - Matchit - lists.gking.harvard.edu

E. Michael Foster

13 Nov 13 Nov

7:14 p.m.

Well, I don't know what -clarify- does, but to me it seems like the bigger question is how to use the weights. (Most likely, the weights are a feature of your data, and you have -pweights-.) At the matching stage, I would use the weight as a covariate for calculating the propensity score. If your weights are a function of covariates already in the model, I wouldn't expect much of an impact. At the analysis stage, you might think of using weights when calculating the mean treatment difference by pooling across strata or levels of the propensity score. I can't see where the weights for the comparison cases would play any role, at least for ATT. If you are actually analyzing the matched data using something like a ranksum test, I'm not sure how I'd go about using the weights. I would use the weight for the treatment case, at least for ATT (since the weight for the two cases wouldn't necessarily be the same one). But I'm not sure what I would do. (I've started doing analyses on a pair basis b/c it lends itself to the types of sensitivity analyses in the Rosenbaum book. But I've not thought through what to do with the weights in that case.) I'd be interested in hearing what others would do-- michael Casey Klofstad wrote:

...

I have matched data via full matching, and I want to analyze the data in Stata. I know I need to use a regression weight becuase I have subclasses, but which one is the appropriate one? Specifically, Stata offers three potential choices: pweights, aweights, and iweights For the record, I get the same results with "a" and "i". The "p" weight slightly alters the results (becuase it switches over to robust standard errors). Also, I setx/estsimp in Clarify will not take "i" and "p" weight estimates. In these cases Clarify will estimate and aggregate the model results, but it will not calculate predicted probabilities or expected values. Clarify has no problem with "a" weights. Thanks!

-- E. Michael Foster getmike.info download vcard: www dot unc dot edu/~emfoster/Foster.vcf Professor of Maternal and Child Health of Health Policy and Administration of Biostatistics (adjunct) Associate Chair for Faculty Advancement (W) 919-966-3773 (F) 919-966-0458 Office: 407C Rosenau Hall << use for FEDEX School of Public Health University of North Carolina, Chapel Hill Rosenau Hall, Campus Box# 7445 Chapel Hill, NC 27599-7445 [the http is throwing off some spam filters) visit the new FAQ section of my blog: fosterfaq dot blogspot dot com 27) [Mollie Ivins] "What do you mean cancer is a gift. You're as a dumb as Shrub. Cancer just killed me. Some gift." "I'm not quite dead yet, sir!" -- Concorde, Monty Python and the Holy Grail -- MatchIt mailing list served by Harvard-MIT Data Center List Address: matchit(a)lists.gking.harvard.edu Subscribe/Unsubscribe: http://lists.gking.harvard.edu/?info=matchit MatchIt Software and Documentation: http://gking.harvard.edu/matchit/

Reply

E. Michael Foster

8:31 p.m.

I read too much into the question--I thought Casey was asking about complex survey data --sorry. I read into his question something I've been thinking about for another project. I'll read up on the procedure before I comment next time-- Onur BaEer wrote:

...

I would use aweights. Onur Baser University of Michigan On Nov 13, 2007, at 5:58 PM, Casey Klofstad wrote:

I have matched data via full matching, and I want to analyze the data in Stata. I know I need to use a regression weight becuase I have subclasses, but which one is the appropriate one? Specifically, Stata offers three potential choices: pweights, aweights, and iweights For the record, I get the same results with "a" and "i". The "p" weight slightly alters the results (becuase it switches over to robust standard errors). Also, I setx/estsimp in Clarify will not take "i" and "p" weight estimates. In these cases Clarify will estimate and aggregate the model results, but it will not calculate predicted probabilities or expected values. Clarify has no problem with "a" weights. Thanks! -- Casey A. Klofstad University of Miami Department of Political Science Coral Gables, FL klofstad(a)gmail.com http://moya.bus.miami.edu/~cklofstad -- MatchIt mailing list served by Harvard-MIT Data Center List Address: matchit(a)lists.gking.harvard.edu Subscribe/Unsubscribe: http://lists.gking.harvard.edu/?info=matchit MatchIt Software and Documentation: http://gking.harvard.edu/matchit/

-- MatchIt mailing list served by Harvard-MIT Data Center List Address: matchit(a)lists.gking.harvard.edu Subscribe/Unsubscribe: http://lists.gking.harvard.edu/?info=matchit MatchIt Software and Documentation: http://gking.harvard.edu/matchit/

-- E. Michael Foster getmike.info download vcard: www dot unc dot edu/~emfoster/Foster.vcf Professor of Maternal and Child Health of Health Policy and Administration of Biostatistics (adjunct) Associate Chair for Faculty Advancement (W) 919-966-3773 (F) 919-966-0458 Office: 407C Rosenau Hall << use for FEDEX School of Public Health University of North Carolina, Chapel Hill Rosenau Hall, Campus Box# 7445 Chapel Hill, NC 27599-7445 [the http is throwing off some spam filters) visit the new FAQ section of my blog: fosterfaq dot blogspot dot com 27) [Mollie Ivins] "What do you mean cancer is a gift. You're as a dumb as Shrub. Cancer just killed me. Some gift." "I'm not quite dead yet, sir!" -- Concorde, Monty Python and the Holy Grail -- MatchIt mailing list served by Harvard-MIT Data Center List Address: matchit(a)lists.gking.harvard.edu Subscribe/Unsubscribe: http://lists.gking.harvard.edu/?info=matchit MatchIt Software and Documentation: http://gking.harvard.edu/matchit/

Reply

Kosuke Imai

15 Nov 15 Nov

7:56 a.m.

Actually, I think you want to use probability weights rather than frequncy weights since each obs actually does not appear more than once in your sample. Kosuke On Wed, 14 Nov 2007, Kosuke Imai wrote:

...

You want to use frequency or probability weights [depending on whether you focus on the insample or population quantity, respectively]. In the terminology of our PA article, you want those two empirical distributions to be the same. In the case of full matching, you need to weight each observation appropriately (using the weights that come out of MatchIt) in order to get any benefit of matching. Kosuke On Tue, 13 Nov 2007, E. Michael Foster wrote:

I read too much into the question--I thought Casey was asking about complex survey data --sorry. I read into his question something I've been thinking about for another project. I'll read up on the procedure before I comment next time-- Onur BaEer wrote:

I would use aweights. Onur Baser University of Michigan On Nov 13, 2007, at 5:58 PM, Casey Klofstad wrote:

I have matched data via full matching, and I want to analyze the data in Stata. I know I need to use a regression weight becuase I have subclasses, but which one is the appropriate one? Specifically, Stata offers three potential choices: pweights, aweights, and iweights For the record, I get the same results with "a" and "i". The "p" weight slightly alters the results (becuase it switches over to robust standard errors). Also, I setx/estsimp in Clarify will not take "i" and "p" weight estimates. In these cases Clarify will estimate and aggregate the model results, but it will not calculate predicted probabilities or expected values. Clarify has no problem with "a" weights. Thanks! -- Casey A. Klofstad University of Miami Department of Political Science Coral Gables, FL klofstad(a)gmail.com http://moya.bus.miami.edu/~cklofstad -- MatchIt mailing list served by Harvard-MIT Data Center List Address: matchit(a)lists.gking.harvard.edu Subscribe/Unsubscribe: http://lists.gking.harvard.edu/?info=matchit MatchIt Software and Documentation: http://gking.harvard.edu/matchit/

-- MatchIt mailing list served by Harvard-MIT Data Center List Address: matchit(a)lists.gking.harvard.edu Subscribe/Unsubscribe: http://lists.gking.harvard.edu/?info=matchit MatchIt Software and Documentation: http://gking.harvard.edu/matchit/

Reply