under one of those models you specific, you're estimating a single
constant treatment effect. so weights will have no effect on the mean
(bias), only on efficiency.
weights in the sense you're talking about are important for descriptive
quantities, like means or variances, but not for causal effects unless
you allow for variation in those effects over the observations. if you
do the latter somehow and have an estimate of it for each observation
(or group), you could then use the weights to produce an overall
number. but the weight is not necessary for estimation (unless you
think it will help with the variance).
Gary
---
http://gking.harvard.edu
On 10/17/2009 9:40 PM, Sinan Gemici wrote:
Hi All,
Thanks a lot for MatchIt, it's helping me out a great deal in my dissertation
research. I'm analyzing a large-scale labor market dataset (NLSY97) and am still not
sure about how to use survey weights when matching.
Prof. King, in a previous posting you stated: "I'd ignore the sampling weights
for estimating causal effects. As long as you condition (in this case match) on the survey
weights or (better) the variables that compose the survey weights, you're not going
to have any bias."
Just to make sure I understand: does this mean that if I follow your advice (i.e. include
the NLSY97 survey weight as a covariate for matching) then the results of any regular
post-matching parametric analysis (ANOVA, log-linear, etc.) will be representative of the
population?
Any guidance/clarification would be much appreciated.
Sinan
Sinan Gemici, PhD Candidate
Department of Workforce Education
University of Georgia
--
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/
Browse/Search List Archive:
http://lists.hmdc.harvard.edu/lists/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/
Browse/Search List Archive:
http://lists.hmdc.harvard.edu/lists/matchit/