You might try using rpart matching.
Quoting Dan Hopkins <dhopkins(a)fas.harvard.edu>du>:
Hi Abby,
I can only speak to the second question. Here, it sounds like a very good
case to match exactly on some covariates and then within some boundary for
age. Both MatchIt and Matching allow you to do this, although I'm not
sure of the syntax off the top of my head.
Best,
Dan
On Fri, 28 Apr 2006, Abby Williamson wrote:
Hi All,
I've got a few matching questions.
1) Does anyone know how to add the "addlvariables" option into the matching
summary command? The default is = NULL and I didn't know how to make it
NOT
null. I didn't see an example in the
MatchIt manual.
2) I was also hoping to hear people's reactions to a curious result in our
efforts to find the best balance. First, we did exact matching on age,
race, educ, and sex. With age in the equation we lose about half of our
observations, but if we take it out, we get this:
#Sample sizes:
# Control Treated
#All 1292 1386
#Matched 1264 1361
#Discarded 28 25
Not bad, but of course, we'd like to have age in the equation, so we
decided
to try nearest neighbor matching. The results of
the "balance
improvements"
outputs suggest that this technique worsened the
balance along almost every
dimension. Any idea why this would be? Is it worth trying other matching
techniques (any particular suggestions)?
Percent Balance Improvement:
Mean Diff. eQQ Med eQQ Mean eQQ Max
distance -27.07 -29.32 -27.494 -173.671
age -19.20 0.00 -15.233 0.000
race -37.39 0.00 -36.585 0.000
educ -24.52 -100.00 -23.777 0.000
sex 64.63 0.00 66.667 0.000
distancexdistance -23.54 -18.54 -23.905 -2.871
distancexage -26.02 -46.51 -26.941 -10.801
distancexrace -24.09 -24.54 -25.474 -17.362
distancexeduc -53.80 -40.59 -52.541 -57.572
distancexsex -29.55 -24.65 -30.105 -36.215
agexage -134.28 -24.84 -14.673 -0.193
agexrace -41.12 -20.00 -39.032 -50.000
agexeduc - 21.88 -15.00 -21.094 -25.000
agexsex 42.65 0.00 -6.838 -16.667
racexrace -40.56 0.00 -39.726 0.000
racexeduc -31.69 -50.00 -30.853 -128.571
racexsex - 33.12 0.00 -31.752 0.000
educxeduc -28.55 -33.33 -27.919 -46.053
educxsex -15.49 0.00 -14.887 0.000
sexxsex 64.63 0.00 66.667 0.000
3) Finally, I don't know if anyone else has encountered this, but if I
include the (discard= "hull.both") command in nearest neighbor matching on
a
dataset about 3000 observations and 24 variables,
a desktop computer has
insufficient memory to do the calculation (even with the max memory command
in R turned on).
m.nearest <- matchit(wave1 ~ age+race+educ+sex, data=GSS8504,
method="nearest", discard= "hull.both")
Many thanks for any suggestions!
Best,
Abby
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l