Hi William,
I guess I’m not super surprised by this given that you only have 11 treatment schools. So
I think it’s probably just very sensitive to the specific variables, and while your
intuition is right I think there’s probably just some strange things happening because of
your tiny treatment group sample size.
So in this example seems like you should match on school size since then you get better
balance on that and on the other covariates.
Liz
On May 2, 2017, at 10:50 AM, William Dudley
<wndudley@uncg.edu<mailto:wndudley@uncg.edu>> wrote:
I am using MatchIT nearest and mahalanobis distance to match 11 treatment school with a
pool of candidate schools with a 1:4 match.
My question :
If I exclude school size from the match, my Std Differences in the matched sample are
higher on a few covariates
than if I leave in Size. I would think that fewer covariates should lead to better
matching on the remaining covariates.
I provide the syntax and output for match with and without Size.
The first analysis provides a good match (All ST Diff < .25) match BUT WHen I drop
SIZE as a covariate
some STD DIffs rise to > .25 .
I appologise for th elenghty output but I wanted to be sure that I provided all
information
Thanks
Bill
x2.out <- matchit(tx ~ schlvl + collcred +
colltake +
+ drop + PovPct + racePct + size +
+ Znat + Zstate,
+ data = MI317, method = "nearest",
+ exact = c("schlvl"), ratio=4,
Distance ="mahalanobis")
x2.data <- match.data(x2.out)
summary(x2.out, standardize=TRUE)
Call:
matchit(formula = tx ~ schlvl + collcred + colltake + drop +
PovPct + racePct + size + Znat + Zstate, data = MI317, method = "nearest",
exact = c("schlvl"), ratio = 4, Distance = "mahalanobis")
Summary of balance for all data:
Means Treated Means Control SD Control Std. Mean Diff. eCDF Med eCDF Mean eCDF
Max
distance 0.0378 0.0213 0.0209 0.6086 0.2245 0.2219
0.4085
schlvl 0.8182 0.7581 0.4287 0.1486 0.0301 0.0301
0.0601
collcred 0.3208 0.4780 0.6604 -0.3593 0.0794 0.0792
0.1565
colltake 0.1534 0.2005 0.1830 -0.2834 0.0867 0.0903
0.2227
drop 0.0178 0.0149 0.0229 0.1530 0.0602 0.0764
0.2412
PovPct 0.4701 0.4663 0.2008 0.0261 0.0523 0.0634
0.1853
racePct 0.2985 0.2208 0.2946 0.2439 0.0654 0.0802
0.2366
size 622.0909 704.9556 489.4964 -0.2255 0.0504 0.0581
0.1569
Znat -0.3471 0.0094 0.9994 -0.4011 0.0806 0.0845
0.2110
Zstate -0.2346 0.0072 1.0008 -0.2900 0.0927 0.0979
0.2700
Summary of balance for matched data:
Means Treated Means Control SD Control Std. Mean Diff. eCDF Med eCDF Mean eCDF
Max
distance 0.0378 0.0382 0.0270 -0.0127 0.0227 0.0376
0.1136
schlvl 0.8182 0.8182 0.3902 0.0000 0.0000 0.0000
0.0000
collcred 0.3208 0.3932 0.6002 -0.1656 0.0455 0.0597
0.1364
colltake 0.1534 0.1537 0.1405 -0.0024 0.0455 0.0622
0.1818
drop 0.0178 0.0165 0.0236 0.0674 0.0455 0.0838
0.2955
PovPct 0.4701 0.4376 0.1763 0.2193 0.0909 0.0860
0.2045
racePct 0.2985 0.2274 0.3155 0.2231 0.0682 0.0818
0.2045
size 622.0909 548.7045 374.6600 0.1997 0.0909 0.1070
0.3182
Znat -0.3471 -0.2540 0.9097 -0.1047 0.0455 0.0553
0.1591
Zstate -0.2346 -0.1521 0.9579 -0.0989 0.0455 0.0628
0.1818
Percent Balance Improvement:
Std. Mean Diff. eCDF Med eCDF Mean eCDF Max
distance 97.9195 89.8776 83.0534 72.1848
schlvl 100.0000 100.0000 100.0000 100.0000
collcred 53.9154 42.7252 24.5279 12.8806
colltake 99.1641 47.5687 31.0922 18.3539
drop 55.9415 24.5053 -9.6510 -22.4924
PovPct -739.5158 -73.7303 -35.4728 -10.3858
racePct 8.5286 -4.2017 -2.0508 13.5554
size 11.4384 -80.3636 -84.1409 -102.8037
Znat 73.8822 43.6364 34.5807 24.5873
Zstate 65.8771 50.9881 35.8666 32.6544
Sample sizes:
Control Treated
All 496 11
Matched 44 11
Unmatched 452 0
Discarded 0 0
*************************************
Matchit without school size NOTE racepct and CollTake get worse
Call:
matchit(formula = tx ~ schlvl + collcred + colltake + drop +
PovPct + racePct + Znat + Zstate, data = MI317, method = "nearest",
exact = c("schlvl"), ratio = 4, Distance = "mahalanobis")
Summary of balance for all data:
Means Treated Means Control SD Control Std. Mean Diff. eCDF Med eCDF Mean eCDF
Max
distance 0.0378 0.0213 0.0208 0.6133 0.2183 0.2186
0.3990
schlvl 0.8182 0.7581 0.4287 0.1486 0.0301 0.0301
0.0601
collcred 0.3208 0.4780 0.6604 -0.3593 0.0794 0.0792
0.1565
colltake 0.1534 0.2005 0.1830 -0.2834 0.0867 0.0903
0.2227
drop 0.0178 0.0149 0.0229 0.1530 0.0602 0.0764
0.2412
PovPct 0.4701 0.4663 0.2008 0.0261 0.0523 0.0634
0.1853
racePct 0.2985 0.2208 0.2946 0.2439 0.0654 0.0802
0.2366
Znat -0.3471 0.0094 0.9994 -0.4011 0.0806 0.0845
0.2110
Zstate -0.2346 0.0072 1.0008 -0.2900 0.0927 0.0979
0.2700
Summary of balance for matched data:
Means Treated Means Control SD Control Std. Mean Diff. eCDF Med eCDF Mean eCDF
Max
distance 0.0378 0.0371 0.0247 0.0265 0.0455 0.0488
0.1364
schlvl 0.8182 0.8182 0.3902 0.0000 0.0000 0.0000
0.0000
collcred 0.3208 0.2884 0.3777 0.0740 0.0455 0.0506
0.1591
colltake 0.1534 0.1282 0.1294 0.1513 0.0682 0.0750
0.1591
drop 0.0178 0.0185 0.0269 -0.0370 0.0455 0.0554
0.1818
PovPct 0.4701 0.4500 0.1593 0.1358 0.0682 0.0694
0.1818
racePct 0.2985 0.2023 0.2746 0.3017 0.0682 0.0868
0.2500
Znat -0.3471 -0.2889 0.7923 -0.0655 0.0455 0.0559
0.1818
Zstate -0.2346 -0.1612 0.7852 -0.0880 0.0455 0.0591
0.1591
Percent Balance Improvement:
Std. Mean Diff. eCDF Med eCDF Mean eCDF Max
distance 95.6857 79.1772 77.6937 65.8245
schlvl 100.0000 100.0000 100.0000 100.0000
collcred 79.4135 42.7252 36.0128 -1.6393
colltake 46.6206 21.3531 16.9131 28.5597
drop 75.8229 24.5053 27.5071 24.6201
PovPct -419.9426 -30.2977 -9.4203 1.8793
racePct -23.7074 -4.2017 -8.2357 -5.6545
Znat 83.6764 43.6364 33.8539 13.8141
Zstate 69.6610 50.9881 39.6640 41.0726
Sample sizes:
Control Treated
All 496 11
Matched 44 11
Unmatched 452 0
Discarded 0 0
--
William N. Dudley, PhD
Professor - Public Health Education
The School of Health and Human Sciences
The University of North Carolina at Greensboro
437-L Coleman Building
Greensboro, NC 27402-6170
Visit my Web
Site<http://www.uncg.edu/phe/faculty/dudley.html>
VOICE 336.256 2475<tel:(336)%20256-2475>
-
---
MatchIt mailing list served by HUIT
List Address:
matchit@lists.gking.harvard.edu<mailto:matchit@lists.gking.harvard.edu>
Subscribe/Unsubscribe:
http://lists.gking.harvard.edu/mailman/listinfo/ei
MatchIt Software and Documentation:
http://gking.harvard.edu/matchit/
Browse/Search List Archive:
http://lists.gking.harvard.edu/mailman/private/matchit/
Matchit mailing list
Matchit@lists.gking.harvard.edu<mailto:Matchit@lists.gking.harvard.edu>
To unsubscribe from this list or get other information:
https://lists.gking.harvard.edu/mailman/listinfo/matchit