Yes, and just to add a note on that…perfect prediction can be a particular problem in the
propensity score context since it implies that the treatment and control groups are
completely different from one another on at least some combination of the covariates. In
other words, this is a potential violation of the assumption that everyone has a positive
probability of getting each treatment, and basically means that some combination of the
covariates is perfectly (or nearly perfectly) confounded with the treatment. So drawing
causal inferences may be difficult in this situation, and you don't necessarily want
to just discard variables from your model to avoid this, since then you are just turning
those variables into unobserved (rather than observed) confounders.
Liz
On Jul 9, 2013, at 8:34 PM, Kosuke Imai <kimai(a)princeton.edu> wrote:
This means that the covariates are perfectly (or
nearly perfectly) predicting the outcome. Perhaps, it's the sign of overfitting
(e.g., too many covariates).
Kosuke
Department of Politics
Princeton University
http://imai.princeton.edu
On Jul 9, 2013, at 8:59 AM, "Dancer, Anthony"
<anthony.dancer09(a)imperial.ac.uk> wrote:
Hi,
Hoping someone might be able to help me with the above warning message.
I have a dataframe with 7 covariates. 6 out of the 7 run without any issues, but any
matching method run involving the 7th returns the following message;
"Warning message: glm.fit: fitted probabilities numerically 0 or 1 occurred"
I've googled the above and despite some fairly detailed explanations I'm none
the wiser as to how to progress.
The variable is numerical. Sample as follows:
"[1] 42.108184 1.925875 42.108184 3.540687 2.331532 2.331532 2.781410 2.781410
1.701947 1.701947
[11] 1.701947 1.701947 1.701947 1.966572 1.966572 10.994199 0.000000 0.748349
3.666423 3.666423
[21] 3.666423 3.666423 0.746897 2.046640 1.019349 0.371999 1.019349 0.208058
0.194368 0.194368
[31] 0.952280 0.952280 0.952280 1.202849 1.202849 1.361906 1.361906 1.694844
0.618513 1.361906
[41] 1.361906 1.361906 0.304016 0.304016 0.641256 0.641256 0.641256 0.234019
0.641256 0.641256"
Any suggestions would be very much appreciated.
Cheers,
Anthony
Anthony Dancer
Imperial College London
Silwood Park Campus
Buckhurst Road
Ascot SL5 7PY
t: +44(0) 7968 836 451
e: anthony.dancer09(a)imperial.ac.uk
s: anthony.dancer1
-
---
MatchIt mailing list served by HUIT
List Address: matchit(a)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(a)lists.gking.harvard.edu
To unsubscribe from this list or get other information:
https://lists.gking.harvard.edu/mailman/listinfo/matchit
-
---
MatchIt mailing list served by HUIT
List Address: matchit(a)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(a)lists.gking.harvard.edu
To unsubscribe from this list or get other information:
https://lists.gking.harvard.edu/mailman/listinfo/matchit