I'm trying to use cem to do matching involving 3 treatments. Prof. King
(http://comments.gmane.org/gmane.comp.lang.r.matchit/393) indicated that
this can be done.
When I tried it, I got an error message -
imbalance(): 3258 <tmp>[1,1] found where void required
<istmt>: - function returned error
There was a discussion on this site on July 24 of 2013 about a similar
issue. In that case, there were only 2 treatments but a treatment of ".",
along with "1" and "2", showed up in the summary table. The suggestion was
to add "if indepvote != ." to the cem call."
While I'm sure that there are no missing values of my treatment variable, I
still tried to add the if clause to my cem call. But it didn't help. I got
another error:
. cem margins1 age, tr(approach1) if approach1 != .
option if not allowed
r(198);
Any ideas of how I can proceed?
Thanks.
Dennis Hanseman
Just trying you again. (Matt & Ariel am trying your emails too)
Any idea why I would be getting such a high % of missing.
In my dataset I have 30,473 cases where "cem_matched" ==1 (25,090 control,
5,383 treated). Of those 14,245 have a "cem_strata" ==. (11,624 control,
2,621 treated).
FYI, here is an output of the weights (cem1w) - treatment is pv, control is
non-pv
Thanks for any help you might be able to offer.
Ben
Ben Hoen
LBNL
Office: 845-758-1896
Cell: 718-812-7589
From: Ben Hoen [mailto:bhoen@lbl.gov]
Sent: Tuesday, September 09, 2014 3:43 PM
To: cem(a)lists.gking.harvard.edu
Subject: cem_matched ==1 and cem_strata==. <missing>
Hi all,
I had been using a cem matching output to run regressions and have just now
found that a large set of the output has the variable "cem_matched" ==1
while the "cem_strata" ==. (a.k.a. missing). For those cases, there is also
a weight stored in "cem_weights".
Is this a common occurance? If so, would you be able to explain when/why
this occurs?
Ben
Ben Hoen
Staff Research Associate
Lawrence Berkeley National Laboratory
Office: 845-758-1896
Cell: 718-812-7589
bhoen(a)lbl.gov <mailto:bhoen@lbl.gov>
<http://emp.lbl.gov/staff/ben-hoen> http://emp.lbl.gov/staff/ben-hoen
Visit our publications at:
<http://emp.lbl.gov/reports/re> http://emp.lbl.gov/reports/re
Sign up for our email list to receive publication notifications at:
<http://visitor.r20.constantcontact.com/manage/optin?v=001kdAkoVU6ITzX4UcDVi
5Hi_AUSvFPTtBN-DZDpyKukOGnrLar2tHOPf4yX2rmlhKRWIxNFc_i_L87BbQfcvFC-EJdikSpk4
uHs_O1_1Cnlt4%3D> LBNL Report Notification List
Hi,
I believe (based on previous correspondence) that the weights that CEM produces should be considered aweights (when used in regressions). This has worked fine because I often use either areg or reg, both of which accept aweights.
But now I find I need to rely on qreg (quartile regression) that does not accommodate aweights. Could you imagine a way to convert the CEM output weights (which sum to 1 within each strata) to pweights, that areg accepts? If so, would you suggest a conversion formula?
Thanks, in advance,
Ben
Ben Hoen
LBNL
Office: 845-758-1896
Cell: 718-812-7589
Hi all,
I had been using a cem matching output to run regressions and have just now
found that a large set of the output has the variable "cem_matched" ==1
while the "cem_strata" ==. (a.k.a. missing). For those cases, there is also
a weight stored in "cem_weights".
Is this a common occurance? If so, would you be able to explain when/why
this occurs?
Ben
Ben Hoen
Staff Research Associate
Lawrence Berkeley National Laboratory
Office: 845-758-1896
Cell: 718-812-7589
bhoen(a)lbl.gov
<http://emp.lbl.gov/staff/ben-hoen> http://emp.lbl.gov/staff/ben-hoen
Visit our publications at:
<http://emp.lbl.gov/reports/re> http://emp.lbl.gov/reports/re
Sign up for our email list to receive publication notifications at:
<http://visitor.r20.constantcontact.com/manage/optin?v=001kdAkoVU6ITzX4UcDVi
5Hi_AUSvFPTtBN-DZDpyKukOGnrLar2tHOPf4yX2rmlhKRWIxNFc_i_L87BbQfcvFC-EJdikSpk4
uHs_O1_1Cnlt4%3D> LBNL Report Notification List