5 Feb
2013
5 Feb
'13
11:42 a.m.
Hi Evan,
First, in Stata, does the "cem" command work the same way as the
"imb"
command in calculating the multivariate L1 distance?
Is there an option
under the "breaks()" argument that I can specify where I can be sure it is
using the user-specified cutoff points?
The "cem" command uses the user-defined cut points to match, but it uses
"scott" or the algorithm given in "imbbreaks()" to perform the
imbalance
checking.
Second, I'm not sure if I understand your point about using different
cutoff points to measure imbalance than the ones used in the matching.
I thought that one of the benefits of CEM was that the user could up front
declare how much imbalance they are willing to tolerate by coarsening the
variables as he/she saw fit. Assuming the researcher is coarsening the
variables with sound reasons, why not use those same cutoff points to
measure imbalance?
Because CEM explicitly minimizes the imbalance for a given set of cut
points, so that comparing across definitions of the cut points would be
meaningless. Thus, instead, we tend to fix the imbalance measure (that is,
fix the coarsening for the L1) and then try multiple user-defined cut
points. But, in general, we don't want to evaluate our matching solution
with something defined by the matching solution. See page 7 of the CEM
paper:
http://gking.harvard.edu/files/gking/files/political_analysis-2011-iacus-pa…
Cheers,
matt.
One of the examples that you and your colleagues have
used is coarsening
a 'years of education' variable.
If the researcher coarsens that variable in meaningful ways for matching
(0-8; 9-12; 12-16; >16), and then measures imbalance using the scott or
sturges break method where the cutpoints are changed to non-meaningful
cutpoints (5.5; 7.5; 9.5; 11.5; etc), I'm unclear as to how that calculated
L1 distance would be more valid than calculating the L1 distance with the
user-set cutoff points.
Any clarification would be greatly appreciated and thanks for your time.
Evan
On Tue, Feb 5, 2013 at 8:53 AM, Matt Blackwell <m.blackwell(a)rochester.edu>wrote;wrote:
Hi Evan,
There's currently no way to set variable-specific cut points for the
"imb" command. You can only change the algorithm for all variables via the
"breaks()" argument. For the imb command, it's best to use a set of cut
points other than the ones you use for the matching and keep that cut point
fixed while you attempt different matching solutions.
Hope that helps!
Cheers,
matt.
On Fri, Feb 1, 2013 at 1:13 PM, Evan Cole <cole.s.evan(a)gmail.com> wrote:
CEM Users,
When using Stata and the imb command, is their anyway to change it so
that it recognizes user-defined cutpoints to calculate the L1 distance
instead of defaulting to the scott method?
I have tried running the imb command after a cem command with user
defined cutpoints but it still uses the scott method.
Any help would be greatly appreciated. Thanks for the time.
Evan Cole
Tulane University
-
--
cem Mailing List, served by HUIT
Send messages: cem(a)lists.gking.harvard.edu
[un]subscribe Options: http://lists.gking.harvard.edu/?info=cem
More information on cem: http://gking.harvard.edu/cem
Cem mailing list
Cem(a)lists.gking.harvard.edu
To unsubscribe from this list or get other information:
https://lists.gking.harvard.edu/mailman/listinfo/cem