|Date: ||Thu, 9 Jan 2003 17:24:37 -0500|
|Reply-To: ||Victor <praszz@YAHOO.COM>|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|From: ||Victor <praszz@YAHOO.COM>|
|Subject: ||Marketing mix seperation issues...|
This is not exactly a SAS issue but more of a Marketing/Statistics domain
Test/Case and Control Matching:
Background: I work at a retail chain and
I am working on analyzing promotional programs. We run some marketing
programs at certain stores(called test group) and then want to test the
performance of that test against a 'control' group.
Frequently we need to pick some other retail stores to use as a
control group to validate a promotion/test program, which
ran at a few 'test' stores.
The difficulty I am facing is "How to identify a control store(s) and what
are the techniques available?". And how to match a control to a test. I
understand the theory of using clustering on matching variables and model
specification is painstakingly slow. I am interested to know other ways
and generate some creative ideas of looking at the test-control issue.
I have looked at the 'match' algorithm posted on this list, using NETFLOW
proc. But that is more for a individual case-control used may be for
epidemilogy. I am interested to match a group of stores. I am looking it
from a statistical perspective of finding a statistical measure to
report "How close is a particular control to the test group".
I am currently working on experimental design and am planning to use
Clustering as a technique to match test and control stores. Can anyone
throw some thoughts/feedback?