Date:         Mon, 3 Jul 2006 06:16:57 -0700
Reply-To:     Andreww <andrew.whittam@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Andreww <andrew.whittam@GMAIL.COM>
Organization: http://groups.google.com
Subject:      Re: Finding control stores
Comments: To: sas-l@uga.edu
Content-Type: text/plain; charset="iso-8859-1"

Hi Hari - There probably is some way of doing it using cluster
analysis, but if all you need is 30 stores you can probably do it by
looking through your list for 30 other similar stores based on as near
as you can to the crieteria you have... or as reasonably near as you
can get. You could put the list into xl and flag stores on size/region
etc and "cluster" them that way.

Mapping them is always a good way to check the basic robustness of the
output.

If you think about it, it is probably going to be impossible to find a
store that is the same in each and every respect. But something like
the above should get you a good classification solution.

I have been involved in doing this sort of thing for Tesco, IKEA, and
others and what usually matters are factors like sales,  size in sq
meters, that it's folloing a similar trading pattern - not brand new or
in an odd place, similar region, urban/not, if you can add demographics
similarity then that will help but probably best to keep it top level -
poor urban, wealthy ethnic etc.

I think the approach you take does depend on the amount of time you
want to put into this.


Any help?

Andrew


Hari wrote:
> Hi,
>
> I have a particular promotion campaign which was run (April/May) in
> lets say 30 stores (test stores). I want to compare the efficacy of
> this campaign by comparing the results from 30 other similar stores
> (control stores) in which this campaign was not run assuming that all
> other conditions were same between these 2 sets of stores.
>
> Presently I need to identify the set of 30 control stores which most
> closely match the set of 30 test stores. To start with I have data on
> 1000 stores (data current as of March of this year), like their last
> year's revenue, share of the competitors in the region in which the
> store is there, average distance of the competitor from that store and
> demand for the product (in terms of Index) in the immediate vicinity of
> the store and lastly, I also have data on a categorical variable -
> RegionType- (6 levels) which is classifying the region in which the
> store is there to "Modest working town", "Affluent suburban spread"
> etc.
>
> How do I find the set of control stores using above information.
>
> My boss said that one can use non-objective segmentation to make
> clusters of stores and from that select control stores from a cluster
> in which a test store falls. I consulted Proc Cluster and in that went
> through Example 23.6: Size, Shape, and Correlation which also had a
> categorical variable (box color) and was thinking that probably in my
> problem I could also use similar dummy assignments to RegionType
> variable. But, wouldnt the RegionType variable in my problem has some
> sort of ordering (ordinal variable). Is it meaninful to somehow use
> this information for doing segmentation (if yes how).
>
> Also, I would like to know some good tutorials for segmentation
> (especially dealing with ?a scenario similar to what I have above).
>
> Morever, are there any other methods also to approach the problem of
> finding control stores
>
> Please guide me.
>
> Regards,
> HP
> India