Date:         Fri, 16 Feb 2001 12:12:00 -0800
Reply-To:     Tina Sanders <tina_s72@YAHOO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Tina Sanders <tina_s72@YAHOO.COM>
Subject:      Repeated measures analyses
Content-Type: text/plain; charset=us-ascii

Dear listserv members,

I am attempting to build a predictive model.  Let me
explain my data and problem.  The data consists of
several hundred thousand merchants who subscribe to a
particular service provided by a company.  Every month
a number of merchants cancel their service with the
company.  That, of course, is a loss of revenue.  My
goal is to create a model that will predict the
merchants who are most likely to cancel, so that the
company may take some proactive measures.  The company
has monthly data for each merchant that consists of
revenue, billing, call volume, rate for service, and
of course status (a=active, c=cancel), etc.  So, each
observation represents each month.  Some merchants
have been subscribed to the service for 2 months, some
12 months, etc.  Some are still active, some have
cancelled.

My problem is that I'm not sure about which
statistical analysis I should use.  Possibly, Cox
proportional hazards as I have staggered entry and
repeated measures (which the time-dependent option
would handle)?  And I could consider "status" as a
censored variable?  However, will that produce a
probability of cancellation per merchant?  Or should I
look at the Genmod procedure with the GEE?  It looks
like that model does not allow for unequal timepoints.


I have looked at the example in SAS help for both
those procedures, and to tell the truth, I can't quite
tell whether either one would be a good approach.

I would be very happy to hear any ideas or advice!

Thank you very much!
Tina Sanders


__________________________________________________
Do You Yahoo!?
Get personalized email addresses from Yahoo! Mail - only $35
a year!  http://personal.mail.yahoo.com/