Date: Tue, 27 Jul 2010 16:02:18 -0400
Reply-To: William Shakespeare <shakespeare_1040@HOTMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: William Shakespeare <shakespeare_1040@HOTMAIL.COM>
Subject: Logistic regression predictors
Suppose I have a list of customers and their sales going back several
years. The list is ever expanding since new customers are always being
added and some customers are repeat purchasers.
Now suppose I have a hunch that a 2010 Q2 purchase can be predicted by Q2
purchases in previous years. I'm only interested in predicting whether or
not someone bought, not the amount, so my outcome is binary while my
predictors are continuous.
The question I have concerns the Q2 purchases in previous years. It's
possible that person A could have multiple purchases, say 2005 Q2, 2008
Q2, and 2009 Q2 as well at the target quarter (2010 Q2). Do the repeated
predictors represent a violation of the independence assumption if they
contain the same subjects? Clearly if I were looking at repeated outcome
measures it would, but here I have repeated predictors. Collinearity
could also be a problem that's dealt with easily enough.
I seem to remember this question being posted a while back but I don't kow
what the answer was and I can't find it.