Date: Fri, 4 Jul 2008 10:29:24 +0530
Reply-To: Praveen <ipkrishna@GMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Praveen <ipkrishna@GMAIL.COM>
Subject: Combining categories and estimating Odds Ratio
Content-Type: text/plain; charset=ISO-8859-1
I am faced with problem where I have the customer survey responses. I am
trying to build up an ordinal logistic regression with the 11-point ordered
categories in response variable. Upto this point it works out fine. Now I
want to compare few clubbed categories, i.e.,
New Category "A" = 1,2,3,4,5 combined from original 11-point responses
New Category "B"=6,7,8,9 combined from original 11-point responses
New Category "C"=10,11 combined from original 11-point responses
I did it using running three different logistic regression comparing (A,B),
(B,C) & (A,C) and got the estimates of odds ratio for each combination which
was making a lot of sense to me.
But other guys in my group said clubbing the categories is not a good idea
as it results in loss of information, I completely agree with them. But the
problem is with the technique. None of us know how to get to different
estimates of odds ratios for different combinations mentioned above. I know
we can compute individual probabilities on each observations using ordinal
logistic regression, but I am interested in odds ratio which are
different/relevant for clubbed categories.
It would really help me if someone can guide me through getting my 'desired'
result by using OLR on 11-point dependent variable.
All measurements are subject to variation.