Date:         Fri, 17 Jan 2003 16:16:56 +0000
Reply-To:     mbabyak@DUKE.EDU
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         mbabyak@DUKE.EDU
Organization: Duke University, Durham, NC, USA
Subject:      Re: How can I calculate the Odds Ratio for continuse variables in
              Logistic Regression!

Rodney Sparapani <rsparapa@mcw.edu> wrote:
: Bin Fan wrote:

:>I run a Logistic Regression in SAS to get the Odds Ratio for each variable,  someone told me that the Odds Ratio for dummy variable is O.K., but  we need to calculate the Odds Ratio for continuse variable by hand
:>
:>
: The ORs for continuous variables from SAS are fine.  You just have to
: remember that they are based on a
: change of +1 in the continuous variable.  So, some scaling may be
: warranted before analysis.  For example,
: Age can be analyzed as Age/10 so the OR reflects a decades advance in Age.


As Rodney points out, you should be using meaningful scaling when it
exists.  For some clinical-decision type scales, for example, I often ask
clinicians something like, "how big a difference in points would you need
to see before you'd really sit up and notice?"  Other times I'll use
something like the standard error of measurement.  I like the age/10
because folks find it intuitive.  For other coninuous variables, when the
relation is linear, Harrell often recommends using comparing the 75th to
25 percentile, ie, dividing by the interquartile range.  You can actually
use the UNITS option in Logistic to rescale the variable(s) directly in
the PROC.  It even allows you to produce simultaneously estiamtes for a
variety of scalings.

Mike Babyak