Date:         Thu, 22 Dec 2005 10:24:39 -0500
Reply-To:     Venita DePuy <depuy001@NOTES.DUKE.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Venita DePuy <depuy001@NOTES.DUKE.EDU>
Subject:      Re: 95% CI for difference between medians?
Comments: To: Deborah Wentworth <debby@CCBR.UMN.EDU>
In-Reply-To:  <Pine.GSO.4.44.0512220910480.20183-100000@sturgeon.ccbr.umn.edu>
Content-Type: text/plain; charset="US-ASCII"

Hi Deborah -
I'm not aware of such a thing, although it may exist anyway.
Assuming you have non-normal data; try the Wilcoxon rank-sum test in Proc
Npar1way, assuming the data are independent.
This tests for differences in centers of distribution, assuming the
underlying distributional shapes are the same.
(Typically this is the test you use to test the difference between
medians).
-Venita




Deborah Wentworth <debby@CCBR.UMN.EDU>
Sent by: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
12/22/2005 10:19 AM
Please respond to
Deborah Wentworth <debby@CCBR.UMN.EDU>


To
SAS-L@LISTSERV.UGA.EDU
cc

Subject
95% CI for difference between medians?






Greetings -

Someone I'm working with suggested that I test the difference between two
medians rather than two means, and to report the 95% CI for the
difference.  I have no idea how to do this, and in my perusal of both
the archives and SAS online help, I haven't been able to find a
procedure that will give the CI for a difference between medians.

Is there such an animal?

Thanks if you can help.

And whether you can or can't, happy holidays to all of you who
generously offer help and guidance to those of us who have a little less
SAS savvy! (And for your holiday trivia party - despite how we tend to use
it, savvy is a noun and something you have rather than something you are.)

Deb