Date: Wed, 14 Jan 2009 22:45:06 +0800
Reply-To: Murphy <goladin@GMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Murphy <goladin@GMAIL.COM>
Subject: Re: Graphical summaries of 3 level discriminants
Content-Type: text/plain; charset="UTF-8"
I understand the problem now. Not sure whether there are any other measures that can be used.
I think Nat's solution is pretty good.
Certified Base SAS Programmer for SAS V9
Certified Advanced SAS Programmer for SAS V9
Datashaping Certified SAS Professional
From: Peter Flom [mailto:firstname.lastname@example.org]
Sent: Wednesday, January 14, 2009 10:31 PM
To: Murphy; SAS-L@LISTSERV.UGA.EDU
Subject: RE: Graphical summaries of 3 level discriminants
Murphy <email@example.com> wrote
>I am not sure whether I am being correct here but I think you can manually create the ROC chart by calculating the number of correct hits against the misclassified.
Thanks for your reply.
One could, indeed, calculate the number of correctly classified vs. misclassified, but with a 3 level discriminant it's problematic. Suppose the three levels are MARRIED, SINGLE, and OTHER. (These are not my categories, but they will do .... ).
One could compute the number of married people classifed as M vs. misclassified, the number of single people classified as S vs. misclassified, and the number of others classified as other vs. misclassifed. But then, what is a false positive, true positive, and so on? Wouldn't you have to have true married, false married, true single, false single, and so on? And, given that, what would specificity and sensitivity be?
Peter L. Flom, PhD
www DOT peterflom DOT com