Date:         Thu, 2 Sep 2004 16:52:43 +0000
Reply-To:     sajeel manzoor <sajeelmanzoor@HOTMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         sajeel manzoor <sajeelmanzoor@HOTMAIL.COM>
Subject:      Re: stats question
Content-Type: text/plain; format=flowed

Yes it is ordinal....
How to handle such a variable in clustering ?? It depends, there are several
options available, however each of the treatments revolve around the
question as to what are the other variables, in addition to age, being
considered and whether the dispersion i.e interval in one variable is more
important than the other. In other words, you may have a variable Income
which is on a continuous scale and Age on an ordinal scale then in
clustering, if no variable standardization is done, then Income will turn
out as the main dimension and will get more weight in clustering because of
the large dispersion due to scale.
So I would recommend standardizing the variables before using them in your
clustering exercise..
There are several methods available for it .. I use "proc standard"

I hope this helps!!

Sajeel
Statistical Modeler
ABC


>From: "DePuy, Venita" <depuy001@DCRI.DUKE.EDU>
>Reply-To: "DePuy, Venita" <depuy001@DCRI.DUKE.EDU>
>To: SAS-L@LISTSERV.UGA.EDU
>Subject: Re: stats question
>Date: Thu, 2 Sep 2004 09:58:34 -0400
>
>  assuming if age < 20 ageband=1, else if age < 40 ageband=2, else if age <
>60 ageband=3, sort of thing.
>
>Then ageband is ordinal.  ie order matters, as opposed to nominal (1=male,
>2=female, order is meaningless).
>
>I've done this several times and seen it done as well . . ie maybe being
>older (70+ ) might have an effect on health, but the specific age doesn't
>really matter.  THe most recent paper that comes to mind had 4 categories
>for age (all adult).
>
>Not sure how to answer the clustering question though.
>HTH
>Venita
>
>-----Original Message-----
>From: Andrew Bolton
>To: SAS-L@LISTSERV.UGA.EDU
>Sent: 9/2/2004 9:11 AM
>Subject: ot: stats question
>
>Getting a whiles since I did stats at school so apologies if this is an
>obvious question.
>
>How would I classify the data type 'age-band', i.e. it's not continous
>but
>seem to remeber it's not categorical (is is 'meristic'?).  I'm wondering
>the best way to handle age-band in clustering, and whether if would be
>valid to use a single variable which is coded up as 0,1,2,3 etc for
>progressivley older bands (i.e. an implied order)?
>
>Any help/links much appreciated.
>
>Cheers,
>
>Andy

_________________________________________________________________
Protect your PC - get McAfee.com VirusScan Online
http://clinic.mcafee.com/clinic/ibuy/campaign.asp?cid=3963