Date:         Fri, 14 Mar 2003 12:01:25 -0800
Reply-To:     Vadim Pliner <Vadim.Pliner@VERIZONWIRELESS.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Vadim Pliner <Vadim.Pliner@VERIZONWIRELESS.COM>
Organization: http://groups.google.com/
Subject:      Re: varclus on categorical data
Content-Type: text/plain; charset=ISO-8859-1

William Kossack <kossackw@njc.org>:
> I have a user that has coded all his data as 0 or 1 as real numbers not
> integers.
>
> Anyone point me to a solution on how to cluster categorical/binary
> variables.

You can use %polychor macro to compute tetrachoric correlations which
are suitable for binary data. I think this macro is still available at
the SI web site for download. Among other parameters of this macro,
you should specify the name of the correlation matrix, say, cor_name
(out=cor_name) and type=distance. Then, to cluster your binary
variables you can use proc cluster with data=cor_name option.

Vadim