Date:         Wed, 27 Feb 2008 05:01:49 -0800
Reply-To:     Amw5Gster <amw5gster@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Amw5Gster <amw5gster@GMAIL.COM>
Organization: http://groups.google.com
Subject:      Re: Missing data that means something
Comments: To: sas-l@uga.edu
Content-Type: text/plain; charset=ISO-8859-1

Thanks, all (and those who emailed me separately).  Good stuff that I
hadn't wholly considered.  I think my approach is going to be a bit of
a blend of these suggestions.  I'm going to create binary flags for
those variables where a missing indicates NULL or n/a.  And cluster
those separately, then cluster the clusters on those variables that
are common amongst the entire population.
For example,
everyone who purchased product A, cluster the observations only on
those variables that pertain to purchase of product A
repeat for product B, C, etc.
Can do the same thing for presence of children/children ages
Then use the cluster identifiers for all these segments as a nominal
input variable into the master clustering

Haven't fully thought it through, but it sounded good when I first
thought of it (like so many of my plans....)