|Date: ||Tue, 9 Jul 1996 09:03:27 -0700|
|Reply-To: ||"William D. Allen" <WALLEN@CHE2.CHE.UMN.EDU>|
|Sender: ||"SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>|
|From: ||"William D. Allen" <WALLEN@CHE2.CHE.UMN.EDU>|
|Organization: ||University of Minnesota|
|Subject: ||Re: WAVERAGE in Cluster|
Joe Wheaton wrote:
> I have been experimenting with different clustering solutions using the
> simple matching coefficient with 13 dichotomous variables on 710 subjects.
> The solution that makes the most sense uses the WAVERAGE method. I have
> found a good deal about BAVERAGE (or UPGMA) but nothing about the within
> group clustering method other than a statement by Aldenderfer & Blashfield
> (1984, Sage) saying there are a number of variants to the average clustering
> method. Can anyone give me more information?
I am also using Within-Groups Average Linkage (WAVERAGE) in my analysis of 400
married couples. Although others
who have worked with the data set I am using also used WAVERAGE, their
rationale for this specific method
versus others (say, Ward's) was unclear to me. The method does make intuitive
sense, determining clusters based
on a measure of how they hang together...
In addition to the texts that have been mentioned here, I would suggest getting
hold of a SAS manual (gasp!)
and reading the chapters on CLUSTER and FASTCLUS. Rather than confuse me, I
found that these short sections
helped me to better understand the differences between methods. They also
provided a different prospective on
how to go about clustering including giving some examples.
I don't know whether this is feasible for you but you might want to consider
using both programs and validating
what you find in one with the other. It would mean a bit more time and effort
but given the current state of
uncertainty concerning concensus on cluster analytical techniques, it might be
a good idea.
Bill Allen (who is by no means an expert on CA; just another explorer...;^)