Date:         Mon, 26 Sep 2005 16:20:32 -0700
Reply-To:     Aric Zion <Aric.Zion@asu.edu>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Aric Zion <Aric.Zion@asu.edu>
Subject:      Re: Cluster Analysis - best practices
Content-type: text/plain; charset=utf-8

A clear overview is provided by H. Charles Romesburg's "Cluster Analysis for Researchers", (2004) Lulu Press.  While this doesn't give guidance on how to specifically run Cluster Analysis in SPSS, it does offer an very clear view of how cluster analysis operates.

Aric


        -----Original Message-----
        From: SPSSX(r) Discussion on behalf of Bob Schacht
        Sent: Mon 9/26/2005 1:22 PM
        To: SPSSX-L@LISTSERV.UGA.EDU
        Cc:
        Subject: Re: Cluster Analysis - best practices



        At 04:32 AM 9/26/2005, cristiano wrote:
        >Dear listers,
        >  I'm a statistician but I'm looking for some books/resources/example for
        >using Cluster Analysis with SPSS: i'd like to know the models and methods
        >behind this analysis.
        >  In your experience, could you suggest to me some stuff?
        >  Thanks in advance
        >  Cristiano

        Cristiano,
        As a prelude to your reading, let me comment in general. Cluster analyses
        fall into two approaches: One is polythetic agglomerative in nature, the
        other monothetic subdivisive.

        Polythetic agglomerative methods start with every case as an individual,
        and proceed to cluster by combining cases that most closely resemble each
        other. In each step of the analysis, the similarity between remaining cases
        and clusters is measured, and those most closely resembling each other are
        combined. This proceeds by steps as far as one wants to go, based on
        measures of cohesion or similarity.

        Monothetic subdivisive methods, on the other hand, start with all cases
        combined into one supergroup. The procedure in this case is how to
        subdivide the supergroup in to two groups in a way that maximizes the
        *difference* between the two groups. I'm not clear on how this procedure
        works, but it may begin with variables with the highest degree of
        variability, and splitting the cases at the mean. Again, the process
        proceeds stepwise until some threshold criterion is reached.

        You may have some a priori reason for preferring one approach over the
        other. Descriptions of the methods may not identify themselves clearly with
        these alternatives, so this overview might prove helpful.

        Bob


        Robert M. Schacht, Ph.D. <schacht@hawaii.edu>
        Pacific Basin Rehabilitation Research & Training Center
        1268 Young Street, Suite #204
        Research Center, University of Hawaii
        Honolulu, HI 96814