Date: Fri, 13 Dec 2002 12:26:13 -0500
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Arthur J. Kendall" <Art@DrKendall.org>
Organization: Social Research Consultants
Subject: Re: dissimilarities in Clustering
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
open a syntax window
key in the word
highlight it and then click the syntax icon
it shows you how to specify the nature of your (dis)similarity
coefficient and how to specify matrix input.
You can put any data through the Proximities procedure, write out the
matrix, and see how it is formatted.
also try <help> and "matrix"
Hope this helps.
Social Research Consultants
University Park, MD USA
Don Eduardo Miranda wrote:
> I am trying to cluster some variables using hierarchical clustering. However,
i only possess the matrix of dissimilarities, not the actual data matrix
(variables * cases) . I was wandering if i could pass this
dissimilarity matrix as the argument of the CLUSTER command. I have
seen this done in some examples which use the Proximities command
before, but i wonder on the correctness of this approach. My confusion
comes from the fact that the cluster function also expects a
dissimilarity measure type and asumes SEUCLID as default, and sometimes
(actually in most of the cases) i have seen this command applied to the
original data matrix , so i assume this function internally generates a
dissimilarity matrix based on its parameter matrix. So if i pass my
dissimilarity matrix as the argument, the actual algorithm would be
working on the dissimilarities of the dissimilarities matrix (something
like a meta-dissimilarity) and well, once on that level i have no idea
of what to expect from my clustering. Additional to this, the
dissimilarity measure i am using does not belong to the set of
dissimilarity measures supported by SPSS, so i dont know whether
assuming the dissimilarity measure to be the default one (if i dont
indicate any dissimilarity measure) will alter my results as some steps
of the clustering algorithm will need to re-calculate the distances once
the clusters are being formed.
> Could you please help me clarify this?
> thank you very much
> Eduardo Miranda
> Departamento de Informatica da FCT/ UNL
> Quinta da Torre, 2829-516 Caparica, Portugal
> Tel: +351-21 294 85 36 - Ext. 10731
> Fax: +351-21 294 85 41
> E-mail: email@example.com