Hierarchical Cluster Analysis

[Joel M SHERMAN]

Dear Listers,

I am familar with the SPSS routines to do cluster analysis, but I'm
wondering if anyone is familiar with how this method compares to
geospatial approaches in clustering, which typically only use a latitude
and longitude (x,y) variables to characterize clusters.

Using H-clustering in SPSS, If I cluster a set of drinking water
sources based on their arsenic levels and some other meaningful
variables, using euclidian distances, the output (dendrogram or some
other tool) would show the clusters and which drinking water systems
belong to which clusters based on an aggregation of the proximity
matrices of each of the variables I use to cluster the systems.

However, if I use some GIS tool, like CrimeStat or SatsScan, it seems
that only a spatial component (proximity of lat/long points) is used to
form clusters, rather than known, meaningful attributes of the water
systems.

Am I correct to infer that these are two distinct types of clustering?
What methods, if any, can reconcile these two?

TIA.

Joel M. Sherman
Environmental Public Health
Department of Human Services
800 NE Oregon Street, Suite 608
Portland, Oregon 97232
971.673.0441 - phone
[hidden email]

[Art Kendall]

Yes, these are distinct types of clustering.

The Classification Society list class-l   would be a more focused forum
to post your question.

http://www.classification-society.org/csna/csna.html

http://www.classification-society.org/csna/lists.html#class-l

http://www.classification-society.org/csna/lists.html#class-l

One approach find points that are close together in a two (or three)
dimensional space.

The other approach find points that are close together in a space in
which the dimensions are the variables (attributes) that describe the
entities.

Conceptually, physical location (and time) can be variables that
describe cases (entities) in addition to color concentrations of
chemicals etc..

It sounds like something the ecologists or others on class-l may have
dealt with.  I suggest you post your query there.


Art



Joel M SHERMAN wrote:

>Dear Listers,
>
>I am familar with the SPSS routines to do cluster analysis, but I'm
>wondering if anyone is familiar with how this method compares to
>geospatial approaches in clustering, which typically only use a latitude
>and longitude (x,y) variables to characterize clusters.
>
>Using H-clustering in SPSS, If I cluster a set of drinking water
>sources based on their arsenic levels and some other meaningful
>variables, using euclidian distances, the output (dendrogram or some
>other tool) would show the clusters and which drinking water systems
>belong to which clusters based on an aggregation of the proximity
>matrices of each of the variables I use to cluster the systems.
>
>However, if I use some GIS tool, like CrimeStat or SatsScan, it seems
>that only a spatial component (proximity of lat/long points) is used to
>form clusters, rather than known, meaningful attributes of the water
>systems.
>
>Am I correct to infer that these are two distinct types of clustering?
>What methods, if any, can reconcile these two?
>
>TIA.
>
>Joel M. Sherman
>Environmental Public Health
>Department of Human Services
>800 NE Oregon Street, Suite 608
>Portland, Oregon 97232
>971.673.0441 - phone
>[hidden email]
>
>
>
>