Multi-variate Cluster Analysis

Alastair Duncan 100074.213 at compuserve.com
Fri Oct 13 08:00:00 EDT 1995


Hello Everyone,

I've got a large pile of work in my in-tray - my boss wants me to do some
Canberra-metric Group Averaged Cluster analysis on practically our complete
database within the next two weeks !!! (bosses ... don't you just love them
!?!). Has anyone done this kind of thing with GRASS ?? I guess it would be
pretty difficult as the site data format only allows one attribute to be stored
with the object ..... the only way I can figure out doing it at the moment is b
y
using a load of seperate DOS programmes (I need to query the database, take tha
t
data into a spreadsheet and then into WordPerfect in order to format it right
for a cluster analysis programme nicked from a college that only allows 125
points to be analysed, then into a graphics package to printout the dendrogram,
and then into the GIS to map out the clusters ..... it's a bit like making a
satellite out of bits from the back of old televisions).
        Does anybody know of a better way of doing it than that ??? 

        Also does anybody know if the Canberra-metric similarity index has any
set levels of significance for dealing with clusters - after reading the (rathe
r
thin) documentation I've got, it seems to be more of a subjective and
descriptive analysis, rather than say correlation which has tables of set level
s
of significance associated with the number of samples used. It also strikes me
that when using it for geographical objects it would be more realistic to
measure the similarity between adjacent sites only - not the complete data-set,
as in the cluster analysis, unless the result wanted is a broadly based
classification. Has anyone got any experiance, ideas, references along these
lines ??? Also I've got documentation on the Canberra-metric and Bray-Curtis
indexes - but not the McConnaughey Similarity Index which the DOS software
offers as an option - does anyone know what the McConnaghey Index is ??

Thanks a lot for any help on this,

Cheers, Al.


__________________________________________________________________________

Alastair Duncan,
Spatial Information Scientist,
NRA National Centre for Instrumentation and Marine Surveillance,
Bath, UK.

e-mail :- 100074,213 at compuserve.com






More information about the grass-user mailing list