I know this is late but I just saw it today......

Last year it was suggested to me to use hierarchical clustering to derive
initial centers for a kmeans application. Basically, given a large dataset
with at least interval (better if ratio) measurements it makes indeed sense
to:

1) Select randomly a sample of a few hundred cases

2) Run a hierarchical clustering procedure and save the centers generated

3) Feed into a kmeans procedure such centers, so that the final solution
applied to the entire dataset is less influenced by the arbitrary selection
of the initial centers

On http://www.listserv.uga.edu/cgi-bin/wa?A2=ind0404&L=spssx-l&P=R4505 you
shall find a reply I have got from Wim Beyers about my intention to use SPSS
twostep clustering method and some references to the above mentioned
alternative method.

Best regards,

Mr. Luca MEYER				
Survey research, data analysis & more: http://www.lucameyer.com/
Tel: +390122854456 - Fax: +390122854837 - Mobile: + 393394950021

"If you can't feed a hundred people, then feed just one." - Mother Teresa -
www.thehungersite.com


> -----Messaggio originale-----
> Da: Classification, clustering, and phylogeny estimation
> [mailto:[log in to unmask]] Per conto di jessie jessie
> Inviato: venerdì 9 settembre 2005 22.31
> A: [log in to unmask]
> Oggetto: Using more than one clustering method...
>
>
> If I use more than one clustering method, e.g. Kmeans
> and Hierarchical, to cluster the same set of data,
> does the joint operation necessarily increase my
> chance of finding the underlying clusters that might
> be verified in the future? Is there any books/papers
> that talk about this topic? Thanks a lot in advance!
>
> Jessie
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
>