Determining the number of clusters

[cbautista]

I'm applying a hierarchical cluster analysis in order to determine groups
of cases. As you know SPSS give us a range of potential number of clusters.
However, is there any methodology or specific procedure to follow to define
what is the most correct number of clusters?

Thanks,

/Christian

[Dogan, Enis]

SPSS 14 has a new clustering procedure: 2-steps clustering, that
determines the numb of clusters automatically. It is not very powerful
though. It usually produces two clusters (my experience) but the nice
thing about it is that it handles categorical variables as well.

To determine the best number of clusters I would also seek external
validation.
Find an external variable (that does not enter the clustering
procedure); run the clustering with 2, 3, 4, etc... number of clusters
and see if cluster membership explains the variance in this external
variable. The external variable should, of course, be related to the
clusters to begin with.

Hope this helps

You might also want to consider Latent Class analysis for this
purpose...


Enis



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Christian Bautista
Sent: Friday, June 16, 2006 10:55 AM
To: [hidden email]
Subject: Determining the number of clusters

I'm applying a hierarchical cluster analysis in order to determine
groups
of cases. As you know SPSS give us a range of potential number of
clusters.
However, is there any methodology or specific procedure to follow to
define
what is the most correct number of clusters?

Thanks,

/Christian

[LUCINDA M TEAR]

The following paper is an application of that kind of approach to clustering
sites using the frequency of occurrence and abundance of species at sites:

Dufrene, M. and P. Legendre.  1997.  Species assemblages and indicator
species:  the need for a flexible asymmetrical approach.  Ecological
Monographs 67:345-366.


----- Original Message -----
From: "Dogan, Enis" <[hidden email]>
Newsgroups: bit.listserv.spssx-l
To: <[hidden email]>
Sent: Friday, June 16, 2006 9:25 AM
Subject: Re: Determining the number of clusters

> SPSS 14 has a new clustering procedure: 2-steps clustering, that
> determines the numb of clusters automatically. It is not very powerful
> though. It usually produces two clusters (my experience) but the nice
> thing about it is that it handles categorical variables as well.
>
> To determine the best number of clusters I would also seek external
> validation.
> Find an external variable (that does not enter the clustering
> procedure); run the clustering with 2, 3, 4, etc... number of clusters
> and see if cluster membership explains the variance in this external
> variable. The external variable should, of course, be related to the
> clusters to begin with.
>
> Hope this helps
>
> You might also want to consider Latent Class analysis for this
> purpose...
>
>
> Enis
>
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Christian Bautista
> Sent: Friday, June 16, 2006 10:55 AM
> To: [hidden email]
> Subject: Determining the number of clusters
>
> I'm applying a hierarchical cluster analysis in order to determine
> groups
> of cases. As you know SPSS give us a range of potential number of
> clusters.
> However, is there any methodology or specific procedure to follow to
> define
> what is the most correct number of clusters?
>
> Thanks,
>
> /Christian
>