The issue is not so much whether the "data are of a naturally
hierarchical nature" but of the definition of what one wishes to
call a cluster. Ward's method like the complete link method
assume that clusters should be round and compact. Other methods
only assume that there are gaps between cluster and that the
shape of the cluster does not matter. There are, of course, many
methods in between. Then there is also the issue of how to define
the distances between points. This makes it difficult to agree on
what is the "best" hierarchical solution.

Jim

----------------------------------------------
F. James Rohlf, Distinguished Professor
Dept. Ecology & Evolution, Stony Brook University
Stony Brook, NY 11794-5245 
http://life.bio.sunysb.edu/ee/rohlf


 

> -----Original Message-----
> From: Classification, clustering, and phylogeny estimation 
> [mailto:[log in to unmask]] On Behalf Of leo horseman
> Sent: Saturday, September 16, 2006 2:07 PM
> To: [log in to unmask]
> Subject: Re: clustering with distance matrix as input
> 
> Ward's method as initially implemented by Ward generally does 
> not produce the best hierarchical solutions.  Unless your 
> data are of a naturally hierarchical nature, I would suggest 
> looked at some other method.  You may wish to check the 
> website for "CLUSTAN" package of programs; also obtain an old 
> CLUSTAN manual which would explain in some detail the 
> operation of many distance measures as well as a good variety 
> of clustering methods.
> 
> Leo Horseman
> 
> >From: "Kejun (Kevin) Mei" <[log in to unmask]>
> >Reply-To: "Classification, clustering, and phylogeny 
> estimation"            
> >   <[log in to unmask]>
> >To: [log in to unmask]
> >Subject: clustering with distance matrix as input
> >Date: Fri, 15 Sep 2006 15:17:57 -0500
> >
> >Can someone please tell me a clustering method that use a
distance 
> >matrix as input? I tried implementing the Ward method, but
found the 
> >final clusters are not so good in my application, when I trace
how 
> >clusters are grouped and how cluster centers are updated. I 
> am seeking 
> >an improved version of the Ward method, or any others.
> >
> >Thank you so much and have a great weekend.
> >
> >-Kevin
> >
> >----------------------------------------------
> >CLASS-L list.
> >Instructions: 
> >http://www.classification-society.org/csna/lists.html#class-l
> 
> ----------------------------------------------
> CLASS-L list.
> Instructions: 
> http://www.classification-society.org/csna/lists.html#class-l
> 

----------------------------------------------
CLASS-L list.
Instructions: http://www.classification-society.org/csna/lists.html#class-l