Dear All,

I am new to Matlab and was wondering whether it is possible to calculate p-values for each division in a dendrogram resulting from hierarchical clustering with standard methods (average-linkage, ward, etc.) and distance functions (euclidian, canberra, manhattan, etc.)

Best regards,
Hans

On 8/16/2010 1:20 PM, Hans wrote:
> I am new to Matlab and was wondering whether it is possible to calculate
> p-values for each division in a dendrogram resulting from hierarchical
> clustering with standard methods (average-linkage, ward, etc.) and
> distance functions (euclidian, canberra, manhattan, etc.)

Hans, hierarchical clustering as implemented by the LINKAGE function 
does not involve any statistical model, so no p-values.  You might be 
able use GMDISTRIBUTION to cluster your data, but it is not a 
hierarchical method per se.

Hi Peter,

thanks for your answer. I just had some time to go through the Matlab documentation and I came across the cluster, linkage, dendrogram and pdist package which provide everything for hierarchical clustering.  but indeed I did not find anything regarding p-values. So, apart from the GMDIST. it seems that there is no other simple way to calculate p-values for dendrograms/clusters.

Cheers

On 8/16/2010 3:14 PM, Hans wrote:
> indeed I did not find anything regarding p-values. So, apart from the
> GMDIST. it seems that there is no other simple way to calculate p-values
> for dendrograms/clusters.

There may be some way to do this with permutation tests, but you're on 
your own.