To use Mahalanobis distances for an agglomerative clustering algorithm is a good thing. But this shouldn't be called UPGMA, which measures the distance between two groups as the average of all distances (however defined) between individual members of the groups. The Mahalanobis distance is not taken between individual members, but between the groups as wholes.
As far as I understand. Oyvind Hammer Geological Museum University of Oslo On Sun, 1 May 2005, morphmet wrote: > > Hi all, > > in literature, several people have been using Procrustes distances as > the data for a cluster-analysis, more specifically UPGMA. When doing so, > the Procrustes distances between the consensus configurations of each > OTU are used to calculate a tree based on shape similarities between the > consensusses. However, when using Procrustes distances, one is ignoring > possible differences in the amount of within-group variation (or is > assuming it is the same for all the OTUâs), not? The Procrustes distance > between the means of three groups can be the identical, but because of > overlapping variations within two of them, the shape distance will in > many cases be much lower. > > So my question is, whether it is not wiser to use the Squared > Mahalanobis distances of a canonical variate analysis (based on weight > matrix), as a measure of shape distance between group means for a UPGMA, > as this does take into account the amount of within-group variation (and > as I understood even standardises it). Does this make sense or is there > some mathematical-statistical pitfall behind it? > > cheers > > Dominique Adriaens > > > > > > Prof. Dr. Dominique Adriaens > > Ghent University > > Evolutionary Morphology of Vertebrates & Zoology Museum > > K.L. Ledeganckstraat 35, B-9000 Gent > > BELGIUM > > tel: +32 9 264.52.19, fax: +32 9 264.53.44 > > E-mail: [EMAIL PROTECTED] > > URL: http://www.fun-morph.ugent.be/ > > http://www.zoologymuseum.ugent.be/ > > > > > > -- > Replies will be sent to the list. > For more information visit http://www.morphometrics.org > > -- Replies will be sent to the list. For more information visit http://www.morphometrics.org
