Hi Juan, Thank you for your answer - this makes sense to me. However, could the same not be said for the unormalised eigenvectors of A as they preserve the original similarities among the taxa, have the same units etc.? Also, since solve(A) has the same eigenvectors as A but with reciprocal eigenvalues it has some of these properties too.
I'm not saying that PCO of J-A is bad, but what I would like to know is whether PCA of solve(A) could be considered as an equal alternative. Cheers, Jarrod On Wed, 2012-03-14 at 11:58 +0100, Juan Antonio Balbuena wrote: > Hello, > I think the reason for using principal coordinates in parafit is that > taken all together the PCO axes preserve the original dissimilarities > among the taxa and can thus the resulting ordination is an exact > representation of the host and parasite phylogenies in the hyperspace. > In addition, the scales of the PCO axes are interpretable in the units > of the original resemblance measure. > > Cheers > > Juan A. Balbuena > > > > > Hello Everyone, > > > > I am fitting mixed models in which I have two phylogenies and I would > > like to understand them in the context of earlier work. In parafit > > (Legndre 2002) principal coordinates of the distance matrices (J-P) are > > used, where J is a matrix of ones and P the phylogenetic correlation > > matrix. For me, using unnormalised eigenvectors of solve(P) would > > result in nicer properties, and I wonder whether anyone could tell me > > whether there is some deep motivation for using the principal > > coordinates of J-A that I'm missing? > > > > Kind Regards, > > > > Jarrod > > > > _______________________________________________ > > R-sig-phylo mailing list > > R-sig-phylo@r-project.org > > https://stat.ethz.ch/mailman/listinfo/r-sig-phylo > > > > > > _______________________________________________ R-sig-phylo mailing list R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
