Xiao,
The short answer to your question is that no, it is not correct to use a population-level phylogeny to conduct a phylogenetic ordination (PACA, a phylomorphospace, phylogenetic PCA). This is for the same reason that one should not use a population or individual
level phylogeny in any phylogenetic comparative method. Here is why.
For phylogenetic comparative analyses, the goal is to account for the non-independence of observations (species) during the analysis. Here the species means are the data, and we estimate the nonindependence among them using an object covariance matrix, which
is based on the phylogeny. That phylogeny is usually bifurcating, and each bifurcating event describes the separation of those species after their speciation event. In other words, after the bifurcation, the two species evolve independently.
With populations, this is not the case, as there is frequently gene flow between sets of populations. What that means is that their nonindependence is more of a network, and not a phylogeny, and using a phylogeny to describe their expected covariation is incorrect.
In other words, a phylogeny is the wrong model of expected nonindependence. Instead, we require an object covariance matrix that that describes the network of gene flow among populations, and this will not be bifurcating. For instance, using a migration matrix
to account for nonindependence would be far more correct biologically. This issue was described at length in Felsenstein, 2002; Stone et al. 2011.
Now the same logic applies to phylogenetic ordination methods. One can represent populations by a bifurcating phylogeny, but that ignores any gene flow between them (which for populations with a species can occur). This is incorrect. So a better option is to
represent the nonindependence using some other matrix, such as a migration matrix.
To implement this, one can estimate the object covariance matrix using some data that allows the migration matrix to be estimated (genetic correlations between populations work to a first-order approximation), and then use the function 'ordinate' in RRPP (which
geomorph calls) to perform the population-level 'phylogenetic' ordination. As data one uses the population means.
Then, if you want to have the individual level data in the plot, one can simply project them into the ordination space directly. Fortunately, the 'ordinate' function allows this, with the 'newdata' component of the function.
Hope this helps,
Dean
--
Dr. Dean C. Adams
Distinguished Professor
Department of Ecology, Evolution, and Organismal Biology
Iowa State University