communities (Kembel and Hubbell, 2006). Observed values smaller or larger than 975 of the randomizations were considered significantly structured (P < 0.05).
Compositional similarity between all pairwise comparisons of communities was quantified with the Sørensen Index:
where Sij is the number of taxa common to both communities i and j, and Si and Sj are the total number of species found in community i and j, respectively (Krebs, 1998). By analogy, phylogenetic similarity between two communities was quantified by using an index, coined PhyloSor:
Here, BLij is the branch length common to both communities i and j, and BLi and BLj are the total branch lengths of community i and j, respectively.
The PhyloSor index ranges from indefinably close to 0 (two communities only share a very small root) to 1 (both communities are composed of the same taxa). Similar approaches have been carried out by Lozupone and Knight (2005) and Ferrier et al. (2007) when considering the closely related Jaccard and Bray-Curtis similarity indices. Using PhyloSor, one can test whether two communities are phylogenetically more or less similar than what is expected given their taxa similarity. This is done by comparing the phylogenetic similarity of the observed communities to a null expectation obtained by randomly sampling the pool of all of the species identified in the study while constraining the number of taxa in each community and the number of taxa shared by the two communities.
By analogy with the well-established distance–decay relationship, which describes the decrease in compositional similarity between two communities with increasing geographic distance (or equivalently elevational separation) between them (Soininen et al., 2007), we used PhyloSor to quantify the decrease in phylogenetic similarity with distance (phylogenetic distance–decay). We tested whether the slope in the decay of phylogenetic similarity was greater or less than what was expected given the taxonomic decay in similarity by comparing the observed slope with a distribution of distance–decay slopes obtained by randomizing the loca-