Richness‐dependence of phylogenetic diversity indices

Phylogenetic diversity indices are widely used to characterize the structure and diversity of ecological communities. Most indices are based on a metric that is expected to vary with species richness, so they are standardized to remove this richness‐dependence. With this standardization, values of 0 are consistent with random phylogenetic structure, while phylogenetic clustering is associated with either negative or positive values (depending on the index). One common interpretation of phylogenetic clustering is that it indicates some combination of environmental and biological filtering that restricts the species that can be present in a community. Increasingly, studies are comparing phylogenetic indices along environmental gradients to infer differences in the factors structuring communities. This comparison implicitly assumes that index values are comparable among communities with different numbers of species. Using a set of simulations, I show here that this assumption is incorrect. Holding the strength of filtering constant, communities composed of more species show larger absolute index values. This problem is most pronounced when the environmental filter favors a moderate‐sized clade strongly over others and when using the net relatedness index (NRI) to measure clustering. This bias potentially casts doubt on studies studying phylogenetic index patterns along gradients where richness also varies. Fortunately, the arising generality that more stressful environments have lower species richness and stronger clustering is opposite to this bias and therefore robust. I also show that a simple rarefaction can remove the richness‐dependence of these indices, at the expense of increased error.