The influence of non‐random species sampling on macroevolutionary and macroecological inference from phylogenies

Non‐random species sampling is the rule rather than the exception in phylogenetics, but most phylogenetic methods to infer macroevolutionary and macroecological processes assume that the tips of the phylogenetic tree are either completely sampled or randomly sampled. In this study, we focus on extending the popular BiSSE framework to better account for non‐random sampling of species. The existing BiSSE correction (which we describe hereafter as the unresolved clade correction) cannot be used on trees with clades of more than about 200 species, or when lineages that originate near the root are not sampled. We propose new correction that does not have these two limitations. To assess the performance of our correction relative to the unresolved clade correction, we simulate trees using a common sampling strategy in which representative species of higher clades (e.g. genera) are sampled to include in a phylogeny. Compared to the unresolved clade correction, we show that our new correction gives less biased parameter estimates; has higher power but a slightly elevated false positive rate to detect state dependence in speciation and extinction rates; and is less sensitive to a failure to sample all extant groups of taxa. Over all simulation scenarios, our correction perform equally well under conditions where the unresolved clade correction is applicable and conditions where the unresolved clade correction is inapplicable. Given that both our correction and the unresolved clade correction have their own advantages and disadvantages, we suggest combining the two corrections. This can be done by applying our correction to groups that exceed the size limit of the unresolved clade correction or to account for the uncertainties in the placement of the lineages that originate near the root.