PERFORMANCE OF MACROEVOLUTIONARY RATE ESTIMATION APPROACHES: A SIMULATION REVIEW

Rates of speciation, extinction, and sampling form an important part of many macroevolutionary studies, and a wide variety of methods are now available with which these rates can be estimated using fossil data. In particular, several Bayesian phylogenetic approaches have recently been made available. When new rate estimation approaches are initially presented, they often include validation of the method using simulation. However, simulation studies that compare the accuracy and precision of the outcomes from each method to one another, and under different and realistic evolutionary scenarios, have been lacking. Here we compare the success of rate estimation methods that have been used in macroevolutionary studies in the last five years, including both occurrence based and phylogenetic approaches. We simulated fossil records designed to resemble carnivoran, dinosaurian and molluscan clades, and evolutionary scenarios including mass extinction, radiation, and stochastic variation, among others. We also investigated the effect of timebinning schemes, and of topological inaccuracy in phylogenetic approaches. The methods tested include: Alroy’s gap-filler and three-timer; Foote’s likelihood and boundary-crosser, with and without phylogenetic range extensions; capture-mark-recapture; Sakamoto et al’s Bayesian phylogenetic approach; the fossilized-birth-death range skyline model; and PyRate. The most appropriate choice of method depends on the clade and time period under investigation. Many occurrence based approaches are biased towards low estimates of both speciation and extinction rates, whereas time varying capture mark recapture models and phylogenetic approaches overestimate rates when sampling rate is low (e.g. dinosaurs) but are more accurate when sampling is higher (e.g. molluscs). There is broad variability in rate estimates across birth-death and sampling realisations, for all of the methods. Our aim for this comprehensive assessment of the approaches currently available to estimate evolutionary rates is to provide a useful guide to those wishing to implement these analyses on their own data.