Tree size and relative clade age influence estimation of speciation rate shifts
The development of methods to estimate rates of speciation and extinction from time-calibrated phylogenies has revolutionized evolutionary biology by allowing researchers to correlate diversification rate shifts with causal ecological factors. We use rigorous simulations to evaluate the statistical performance of three widely used modelling approaches - BiSSE, BAMM and MEDUSA - in relation to detection of speciation rates shifts. We simulated sets of trees with each tree having a single increase in speciation rate. We varied the location of shifts, the degree of increase in speciation rate and the total age of the tree. We then used BiSSE, BAMM and MEDUSA to estimate rate shifts. For BiSSE, we assigned different character states for the lineages with different simulated speciation rates. We show that all methods are better at detecting rate shifts when the change in speciation rate is higher, but had high Type II errors (non-detection of rate shifts). While the algorithms more accurately identified rate shifts close to the root of the tree, both perform poorly when the rate shift occurred more recently. All methods performed better with increase in the overall number of tips and the number of tips in the clade with rate shift, both of which are correlated with tree age and speciation rate asymmetry. We discuss the implications of this study for the use and development of methods for hypothesis testing based on diversification rate shifts.