BAMM gives misleading rate estimates in simulated and empirical datasets.

Abstract

In a previous paper, we used simulations and empirical data to show that BAMM (Bayesian Analysis of Macroevolutionary Mixtures) can give misleading estimates of rates and rate shifts. In simulations, BAMM underestimated rate shifts across every tree analyzed, and assigned incorrect rates to most clades in most trees. In empirical analyses, BAMM behaved as expected from simulations, and assigned different rates to clades when clades were analyzed alone versus across the tree (i.e., with rate heterogeneity). Rabosky recently criticized our paper, focusing primarily on the idea that our comparison of BAMM to another approach (method-of-moments estimators of Magallón and Sanderson, or MS estimators) was unfair to BAMM. Here, we provide further evidence that BAMM gives misleading rate estimates in empirical studies. We then describe how Rabosky's rown method comparisons were either acknowledged as being problematic or were described inaccurately (to favor BAMM). Finally, we show that the MS estimators can perform well when rates vary over time, despite untested assertions that they require constant rates to be accurate. Many other methods are available for analyzing diversification rates: we argue that BAMM should be avoided for estimating both diversification rates and rate shifts.