Fig. 1

Median absolute percentage error of node times (on log scale) reconstructed from data sets simulated under three simulation types with different relationships between molecular evolution and speciation. The mean error is calculated only for branches whose bipartitions are shared between the true and inferred trees, and therefore ignores errors in topology which may occur in the BEAST 2 analyses. The three simulation models are Unlinked (instantaneous covariance of molecular rates and speciation rates = 0), Continuous (instantaneous covariance = 0.0044), and Punctuated (instantaneous covariance = 0, bursts of substitutions added at speciation events). Topologies and node times were reconstructed using three different analytical methods, with an uncorrelated lognormal ‘relaxed clock’ rate prior (UCLN) in BEAST 2, the autocorrelated lognormal rate prior (ACLN) in PAML, and the UCLN in PAML