Bayesian inference provides a powerful framework for integrating different sources of information (in particular, molecules and fossils) to derive estimates of species divergence times. Indeed it is currently the only framework that can adequately account for uncertainties in fossil calibrations.
We use two Bayesian Markov chain Monte Carlo (MCMC) programs, MULTIDIVTIME and MCMCTREE, to analyze three empirical datasets to estimate divergence times in amphibians, actinopterygians, and felids. We evaluate the impact of various factors, including the priors on rates and times, fossil calibrations, substitution model, the violation of the molecular clock and the ratedrift model, and the exact and approximate likelihood calculation.
Assuming the molecular clock caused seriously biased time estimates when the clock is violated, but two different ratedrift models produced similar estimates. The prior on times, which incorporates fossilcalibration information, had the greatest impact on posterior time estimation. In particular, the strategies used by the two programs to incorporate minimum and maximumage bounds led to very different time priors and were responsible for large differences in posterior time estimates in a previous study.
The results highlight the critical importance of fossil calibrations to molecular dating and the need for probabilistic modeling of fossil depositions, preservations and sampling to provide statistical summaries of information in the fossil record concerning species divergence times.
