The unbearable uncertainty of Bayesian divergence time estimation

摘要: Divergence time estimation using molecular sequence data relying on uncertain fossil calibrations is an unconventional statistical problem. As the provide information about distances only, of absolute times and rates has to rely in prior, so that model only semi-identifiable. In this paper, we use a combination mathematical analysis, computer simulation, real analysis examine uncertainty posterior estimates when amount increases. The extends infinite-sites theory Yang Rannala, which predicts distribution divergence rate approaches infinity. We found credibility interval general decreases reaches non-zero limit size However, for node with most precise calibration (as measured by width divided mid value), do not really make estimate any more precise. propose finite-sites square its infinite-data at 1/n, where n length. suggest procedure partition into due uncertainties sampling errors data. evaluate impact conflicting point out narrow intervals or overly can be produced erroneous calibrations.