Implications of uniformly distributed, empirically informed priors for phylogeographical model selection: A reply to Hickerson et al

摘要

Establishing that a set of population-splitting events occurred at the sametime can be a potentially persuasive argument that a common process affectedthe populations. Oaks et al. (2013) assessed the ability of anapproximate-Bayesian method (msBayes) to estimate such a pattern ofsimultaneous divergence across taxa, to which Hickerson et al. (2014)responded. Both papers agree the method is sensitive to prior assumptions andoften erroneously supports shared divergences; the papers differ about theexplanation and solution. Oaks et al. (2013) suggested the method's behavior iscaused by the strong weight of uniform priors on divergence times leading tosmaller marginal likelihoods of models with more divergence-time parameters(Hypothesis 1); they proposed alternative priors to avoid strongly weightedposteriors. Hickerson et al. (2014) suggested numerical approximation errorcauses msBayes analyses to be biased toward models of clustered divergences(Hypothesis 2); they proposed using narrow, empirical uniform priors. Here, wedemonstrate that the approach of Hickerson et al. (2014) does not mitigate themethod's tendency to erroneously support models of clustered divergences, andoften excludes the true parameter values. Our results also show that thetendency of msBayes analyses to support models of shared divergences isprimarily due to Hypothesis 1. This series of papers demonstrate that if ourprior assumptions place too much weight in unlikely regions of parameter spacesuch that the exact posterior supports the wrong model of evolutionary history,no amount of computation can rescue our inference. Fortunately, more flexibledistributions that accommodate prior uncertainty about parameters withoutplacing excessive weight in vast regions of parameter space with low likelihoodincrease the method's robustness and power to detect temporal variation indivergences.