MODEL AVERAGING IN PHYLOGENETIC PALEOBIOLOGY

Bayesian methods for palaeontological phylogenetics rely on appropriate model choice to be statistically consistent. Inappropriate model choice for a particular dataset can lead to error in both the tree topology, and any associated model parameters.

In recent years, more complex models for integrating fossil data in phylogenetic estimations have been proposed. For example, the fossilized birth-death model (FBD) is a hierarchical model that incorporates a model of character evolution, a model describing the distribution of rates of evolution across the tree, and a model describing the process of speciation, extinction and fossil recovery that generates the observed tree. While the FBD model is a powerful tool, it also poses new challenges to how we select appropriate models for phylogenetic analyses. In the FBD model, fossil ages are both data and model outputs (posterior distributions of the fossil ages). This can generate support for overly-conservative model estimates.

In this talk, we explore the use of model averaging approaches, particularly reversible jump Markov Chain Monte Carlo, to choose among models. This approach is lighter-weight than other model selection methods, and avoids the issue of overly-conservative model selection. We will preview results from several groups of Cambrian Echinoderms to demonstrate the power of the approach and contrast it with results from prior studies.