The website for this project is http://phyletica.org/ecoevolity-model-prior
This repository serves as an open-science notebook for research conducted by the Phyletica Lab to assess the performance new model priors implemented in the software package, Ecoevolity.
Ecoevolity is a software package for full-likelihood Bayesian comparative phylogeographic analyses. It compares the timing of evolutionary events across an arbitrary number of taxa, or "comparisons," to infer whether such events were temporally clustered, which might suggest a shared biogeographic process. These evolutionary events can comprise the divergence between two populations or species, or the change in the effective size of a population.
To do this, ecoevolity uses Bayesian model choice to approximate the posterior probability of all possible ways the taxa can be partitioned into event time categories. Each possible partitioning of the taxa is a unique model of evolutionary events, or "event model." To do this, we must specify the probability of each possible event model a priori.
For more background about ecoevolity, please see http://phyletica.org/ecoevolity/background.html.
Initially, ecoevolity, used a Dirichlet process (DP) as a mathematically convenient and flexible way to assign the prior probability of every possible event model. We have since implemented a Pitman-Yor process (PYP) prior, which is a generalization of the Dirichlet process. We also implemented a prior that by default assigns equal prior probability to all possible event models, which we'll refer to as a uniform prior. However, this "uniform" prior has a "split weight" (SW) parameter to allow classes of event models with more event time categories to be favored (or disfavored) a priori.
The goal of this project is to compare the performance of these three priors on event models (DP, PYP, and SW) using simulations.
This work was made possible by funding provided to Jamie Oaks from the National Science Foundation (grant numbers DBI 1308885 and DEB 1656004).
This work is licensed under a Creative Commons Attribution 4.0 International License.
