Fig. 3: Comparison of parameter exploration efficiencies between STACEY and StarBeast3, under the birth-collapse tree prior.
From: Quantitatively defining species boundaries with more efficiency and more biological realism
The average effective samples generated per hour ( ± 1.96 se) are plotted for each term. Top of each plot: the mean relative difference compares StarBeast3 with STACEY. These terms are coloured orange if StarBeast3 outperformed (and purple if STACEY outperformed) at a significant level across 20 MCMC replicates (p < 0.05 from a two-sided t test). All means and standard errors were computed in log-space. We evaluated ESS/hr across an array of parameters, including general inference, species tree parameters, and parameters describing gene trees and substitution models. Notation -- p(θ∣D): posterior density; p(D∣θ): likelihood; p(θ): prior density; hS: species tree height; lS: species tree length; λ: species tree birth rate; ω: species tree collapse weight; \({{{{{{{\mathcal{O}}}}}}}}\): species tree origin; hG: gene tree height; lG: gene tree length; μN: mean effective population size; κ: gene tree transition-transversion ratio; f: gene tree nucleotide frequencies; ν: gene tree substitution rate; min: minimum ESS/hr across all other terms.
