A critical comment on the ‘multiple variance Brownian motion’ model of Smaers et al. (2016)

Abstract

Smaers, Mongle & Kandler (2016) (Biological Journal of the Linnean Society, 118: 78-94) introduced a new phylogenetic comparative method, multiple variance Brownian motion (mvBM), for reconstructing ancestral states given a phylogenetic tree and continuous trait data. The authors conducted a simulation study and argued that mvBM outperforms constant variance Brownian motion (BM) when rates of evolution vary across the phylogeny. In this comment, we argue that mvBM is not a viable statistical method because it is fundamentally a circular analysis that overfits phylogenetic branch lengths to the data. We further argue that the comparison of mvBM to BM under conditions where the assumptions of BM are clearly violated is not an informative performance analysis, and that the simulation study of Smaers et al. (2016) exaggerates the performance of mvBM by focusing on a narrow range of simulation conditions and reporting aggregated accuracy metrics that obscure severe inaccuracy and bias in its ancestral state estimates. Our arguments are supported by simulation results. We conclude that mvBM is not a viable phylogenetic comparative method.