Using BayesModelS to provide Bayesian- and phylogenetically-informed primate body mass predictions

Abstract

An accurate prediction of the body mass of an extinct species can greatly inform the reconstruction of that species' ecology. Therefore, paleontologists frequently predict the body mass of extinct taxa from fossilized materials, particularly dental dimensions. Body mass prediction has traditionally been performed in a frequentist statistical framework, and accounting for phylogenetic relationships while calibrating prediction models has only recently become more commonplace. In this article, we apply BayesModelS—a phylogenetically informed Bayesian prediction method—to predict body mass in a sample of 49 euarchontan species (24 strepsirrhines, 20 platyrrhines, 3 tarsiids, 1 dermopteran, and 1 scandentian) and compare this approach's body mass prediction accuracy with other commonly used techniques, namely ordinary least squares, phylogenetic generalized least squares, and phylogenetic independent contrasts (PICs). When predicting the body masses of extant euarchontans from dental and postcranial variables, BayesModelS and PICs have substantially higher predictive accuracy than ordinary least squares and phylogenetic generalized least squares. The improved performances of BayesModelS and PIC are most evident for dentally derived body mass proxies or when body mass proxies have high degrees of phylogenetic covariance. Predicted values generated by BayesModelS and PIC methods also show less variance across body mass proxies when applied to the Eocene adapiform Notharctus tenebrosus. These more explicitly phylogenetically based methods should prove useful for predicting body mass in a paleontological context, and we provide executive scripts for both BayesModelS and PIC to increase ease of application.