PhyBaSE: A Bayesian structural equation model approach to causal inference in phylogenetic comparative analyses

Abstract

One of the main limitations of phylogenetic comparative analyses is that associations between traits can only be interpreted as correlations. Here, we present a novel Bayesian structural equation model (PhyBaSE) which allows us to disentangle direct from indirect relationships among variables to propose potential causal hypotheses while accounting for phylogenetic non‐independence. Compared with the existing maximum‐likelihood based approach, PhyBaSE models are more flexible, allowing the inclusion of trait and phylogenetic uncertainty, as well as non‐continuous variables. To facilitate the application of the method, we provide worked examples, data and code. We exemplify the method both with simulated as well as empirical data. Our analyses with simulated data indicate that PhyBaSE models have higher power than classic Phylogenetic Path Analysis to discriminate between competing models. As an example of PhyBaSE using empirical data, we revisit different hypotheses proposed to explain the relationship between relative brain size and group size in Bovids. Our results challenge the previously supported social brain hypothesis and provide support for an allometric effect of body size on social group size and an effect of brain size on life span, as predicted by the cognitive buffer hypothesis. The flexibility of PhyBaSE models will allow researchers to explore more complex hypotheses on the evolution of behavioural, ecological and life history traits at a macroevolutionary level and how these are linked to anthropogenic drivers of biodiversity loss and extinction, taking full advantage of the increasing number of publicly available species‐specific datasets.