Continuous traits and speciation rates: Alternatives to state‐dependent diversification models

Abstract

Many quantitative traits, for example body size, have been hypothesized to influence the diversification dynamics of lineages over macroevolutionary time‐scales. The Quantitative State Speciation‐Extinction (QuaSSE) model and related methods provide an elegant framework for jointly modelling the relationship between change in continuous traits and diversification. However, model misspecification and phylogenetic pseudoreplication can result in elevated false discovery rates in this and other state‐dependent speciation‐extinction models. Here, we evaluate alternative trait‐dependent diversification methods that do not formally model the relationship between traits and diversification, but instead test for correlations between summary statistics of phylogenetic branching patterns and trait variation at the tips of a phylogenetic tree (hereafter tip‐rate correlations or TRCs). We compare alternative branching pattern statistics and significance tests, and we evaluate their performance relative to QuaSSE under a range of evolutionary scenarios. We found that a simple statistic derived from branch lengths (inverse equal splits) can detect trait‐associated rate variation, and that a simulation‐based method performs better than phylogenetic generalized least squares for testing the significance of trait‐rate correlations. This test (ES‐sim) had better power to detect trait‐dependent diversification than other TRCs. By testing the approach across a diverse set of simulation scenarios, we found that ES‐sim is similar to QuaSSE in statistical power. However, the approach rarely led to false inferences of trait‐dependent diversification, even under conditions that are problematic for formal state‐dependent models. We illustrate the application of ES‐sim to real data by re‐assessing the relationship between dispersal ability and diversification in Furnariid birds. We conclude that simple, semi‐parametric tests like ES‐sim represent a promising approach for trait‐dependent diversification analyses in groups with heterogeneous diversification histories and provide a useful alternative or complement to formal state‐dependent speciation‐extinction models.

Larger brains spur species diversification in birds

“…Because this method can sometimes detect false positives (Harvey and Rabosky 2018), we compared the outcome of this model with a model considering a randomly generated trait instead of brain size. We first used a Quantitative-state speciation extinction model (QuaSSE) that allows explicitly testing whether speciation and extinction rates have been influenced by a quantitative trait (in this case, relative brain size) along the phylogenetic tree.…”

“…This model is based on a birth and death process where parameters can depend on the state of the trait (FitzJohn 2012). Because this method can sometimes detect false positives (Harvey and Rabosky 2018), we compared the outcome of this model with a model considering a randomly generated trait instead of brain size. To further test for results consistency and robustness, we also used an alternative method: we built phylogenetic linear mixed models to test the effect of relative brain size on a DR metric that summarizes all the splitting events along the tree (Redding and Mooers 2006;Jetz et al 2012) and that reliably estimates speciation rates (Title and Rabosky 2018).…”

Corrigendum for Rojas et al. (2018) DOI: 10.1111/ele.12911

“…Another method, the quantitative state speciation and extinction approach (QuaSSE; FitzJohn ) tests speciation rate as different explicit functions of the trait states. Limitations of these methods have been analyzed elsewhere (e.g., Harvey & Rabosky ). For example, STRAPP requires large phylogenies to detect significant associations between traits and speciation rates (Rabosky & Huang ), while QuaSSE can lead to spurious correlations if there are different macroevolutionary regimes in the phylogeny (Rojas et al .…”

“…Firstly, we used the ES‐sim test from Harvey & Rabosky () to analyze the correlation between a metric of tip‐rate of speciation – the inverse equal splits (Redding & Mooers ) – and marginality. The ES‐sim method is a semi‐parametric method in which the observed correlation between the inverse equal splits and the trait is tested against a null set of associations (we used 10,000).…”

“…Although ES‐sim is less biased towards false positives than other trait‐dependent diversification methods (Harvey & Rabosky ), it cannot be used to analyze different macroevolutionary regimes within a tree. If we were able to prune the tree to generate the different subtrees that correspond to the regimes, and then perform ES‐sim on each subtree for comparison, we will need more tips per subtree to achieve a high statistical power (e.g., 0.38 for 50 tips vs. 0.93 for 250 tips) (Harvey & Rabosky ). This is not always feasible.…”