Abstract
ABSTRACT Within-species variation may be the result of genetic variation, environmental variation or measurement error for example. In phylogenetic comparative studies, failing to account for intraspecific variation has many adverse effects, such as increased error to test hypotheses about evolutionary correlations, biased estimates of evolutionary rates, and inaccurate inference of the mode of evolution. These adverse effects were demonstrated in studies that considered a tree-like underlying phylogeny. Comparative methods on phylogenetic networks are still in their infancy. The impact of within-species variation on network-based methods has not been studied. Here, we extend the phylogenetic linear model available in the Julia package PhyloNetworks to account for within-species variation in the continuous response trait, assuming equal intraspecific variances across species. We show how inference based on the individual values can be reduced to a problem using species-level summaries, even when the within-species variance is estimated. Using simulations, we find that our method performs well under various settings, and is robust when intraspecific variances are unequal across species. When phenotypic correlations differ from evolutionary correlations, we find that estimates of evolutionary coefficients are pulled towards the phenotypic coefficients, for all methods we tested. Also, evolutionary rates are either underestimated or over-estimated, depending on the mismatch between phenotypic and evolutionary relationships. We applied our method to morphological and geographical data from Polemonium . We find a strong negative correlation of leaflet size with elevation, despite a positive correlation within species. Our method can explore the role of gene flow in trait evolution. We find marginal evidence for leaflet size being carried along in gene flow, supporting previous observations on the challenges of using individual continuous traits to infer inheritance weights at reticulations.