ABSTRACT
Our current understanding of how species have evolved is mainly based on comparative phylogenetic methods, which use phylogenies to infer the evolution of traits. The development of ancestral state reconstruction (ASR) methods has provided the tools to reconstruct trait evolution, which are widely used in fields like evolutionary biology, macroecology and paleontology. As there are different elements involved in those analyses, with different levels of uncertainty (i.e. relating to branch length estimation, trait coding, statistical framework, taxon sampling or software), the various combinations of these elements likely have a strong impact on the reconstruction of the evolution of traits, potentially leading to opposite conclusions. To assess the impact of these different elements in ASR, we performed a set of analyses, including all possible combinations of such elements and using the evolution of migratory behavior in Sylvia warblers as a case study, which was coded as a continuous or as a discrete character. Our results show that taxon sampling, character coding, tree shape, statistical framework and software all significantly affect ASR, both individually and in combination. Not all reconstructed tree nodes show discrepancies, but in the critical ones most pairwise comparisons between analyses lead to conflicting and unexpectedly antagonistic results (zero migration vs fully migratory), thus challenging interpretations of trait evolution. We propose some possible solutions to partly inform decisions, involving the method selection and the incorporation of biological or fossil evidence regarding how traits evolve, but our results demand serious rethinking about how the research community currently uses ASR.
