Revisiting phylogenetic signal; strong or negligible impacts of polytomies and branch length information?

Abstract

Background: Inaccurate estimates of phylogenetic signal may mislead interpretations of many ecological and evolutionary processes, and hence understanding where potential sources of uncertainty may lay has become a priority for comparative studies. Importantly, the sensitivity of phylogenetic signal indices and their associated statistical tests to incompletely resolved phylogenies and suboptimal branch-length information has been only partially investigated. Methods: Here, we use simulations of trait evolution along phylogenetic trees to assess whether incompletely resolved phylogenies (polytomic chronograms) and phylogenies with suboptimal branch-length information (pseudo-chronograms) could produce directional biases in significance tests (p-values) associated with Blomberg et al.'s K and Pagel's lambda (λ) statistics, two of the most widely used indices to measure and test phylogenetic signal. Specifically, we conducted pairwise comparisons between the p-values resulted from the use of "true" chronograms and their degraded counterparts (i.e. polytomic chronograms and pseudo-chronograms), and computed the frequency with which the null hypothesis of no phylogenetic signal was accepted using "true" chronograms but rejected when using their degraded counterparts (type I bias) and vice versa (type II bias). Results: We found that the use of polytomic chronograms in combination with Blomberg et al.'s K resulted in both, clearly inflated estimates of phylogenetic signal and moderate levels of type I and II biases. More importantly, pseudo-chronograms led to high rates of type I biases. In contrast, Pagel's λ was strongly robust to either incompletely resolved phylogenies and suboptimal branch-length information. Conclusions: Our results suggest that pseudo-chronograms can lead to strong overestimation of phylogenetic signal when using Blomberg et al.'s K (i.e. high rates of type I biases), while polytomies may be a minor concern given other sources of uncertainty. In contrast, Pagel's λ seems strongly robust to either incompletely resolved phylogenies and suboptimal branch-length information. Hence, Pagel's λ may be a more appropriate alternative over Blomberg et al.'s K to measure and test phylogenetic signal in most ecologically relevant traits when phylogenetic information is incomplete.