Realistic scenarios of missing taxa in phylogenetic comparative methods and their effects on model selection and parameter estimation

Abstract

Model-based analyses of continuous trait evolution enable rich evolutionary insight. These analyses require a phylogenetic tree and a vector of trait values for the tree’s terminal taxa, but rarely do a tree and dataset include all taxa within a clade. Because the probability that a taxon is included in a dataset depends on ecological traits that have phylogenetic signal, missing taxa in real datasets should be expected to be phylogenetically clumped or correlated to the modelled trait. I examined whether those types of missing taxa represent a problem for model selection and parameter estimation. I simulated univariate traits under a suite of Brownian Motion and Ornstein-Uhlenbeck models, and assessed the performance of model selection and parameter estimation under absent, random, clumped or correlated missing taxa. I found that those analyses perform well under almost all scenarios, including situations with very sparsely sampled phylogenies. The only notable biases I detected were in parameter estimation under a very high percentage (90%) of correlated missing taxa. My results offer a degree of reassurance for studies of continuous trait evolution with missing taxa, but the problem of missing taxa in phylogenetic comparative methods still demands much further investigation. The framework I have described here might provide a starting point for future work.

Anthropogenic extinctions conceal widespread evolution of flightlessness in birds

“…Extant species make up 95% of bird species in our study, but strong biases still arise from the imbalance in trait distributions between extant and extinct species. These results add further insight into the general problem of sampling biases in comparative phylogenetic analyses (37,38) and highlight the need for better integration of paleontological, ecological, and evolutionary studies. Despite previous calls to restrict macroevolutionary analysis to species with genetic information (39), doing so could potentially lead to a greater bias if the excluded species represent a nonrandom sample of the total.…”

Female‐driven intersexual coevolution in beetle genitalia

“…Our taxon sampling is evenly distributed across the beetle tree of life, ensuring trait-state configurations (Table 1) are representative of the real distribution of these traits in nature. This is important because trait modelling approaches tend to be more robust when taxon sampling is evenly distributed across a phylogeny, even if this sampling is highly sparse (Marcondes, 2019). It is also noteworthy that rare trait-state combinations occur in nature and can reflect evolutionary instability of some trait configurations.…”

Historical and future climate change fosters expansion of Australian harvester termites, Drepanotermes

“…Estimates of phylogenetic signals can be distorted due to incomplete sampling of species and/or taxonomic oversplitting (Losos 2008; but see also Marcondes 2019). In both cases, levels of phylogenetic signal may be inflated because too many ecologically similar species cluster together in reconstructed phylogenies.…”