Abstract
All models are wrong and sometimes even the best of a set of models is useless. Modern phylogenetic comparative methods (PCMs) are almost exclusively model-based and therefore making robust inferences from PCMs requires using a model of trait evolution that is a good explanation for the data. To date, researchers using PCMs have evaluated the explanatory power of a model only in terms of relative, not absolute, fit. Here we develop a general statistical framework for assessing the absolute fit, or adequacy, of phylogenetic models for the evolution of quantitative traits. We use our approach to test whether commonly used models are adequate de scriptors of the macroevolutionary dynamics of real comparative data. We fit models of trait evolution to 337 comparative datasets covering three key Angiosperm functional traits and evaluated the absolute fit of the models to each dataset. Overall, the models we used are very inadequate for the evolution of these traits; this was true for many different groups and at many different scales. Furthermore, the relative support for a model had very little to do with its absolute adequacy. We argue that assessing model adequacy should be a key step in comparative analyses.
There are known knowns; there are things we know we know. We also know there are known unknowns; that is to say we know there are some things we do not know. But there are also unknown unknowns — the ones we don’t know we don’t know.
— Former U.S. Secretary of Defense, Donald Rumsfeld
