Phylogenetic comparative methods are often used to test functional relationships between traits. However, million-year macroevolutionary observational datasets cannot definitively prove causal links between traits --- correlation does not equal causation and experimental manipulation over such timescales is impossible. While this caveat is widely understood, it is far less appreciated that different phylogenetic approaches make different causal assumptions about the functional relationships of traits. In order to make meaningful inferences, it is critical that our statistical methods make biologically reasonable assumptions. Here we illustrate the importance of causal reasoning in comparative biology by examining a recent study by Avaria-Llautureo et al. (2019) that tested for the evolutionary coupling of metabolic rate and body temperature across endotherms and made the notable discoveries that these traits were unlinked through evolutionary time and that body temperatures were, on average, higher in the early Cenozoic than they are today. We argue that the causal assumptions embedded into their models made it impossible for them to actually test the relevant functional and evolutionary hypothesis. We then re-analyze their data using more biologically appropriate models and find support for the exact opposite conclusions, corroborating previous evidence from physiology and paleontology. We highlight the vital need for causal thinking, even when experiments are impossible. 

This dataset includes scripts and data from the original cited literature, as well as simulated data with the generating code.

Supplementary Text and supplementary Figure 1 are available in the SupplementaryMaterial.* files. Version history, including up to date code, is available at the git repository: https://github.com/uyedaj/Tb_ALEA.

Supplementary Figure 1. Simulated values of contrast correlations from phylogenetically-distributed rate scalars demonstrates reasonable power to detect significant relationships between rates directly from contrasts in the presence of rate variation. Open circles indicate that the contrast Pearson-correlation coefficient was non-significant, whereas filled black circles indicate it is significantly correlated in X and Y. Red outlines indicate that the true correlation coefficient among branch rates are significant, whereas black outlines indicate even with perfect knowledge of the branch rates, the obesrved correlation was non-significant. Observed values from the empirical data are shown, while note that the data are simulated to match the Mammalian, not bird data.

Usage information and a description of the contents of the repository, and the data sources used, is provided in "README.txt".