ABSTRACT
Instrumental variables are natural experiments that impersonate randomized experiments. Phylogenetic distance is a proxy for unobserved variables that generate covariation among species, even when selection matters. Phylogeny is associated with the patterns of covariation across species, because recently diverged species tend to be more similar. The original phylogenetic regression approach treats phylogenetic distance in a highly constrained and unrealistic way, based on a neutral model of divergence with time. The Brownian motion model is a special kind of Gaussian process in which the covariance declines in a very rigid way with increasing distance. There is no need to be so rigid and good reason to think evolution is not well-described by Brownian motion.
