Controlling for body size leads to inferential biases in the biological sciences

Many traits correlate with body size. Studies that seek to uncover the ecological factors that drive evolutionary responses in traits typically examine these responses relative to associated changes in body size using multiple regression analysis. However, it is not well appreciated that in the presence of strongly correlated variables, the partial (i.e., relative) regression coefficients often change sign compared to the original coefficients. Such sign reversals are difficult to interpret in a biologically meaningful way, and could lead to erroneous evolutionary inferences if the true mechanism underlying the sign reversal differed from the proposed mechanism. Here, we use simulations to demonstrate that sign reversal occurs over a wide range of parameter values common in the biological sciences. Further, as a case-in-point, we review the literature on brain size evolution; a field that explores how ecological traits relate to the evolution of relative brain size (brain size relative to body size). We find that most studies show sign reversals and thus that the inferences of many studies in this field may be inconclusive. Finally, we propose some approaches to mitigating this issue.