Meeting Abstract
It has long been understood that phylogenetic relationships must be accounted for in the analysis of comparative data due to non-independence resulting from common ancestry. However, most available analysis options apply only to linear relationships between variables. To expand upon these approaches, we developed a nonlinear phylogenetic regression model. Building upon a widely used linear phylogenetic least squares regression model (PGLS), we developed a three-parameter nonlinear model that fits a sigmoidal logistic function while accounting for phylogenetic signal in the residuals. Unlike already available logistic regression approaches, ours handles continuous, as opposed to binary, response variables. We conducted a power analysis of this model using simulated data to determine the model’s ability to detect sigmoidal patterns at varying species sample sizes (n = 15, 30, 50, 100, 200), variance in the error of the data (SD = 0.5, 1, 2), and slopes at the inflection point of the logistic function (slope = 5, 10, 15, 25). We also quantified the type I error, or the rate at which the model produces false positive results. The nonlinear phylogenetic regression model provides good power in most situations, but drops notably when the number of species falls below 30 with increasing error in the data. This approach can also be extended to any other nonlinear model to allow for greater flexibility in analyzing complex comparative datasets with nonlinear relationships.
