Analyses across plant families
First, we used phylogenetic principal components analysis (pPCA) with scaling to ordinate and derive two orthogonal variables, accounting for the most variation among them (Fig. 2, see Supporting information). We then fit linear models of ERR slopes as a function of each variable (Table 1), as well as the first two principal components at the family level, using phylogenetic independent contrasts (PIC). For across-family analyses, all variables were transformed into z-scores to facilitate comparisons of factors at different scales (reported in Table 1), meaning results remain consistent (adj. R2,F -statistic, and P -values), but the intercept becomes zero and slopes of said relationships can be described as positive (slope of 1), negative (slope of -1), or neutral (0). Note, we did not use scaling for reporting individual family ERR slopes, or for visual representation of scatterplot axes. For extended methods, see Supporting information.