Figure 1. Comparative assessment of statistical methods to quantify variable importance with simulated data (see script in Fig. S1). When one variable (X2 in Panel A) is strongly correlated with the other two (X3 and X4 in Pannel A), the beta weights from the multiple regression approach (Panel B) may overestimate the variable importance of X2. Hierarchical partitioning (Panel C) and Random forests analyses (Panels D and E) can cope with the multicollinearity among variables and the variable importance of X2 decreases. Simulated data were modified from Ray-Mukherjee et al. (2014).