Fig. 8

Multiple linear regression models for grouped variables initially selected for Se and As, with adjusted R² values along the y-axis. The best model was computed for each subset size (e.g., each amount of predictor variables utilized in the overall model). n = 25 because of some samples missing predictive variable measurements and thus being omitted, which also explains the slight discrepancy between multiple linear regression analyses completed with more samples (i.e., Additional file 1: Table A4)