Fig 2. Decreases in the RSS from each lower order polynomial model to each higher order polynomial model are shown as a percentage of the lower order model (normalized to 100%).

If a higher order model retains less than 10% of variation from the lower order model, it is accepted, otherwise it is rejected. Within each panel of the figure, species with solid lines are evaluated to determine if they would be overfit by the subsequent model. In the first panel, all lines are solid because all species are evaluated. The species pig, rat, and mouse are determined to be well fit by the model y = a₂x² and overfit by the model y = a₃x³+a₂x². These three species are shown as dotted lines in the subsequent panel. The remaining seven species are shown as solid lines. These seven species are well fit by the model y = a₃x³+a₂x² but overfit by y = a₄x⁴+a₃x³+a₂x² and therefore shown as dotted lines in the third panel.