Fig. 3. The change in R² (A) and Akaike weights, _i (B, solid line) and the frequency different variables were included in the best-subsets model (for 1000 bootstrap iterations) as the number of parameters (k) included in the model is increased (C). The error and intercept terms make up the first two parameters of the model, and total length (TL) was forced as the first variable included (in order to account for size). The bars in B show the mean Akaike weight for each model size for 10 000 random iterations ± 1 s.e.m.