Same as Fig. C.1 for the non-nested model selection between wabs and torus (instead of powerlaw and wabs). As the differences between the models are subtle, the likelihood-ratio method often remains with the null model (wabs). This yields an overall error rate (red line) of ~30−50%. The Bayesian model selection declares the data insufficient for a distinction in such cases. In the upper right panel, the Bayesian method is able to effectively distinguish between some of the simulated instances, and makes few mistakes (<1% false selection rate, considering the total of panel pairs).