Numbers denote the probability that data generated with model X are best fit by model Y, thus the confusion matrix represents . (A) When there are relatively large amounts of noise in the models (possibility of small values for and ), models 3–5 are hard to distinguish from one another. (B) When there is less noise in the models (i.e. minimum value of and is 1), the models are much easier to identify. (C) The inversion matrix provides easier interpretation of fitting results when the true model is unknown. For example, the confusion matrix indicates that M1 is always perfectly recovered, while M5 is only recovered 30% of the time. By contrast, the inversion matrix shows that if M1 is the best fitting model, our confidence that it generated the data is low (54%), but if M5 is the best fitting model, our confidence that it did generate the data is high (97%). (D) Similar results with less noise in simulations.