Model validation.

A random effects test of the models versus a baseline model was performed by simulating random expected value estimates (10,000 repetitions) and then computing a non-parametric p-value per subject as the fraction of repetitions in which the baseline BIC is lower than the model BIC (indicating a better fit). These p-values were then combined across subjects using Fisher's combined probability test. Only HMM outperforms the baseline model in both the appetitive and aversive sessions.