Fig. 6.

Model comparison demonstrates that observational and instructed threat learning rest on distinct computational mechanisms. (A) Observational threat learning (experiment 2). Model comparison showed that the competing systems model explained the results of experiment 2 better than the alternative social learning models (derived from the literature), which did not posit a competition between systems. (B) Instructed threat learning (experiment 3). In contrast to observational threat learning (experiment 2), model comparison showed that a model where instruction functions as a prior on decision making explained the results of experiment 3 better than alternative social learning models (SI Appendix). Akaike Information Criterion (AIC) weights (wAIC) can be interpreted as the probability that the model provides the best explanation of the data in the candidate set (see SI Appendix for results based on Bayesian random effects model comparison). The dotted green line denotes wAIC = 0.95. IL-D, instructed learning D model.