Fig 5. n = 27 subjects.

A) Model comparison. Comparison of the combined Preference+Value model to models that learn only absolute values or relative preferences. The models are compared by means of the Bayesian Information Criterion (BIC; [36]). Lower BIC values indicate a more parsimonious model fit. B) Modeled Utilization of Value and Preference Learning. Individual subject model parameter fits showing the effects of concurrent and cumulative diversity on the degree to which preference and value learning manifested in subjects’ choices. Each dot represents a subject. Dashed lines mark where utilization of the form of learning is equal for low and high diversity.