Fig. 2: Simulations and behavioral results.
From: Mouse tracking reveals structure knowledge in the absence of model-based choice
a Example simulation of a purely model-free agent (w = 0) in the stochastic transition task: the probability to stay with the same first-stage choice increases with the previous trial’s reward, but does not depend on the previous transition (common (dark blue circles) or rare (light blue diamonds)). b Example simulation of a purely model-based agent (w = 1) in the stochastic transition task: the probability to stay with the same first-stage choice decreases with the previous trial’s reward if the previous transition was rare. c Probability to stay with the same first-stage option in the stochastic transition task: subjects display very little model-based behavior (N = 57 individual subjects). d Model-based weight w by condition (block type): 0.6–0.9 correspond to the common transition probability (stochastic condition, in blue), and 1–4 denote the pattern type (deterministic condition, in red). e Correlation in model-based weight w between stochastic and deterministic conditions. The weights are calculated as the averages of block-wise estimates. f Correlation in reward rate between stochastic and deterministic conditions. g Reward rate by conditions (block type) and pattern difficulty (N = 57 individual subjects). h–i Correlation between the model-based w and reward rate in the stochastic (h) and deterministic (i) conditions. All correlation plots show Pearson correlations, each individual point is one subject, and dotted lines indicate linear regression fits. In bar plots: dots denote individual subjects, error bars denote s.e.m. at the subject level. Source data are provided as a Source Data file.
