Figure 5.

Simulated reward earnings for various regimes of MBCA parameters

(A) Average standardized earning for simulated groups of pure-MB agents. A standard agent duplicated the empirical MBCA parameters of its yoked real participant from the standardized format across all three formats. Similarly, prospective- and retrospective inference agents duplicated MBCA from the prospective inference and retrospective inference presentation formats, respectively (see STAR Methods for full details). Error bars correspond to SEM across experiments calculated separately in each format.

(B) Standardized earnings (ordinate) of simulated pure-MB agents (STAR Methods) are displayed as a function of (1) the total weight of MBCA in the retrospective inference format summed across observed and inferred outcomes (color gradient, increasing from black to yellow in steps of 0.3, with some values marked on the right side), and (2) the fraction of that total that is allocated to the inferred outcome (x axis). Earnings are maximized when the ratio is 0.5, i.e., when MBCA for the inferred outcome is identical to MBCA for the observed outcome (for none of the curves was there a point significantly higher than the central point). Each data point is based on 10,000 simulations of synthetic experimental sessions (see STAR Methods for full details).