Fig. 2: Additive heuristic model.

A Additive Heuristic Model parameterizes use of reward and probability information. The probability information component measures the difference in probability between reaching the better (higher reward) versus worse gamble outcome, contingent on accepting the choice stimulus. The reward information component measures the difference in reward associated with the midpoint between the gamble and safe reward. Note that because of the actual reward used in the task (Fig. 1C) this difference can be computed by considering the trigger and safe rewards, without needing to refer to the non-trigger reward. R* refers to the reward after the non-trigger reward (which simply amounts to common noise along with noise specific to that outcome) has been subtracted from all rewards. Working with R*, the difference between the gamble midpoint and safe reward is computed by dividing the trigger reward by two and subtracting the safe reward. B Additive Heuristic Model captures aggregate patterns in choice data. Each data error bar (gray) shows the across-participant (n = 19) mean (+/− s.e.m.) proportion acceptance for each combination of whether a trial is gain or loss (row), whether the safe reward is higher or lower than the midpoint between the two gamble rewards (column) and the trigger outcome probability contingent on acceptance (x-axis). Values reflect outcome rewards prior to adding common and other noise. Blue and orange lines show predictions of the additive heuristic model and expected value models, at best fit parameters. C Additive Heuristic Model indexes individual differences in weighting of probability and reward information. Left) Model-predictions for individual participants that either rely exclusively on probability information (left) or reward information (right). Right) Parameterization of reward and Probability Weighting (\({{{{{{\rm{\beta }}}}}}}_{{{{{{\rm{Prob}}}}}}}\) and \({{{{{{\rm{\beta }}}}}}}_{{{{{{\rm{rew}}}}}}}\)) place these two participants at extreme ends of continuum over which participants (n = 19) vary in these two strategies.