Fig. 3: Simulated versus experimentally measured bias.
From: Behavior engineering using quantitative reinforcement learning models
a The mean bias for each of the reward schedules, as predicted by the CATIE model versus the mean experimentally measured bias (see mean bias and number of participants in Supplementary Data 1, and full bias distribution in Fig. S2). Dashed line is the diagonal. Error bars are the standard error of the means, centered around the mean bias. b The high correlation between CATIE’s predicted bias and the empirical bias shown in (a) is quantified by the orange bar and compared with the correlation between the empirical bias and the bias predicted by other models used for engineering (see also Fig. S3). Error bars represent the standard error of the mean, centered around the mean correlation coefficient. Individual points show the correlation coefficients in single simulations. In both panels, the simulated standard error was computed by averaging over 1000 simulations, each with the same number of participants as in the competition. In both panels, the color scheme is identical to Fig. 2.
