Fig. 2: Competition results.
From: Behavior engineering using quantitative reinforcement learning models
The bias, average proportion of choices of Alternative 1 for the different reward schedules (see schedules and their description in Fig. S1 and Supplementary Data 1). The winner of the competition (schedule 1, orange) was designed by a choice engineer who utilized the CATIE model, achieving an average bias of 64.3%. Noteworthy are also the results of schedules optimized using a QL model with different sets of parameters (schedules 6, 7, 9, and 10; same color, blue or purple, represent schedules that utilized parameters from the same dataset, see Supplementary Data 1). The number of participants (data points) per schedule is different (see “Methods” section), ranging from n = 87 for schedule 11 to n = 595 for Schedule 1. Error bars are the standard error of the mean, centered around the mean. Bias of individual participants is represented by single data points, with identical values stacked from left to right. Data of schedules 2–5, 8, and 11 are arbitrarily represented in green, with brighter shades representing better performance.
