Fig. 2. RL modeling of choice and reward traces.
Illustration of choice-location trace (CL-trace, a), choice stimulus traces (CS-trace, different colors indicate different choice stimuli, b) and reward trace (R-trace, c). d Model architecture. Black colors indicate components used by all models. The other colors refer to choice and reward traces as illustrated in panels a–c. Note that CL-trace and CS-trace effects are added directly to the DV of the model, whereas R-trace affects the prediction error calculation. e Nested model comparison using BICint (lower values indicate better fit). The full model including R-trace explains the data best. x-labels are the same for panels e and f. f Exceedance probability also favors the full model among our set of candidate models. The dashed red line indicates an exceedance probability of 0.95. g–i Experiment specific comparisons of Bayes factors between full model and second-best model. Bars are individual sessions and direction of the bars (left vs right) indicate better model fit in favor of the full (going to the right) compared to the second-best model (going to the left). Colors indicate individual monkeys. (n = 65 sessions). Source data are provided as Source Data file.