Fig. 4: Model comparison.
From: Behavior engineering using quantitative reinforcement learning models
The 333,200 decision probabilities (collapsed across all trials and participants) as predicted by 5 models. a Average choice probability, E[p] and b Log-likelihood, E[log(p)]. While CATIE (orange bar) was the best model with respect to average choice probability, a QL model (schedule 10, dark purple) was superior with respect to log-likelihood. Error bars are the standard error of the mean. Bar color scheme is identical to Fig. 2. See full distributions in Fig. S4. Complete statistical tests can be found in Tables S1 and S2.
