Relation of perseveration estimation with stay probability.

The left panels show the trial-by-trial RL κ perseveration parameter estimation using t-RNN (blue) and QP-stationarity (green) methods, along with the moving average calculation of the stay probabilities (red; window size of 10 trials) for three example subjects (one from each diagnostic group). The right panels show the corresponding Pearson correlation between the moving average stay probabilities and κ parameter estimation of t-RNN and QP-stationarity model (red dashed line denotes the identity). The results indicate a strong correlation between the stay probabilities and κ parameter estimation of t-RNN (r² > 0.9), but not of the QP-stationarity (r ≈ 0), which fails to detect changes in subject behavior throughout the task.

(PDF)