Article Figures & Data
Figures
- Figure 1.
Experimental task. Participants chose from among four differently colored bandits, which yielded a positive reward between 0 and 100. Payoff varied according to a slowly drifting pseudorandom walk, fixed across subjects (inset, solid line), with added Gaussian noise, different across subjects, of constant SD (inset, shaded region).
- Figure 2.
Lesion overlap in the FP group. Lesions were predominantly right sided and included cortex and white matter outside the FPC; all but one patient had bilateral involvement.
- Figure 3.
Conditional logit (CL) model. Weights in the CL model show the influence of past reward on selection probability, where the choice value of each bandit is given by a weighted sum over recent payoffs from the bandit. Significance levels under the Wilcoxon rank-sum test (H0, zero median) or Kruskal–Wallis tests (H0, equal medians) are indicated (*p < 0.01, otherwise as shown for p < 0.05). Significant group effects are shown (horizontal bar).
- Figure 4.
Separate additive contributions of RL and LD terms to weighing of payoffs as a function of selection lag for each group. Points represent individual fits, and lines indicate median weighting across individuals. Top, RL weighting. Middle, LD weighting. Bottom, Sum RL+LD weighting. Note the absence of a contribution of the LD term in group FP. Results do not change with the exclusion of the outlier in group FP.
- Figure 5.
Mean change in reward (difference between the lag-1 and lag-2 payoffs, Δreward) for stay and switch trials. Mean change is shown for trials in which the same choice was repeated (stay trials) and in which a new bandit was chosen (switch trials). For switch trials, Δreward is shown both for the selected bandit (switch, lag-0 choice) and for the previously selected bandit (switch, lag-1 choice). Significance of group effects in each category is indicated with uncorrected p values. Overall significance for the group effect was p = 0.046 (repeated-measures ANOVA).
Tables
Group Gender Age (years) Ed VIQ PIQ FS-IQ FPC 6 M/2 F 52 (15) 14 (3.4) 114 (20) 105 (19) 111 (19) BDC 3 M/5 F 45 (18) 14 (2) 100 (8) 106 (8) 103 (7) NC 4 M/10 F 53 (10) 16 (1.8) 110 (10) 108 (8) 110 (8) Values age given as mean (SD). All participants were fully right handed. M, Male; F, female; VIQ, verbal IQ; PIQ, performance IQ. IQ data were obtained from the WAIS-III (FP, BDC) or estimated from the NART, WRAT-R or abbreviated WAIS-III (NC). Ed, Years of education.
Parameter Description FPC BDC NC βV RL inverse temperature 0.33 (0.36) 0.53 (0.38) 0.45 (0.39) α RL learning rate 0.52 (0.46) 0.35 (0.48) 0.43 (0.45) βC Recency inverse temperature 0.75 (1.1) 0.66 (0.89) 0.63 (1.31) κ Recency decay 0.40 (0.34) 0.14 (0.55) 0.42 (0.39) βD Lag difference −0.008 (0.015) 0.034 (0.026) 0.026 (0.023) Mean (SD) is shown for parameter values within each participant group.
