Article Figures & Data
Figures
- Figure 1.
Risk prediction error. A, B, Task design. An initial ocular fixation spot predicts the overall risk, derived from the overall reward probability distribution (maroon arrow in B). The subsequent appearance of one of the risk cues defines the current risk for each trial. Each risk cue predicts an equiprobable (p = 0.5 each), high or low amount of liquid reward delivered after saccadic eye movement to the cue location. Thus, each risk cue indicates current risk (defined as standard deviation of each specific reward probability distribution) (blue arrows in B). The risk prediction error on any given trial is the specific risk indicated by the risk cue for the current trial minus the predicted, overall risk indicated by the fixation spot common for all trials (red arrows in B). C, Actual measures used. D, Behavioral responses. As risk increased, both monkeys showed an increasing preference for the risky cues compared with the safe cue (which had an outcome that equaled the expected value of the risky cues). Thus, the monkeys showed risk preference. Data points show mean percentage preference. Error bars show SEM.
- Figure 2.
Orbitofrontal neurons code risk prediction error. A, Single neuron examples. Smoothed histograms show responses from two example neurons coding risk prediction error with positive slope (left) and negative slope (right). B, Population responses. Smoothed histograms show averaged population responses from all neurons coding risk prediction error with positive slopes (left) and negative slopes (right). Gray bars indicate the time windows for analysis. C, Regressions on risk prediction error. Data points show mean firing rates averaged from all neurons during the shaded periods in B. Error bars show SEM. D, SRCs from linear regressions on neurons with significant positive betas (left) and negative betas (right) for RiPE at cue presentation. E, Scatterplot of CPDs for neurons with significant correlation coefficients for risk prediction error (red dots), risk (blue dots), and both (green dots) at cue presentation (left). R2 values from single linear regressions with risk prediction error and subjective value prediction errors as regressors (right).
