Figure 6. Loss Experiment.

(a, bottom) Value and salience are collinear when values are only related to rewards (right half) but it is possible to determine whether each factor has an independent effect by also looking at choices made to avoid losses (both left and right halves). (a, top and b) Four hypothetical effects of distractor on accuracy. The first and second hypotheses suggest that the distractor effect is positive, which is predicted by the mutual inhibition model and is related to the distractor’s value and salience respectively. The third and fourth hypotheses suggest that the distractor effect is negative, which is predicted by the divisive normalisation model, and is related to its value and salience respectively. All four hypothesis are not mutually exclusive – value and salience are orthogonal factors and positive/negative distractor effects can predominate different parts of decision space. (c, right) A plot identical to that in Figure 3e that shows the data from the gain trials of Experiment 3 Hong Kong. Accuracy (light-yellow indicates high accuracy, dark-red indicates low accuracy) is plotted across the decision space defined by decision difficulty (HV-LV) and relative distractor value (D–HV). (c, left) A similar plot using the data from the loss trials of Experiment 7 Loss Experiment is shown. (d) GLM analysis indicates the distractor value D had a negative effect, suggesting that accuracy was more impaired on trials with distractors that were associated with fewer losses or more gains. In contrast, the distractor salience |D| had a positive effect, suggesting that accuracy was more facilitated on trials with more salient distractors (i.e. those related to larger gains or losses). (e, left) The negative value D effect was significant on easy trials (orange) and reversed and became positive on hard trials (blue). (e, right) In contrast, the positive salience |D| effect was significant on both hard and easy trials. In the dual route model, there are two components that guide decision making in parallel. This pattern suggests that the positive distractor effect of the mutual inhibition component is related to the salience and value of the distractor whereas the negative distractor effect of the divisive normalisation component is most closely related to the value of the distractor. *p<0.05, **p<0.01, ***p<0.001. Error bars indicate standard error.