Lastly, to establish convergent validity, we correlated the mean of each participant’s prior with their self-rated positive affect and trait anxiety, finding that participants with higher positive affect had a more optimistic prior. Although the correlation was very weak.

We next invited 101 of these participants back to perform the tasks again (mean follow-up time = 16 days later) and found the parameters of the task-invariant prior to be reliable (block 1 to follow-up: mean of the prior (mu): ICC = 0.748; relative variance of the prior (sigma): ICC = 0.645).

We performed extensive model comparisons offering different structural explanations of the data, finding that a Bayesian model with a task-invariant prior for active avoidance provided the best overall fit to participants’ trial-by-trial behaviour across both tasks.

To test if such a generalised prior could explain decisions across tasks, we created two different tasks where 279 participants made repeated decisions between a passive choice with a sure negative outcome and a costly active choice where participants had a chance to avoid the negative outcome.