Abstract
In a Bayesian framework, decision-making involves combining information available at the time of the decision with prior experiences or expectations. The prior is especially important since available information can often be ambiguous or incomplete. As such, priors are thought to have a profound impact on behaviour. For example, the generalised passive behaviour in patients suffering from learned helplessness and depression has been conceptualised as resulting from a pessimistic prior implying the expectation of failure when performing an action. When measuring these priors in the lab, it is important that they generalise across different situations. Otherwise, the priors captured may only reflect task-specific assumptions that do not apply more widely. Task-invariant perceptual priors have been measured using cognitive tasks, however, task-invariant priors for affective decision-making have not been quantified. To quantify a generalizable prior that actions will be successful, we administered two differently framed decision-making tasks. In the tasks, 54 participants made repeated decisions as to whether to take an active action or not. The passive choice was associated with a sure loss or missing out on a potential reward while the active action was associated with a cost and a probability to avoid the negative outcome. Thereby, the value of the passive choice is fully transparent. We then used computational modeling to characterize each participant’s prior expectation that an active choice will result in avoiding a negative outcome. Parameter and model recovery for all models were satisfactory. We found that a Bayesian prior model fit better than 6 alternative models (Bayesian model with a beta distribution prior vs the best-fit model without Bayesian updating: Task 1, model frequency (MF)=.648, exceedance probability (EP)=.985; Task 2, MF=.658, EP=.990). Furthermore, the means and variances of the priors correlated across tasks (M: r=.407, p=.003, var: r=.271, p=.047). When explicitly modelled, slightly more participants were fitted by a model with a common prior across the tasks than a model with one prior for each task (MF=.535, EP=.692). Furthermore, the mean of this common prior was moderately reliable one week later (N=43, ICC(2,1)=.569). The common prior also correlated with positive mood (M: r=.244, p=.075, var: r=-.310, p=.022). In conclusion, our data show that it is possible to use decision-making tasks to quantify prior beliefs that an active choice will result in avoiding a negative outcome. This method will be used to quantify these priors in clinical populations. This may provide insights into learned helplessness and depression.