Interferences between decisional variables: Behavioural and computational studies

Should I try to grow an orange tree on my balcony? According to modern decision theory, when individuals consider taking on a course of action that involves expending effort and other costs in the pursuit of valuable yet uncertain outcomes, their decision typically follows on a specific, subjective assessment of the overall value of that course of action, relative to the value of alternative courses of action. It is assumed that the overall value of a given course of action is its expected utility, i.e., the sum of the subjective values of all its possible outcomes weighted by their respective probability, discounted by the costs that this course of action entails. Over the last forty years, many violations of the expected utility maximization principle have been pointed out, and almost as many adjusted decision models have been proposed, in an effort to better account for actual human behaviour. Despite these consecutive adjustments, one implicit assumption has remained at the core of decision theory until this day: that of a mutual independence between the decisional variables that are combined together in action valuation, in particular 1) the subjective value of possible outcomes, 2) the subjective probability of these outcomes, 3) the subjective cost of action. Yet, over the past decades, substantial evidence from behavioural psychology and neuroeconomics experiments has accumulated, showing that value, probability and cost judgments can influence, or interfere with, each other. In this PhD work, we investigated such manifestations, with the aim of elucidating the computational and neural mechanisms underlying these interferences. For this, we conducted three studies on healthy human volunteers. In the first study, we had participants anticipate the energetic cost of composite running routes whose later completion could entail more or less valuable monetary outcomes, either framed as obtained gains or avoided losses. We found that a given route was anticipated as more costly when it was paired with a larger monetary stake compared to a smaller one (either in the gain or loss frame). Besides, our behavioural data were most in line with a cognitive scenario according to which individuals’ prospective effort judgment is contaminated by the output of a cost-benefit computation. In the second study, we moved on to the investigation of interferences between value and probability, with a focus on a particular case of probability judgment: confidence, that is, the subjective probability of being correct or successful at a given task. Our participants were trained to perform a motor precision task; next, they underwent testing trials in which they were first presented with information about the upcoming motor challenge: its difficulty (adjusted via the size of the target to be hit), the magnitude of their monetary gain in case of success, and the magnitude of monetary loss in case of failure; then, they were prompted to estimate their prospective chances of success (i.e. to report their confidence); finally, they vi tried to hit the target but did not receive any feedback about their performance. Our main finding was that individuals tend to be more confident when faced with a larger prospective gain, but less confident when faced with a larger prospective loss. This pattern of distortions is evocative of an affect-as-information type of misattribution. In our third study, we drew inspiration from earlier neuroimaging findings to predict that value-confidence misattributions may occur bidirectionally, and between unrelated items. To test our predictions, we had participants answer more or less difficult general-knowledge quiz questions while hearing more or less pleasant musical extracts, after which we asked them to make a confidence rating about their answer to the quiz question, or a likeability rating about the musical background. Partly in line with our predictions, higher confidence led to higher likeability ratings, whereas the opposite effect was more elusive. Overall, we managed to isolate, at the behavioural level, various cases of interference between decisional variables in a well-controlled setting, and we gained some insight about their underlying cognitive mechanisms, which seem to differ from one type of interference to the other.