Learning From ‘What Might Have Been’: A Bayesian Model of Learning From Regret

Regret is a common emotion that might either catalyze or impair decision-making. What determines whether regret will be helpful or harmful in a given situation? We test the hypothesis that regret is more likely to hinder decision-making during the early stages of learning, when information is limited, but help during later stages of learning, when the learner has a better understanding of the environment. We introduce a Bayesian model of learning from regret, in which the “counterfactual weight” parameter – reflecting how strongly individuals update their beliefs about foregone outcomes – predicts both learning outcomes and the intensity of subjective regret. We find that probing regret early in the learning phase leads to worse performance than probing regret later or not at all. This work has important implications for both cognitive and affective science, shedding light on the appraisal mechanisms by which regret influences decision-making.