Regret is a powerful emotion that can either hinder or enhance decision-making. While prior research has documented regret's effects, existing theories remain largely descriptive. My dissertation introduces a formal, computational model of regret that simulates when regret promotes learning and when it does not. I take a multi-method approach to examine regret across lab-based tasks, open-ended narratives, and real-world experience sampling. I first show that the timing of regret critically shapes its effects on learning: early regret can lead to confusion, while later regret improves learning outcomes. I then extend the model through behavioral experiments manipulating counterfactuals and perceived responsibility. These computational and behavioral findings are later validated in a large-scale qualitative analysis of everyday regrets as well as an ecological momentary assessment (EMA) study of regret dynamics over time. Together, these studies outline a testable theory of regret and contribute to the emerging field of Computational Affective Science.
