Now or later: A reinforcement learning model of behavioural delay

(Other) people notoriously delay the initiation and completion of work, sometimes beyond what is optimal. One prominent, and indeed experimentally validated, explanation is based on the fact that rewards delivered in the future are discounted. However, other factors can interact with discounting and affect policies, such as the amount of effort and the probability of successful completion. These have received less empirical attention. Here, we build and fit a new reinforcement learning model to the working trajectories of students over the course of a semester in a real-world task (P. Y. Zhang & Ma, 2024). We show that discounting, effort, and efficacy are all important in explaining students’ delays. In addition, the discount factors inferred from task performance correlate significantly with self-reported measures of impulsivity and procrastination, as well as discount rates estimated from a monetary delay discounting task, highlighting that they robustly capture meaningful individual differences in temporal preferences.