Mood impacts confidence through biased learning of reward likelihood

Abstract

Mood disturbances are accompanied by marked and apparently non-rational changes in self-efficacy and goal-directed behaviour, yet the mechanisms linking affective states to these fluctuations remain unclear. Here we tested whether mood biases confidence in reward-based decision-making and whether this is amplified by trait instability in mood and self-efficacy. By formalising both confidence and mood within a computational model of reward learning, we adjudicated between two accounts of these clinically relevant effects: mood imposing a direct, evidence-independent shift in confidence, or mood distorting the accumulation of reward learning signals from which confidence is then rationally inferred. Across two independent studies--a laboratory experiment (n = 35) and a preregistered online replication (n = 106)--participants completed a reinforcement-learning task with mood manipulation and repeated confidence ratings. Mood systematically biased confidence without affecting objective accuracy. Critically, this effect accumulated gradually across learning rather than emerging immediately after mood induction, even as mood itself returned toward baseline. These dynamics were captured by a computational model in which mood biases the accumulation of reward probability estimates. A single mood-bias parameter accounted for shifts in confidence, biased valuation of stimuli learned under different mood states, and amplification of these effects in individuals with elevated hypomanic traits. These findings suggest that disturbances in confidence arise from mood-biased learning signals, providing a mechanistic account linking these distortions to disrupted goal pursuit and self-efficacy in mood disorders.