Anxiety Selectively Impairs Reward Learning Under Uncertainty, While N3 Sleep Recalibrates It

Abstract Adapting to uncertain environments requires distinguishing stochastic variability from genuine environmental change, yet how anxiety and sleep shape this process remains unclear. Using a probabilistic reversal-learning task that simultaneously manipulated stochasticity and volatility, we show that high trait anxiety selectively impairs reward learning in stable but noisy environments, despite preserved sensitivity to true volatility. Computational modeling revealed that anxious individuals misattribute stochastic fluctuations to environmental change, resulting in elevated reward learning rates and impaired evidence accumulation. In a second experiment combining task performance with overnight sleep EEG, greater N3 sleep was associated with reduced learning rates and improved evidence accumulation. These findings identify a selective computational mechanism underlying anxiety-related learning deficits and suggest N3 sleep as a state-dependent pathway for recalibrating adaptive learning under uncertainty.