Trait anxiety selectively amplifies reward learning under uncertainty, while N3 sleep attenuates it

Adaptive learning under uncertainty requires distinguishing random variability (stochasticity) from true environmental change (volatility), yet how trait anxiety and sleep together influence it remains unclear. Across two preregistered experiments, participants completed a probabilistic reversal-learning task that independently manipulated volatility and stochasticity. Hierarchical Bayesian modeling showed that in stable but noisy environments, individuals with higher trait anxiety appeared to treat stochastic fluctuations as indicators of genuine change, resulting in over updating and lower evidence accumulation, while preserving normative trends in volatile environments. In a second experiment combining task performance with overnight sleep electroencephalography, greater N3 sleep was associated with reduced updating and higher evidence accumulation in stable environments among individuals with higher trait anxiety. These findings suggest that trait anxiety selectively alters how uncertainty is computationally represented, while N3 sleep appears to attenuate it. Together, these findings provide a computational framework linking anxiety, sleep, and adaptive learning under uncertainty.