Performance anxiety is associated with biases in learning from reward and punishment in skilled individuals

Many individuals experience performance anxiety (PA) in high-stakes settings, from public speaking to the performing arts. Although debilitating PA is associated with physiological, cognitive, and affective alterations, its mechanisms remain poorly understood. Using behavioural analysis, computational modelling, and electroencephalography, we examined whether PA predisposes individuals to rely more on punishment than reward signals during learning, particularly under high task uncertainty, where anxiety is proposed to exacerbate maladaptive learning. Across three experiments with 95 skilled pianists, participants learned hidden melody dynamics via graded reward or punishment feedback. Bayesian hierarchical modelling revealed that individuals with higher PA learned faster from punishment under low uncertainty, but increasingly relied on reward as task uncertainty increased. These learning biases were mediated by reinforcement-driven modulation of motor variability, which increased following poor outcomes. A generative model of motor variability regulation further showed that reinforcement-sensitive scaling of motor variability was more influenced by punishment than reward, and varied with PA levels. Exploratory evidence suggested that high-uncertainty contexts were perceived as more aversive, particularly among performers with greater intolerance of uncertainty, offering a potential affective account of this learning shift. At the neural level, a convolution general linear model revealed that the interaction between PA and reinforcement modulated frontal theta (4–7 Hz) activity, encoding feedback changes and signalling upcoming motor adjustments. These findings reveal that PA alters the weighting of reward and punishment signals depending on task uncertainty, providing a neurocomputational account of how PA affects skilled performance.