Integration of ‘what’ and ‘when’ predictions during action-effect processing

Voluntary actions generate predictions about their sensory consequences, but it remains unknown whether predictions arising from different dimensions of intentional action are processed independently or integrated into a unified expectation. We addressed this question by combining action choice (“what”) and action timing (“when”) within a single action-effect learning paradigm. Participants learned independent associations between each action dimension and a distinct auditory feature and performed self-paced actions while electroencephalography was recorded. This design allowed action-choice and action-timing predictions to be fulfilled or violated independently or simultaneously. Event-related potentials were analysed using Bayesian linear mixed-effects models. The P2 component showed a significant interaction between action-choice and action-timing prediction violations: neither violation alone reliably modulated P2 amplitude, whereas simultaneous violations produced a marked reduction, consistent with a conjunctive prediction-error response. The later P3 component exhibited a graded pattern, with action-timing violations increasing amplitude and action-choice violations contributing an additional increment when both predictions were violated. These findings suggest that predictions derived from different dimensions of voluntary action are integrated during early sensory processing, whereas later stages evaluate prediction violations more cumulatively. The present results extend previous work on action-effect prediction by showing that predictions associated with action choice and action timing interact during the processing of their sensory consequences.