Human EEG Decoding Reveals Lapse-Prone States Beyond Selective Attention and Control Failures

Abstract

Attentional lapses are a ubiquitous feature of cognition, yet their underlying causes remain poorly understood. Theories of sustained attention often point to failures of cognitive control in maintaining the task-set, while data-driven approaches suggest that lapses may instead reflect a breakdown in the selection of task-relevant information. This study aimed to characterize the neural mechanisms of sustained attention lapses and to test whether EEG-based signatures of lapse-prone states are distinct from signatures of failures of selective attention and task-set maintenance. Twenty adults completed a sustained attention go/no-go task while focusing on either numbers or letters, with EEG recorded simultaneously. Poor sustained attention was examined at two complementary levels: trial-level lapses, defined as no-go errors, and attentional states, derived from reaction-time variability and categorized as "in-the-zone" versus "out-of-the-zone". Across both levels, suboptimal sustained attention was associated with attenuated event-related potentials, most notably a reduced parietal P3 amplitude and weaker whole-scalp inter-electrode correlation. To isolate a unique EEG marker of lapse-prone state, a machine-learning classifier decoded attentional state from EEG activity. Cross-validated accuracy reached [~]80% and remained robust after controlling for reaction time. Finally, representational similarity analysis confirmed that this neural signature was dissociable from stimulus-side selection and task-set maintenance.