Robust EEG brain-behavior associations emerge only at large sample sizes

Abstract Electroencephalography (EEG) offers unique access to the oscillatory dynamics that shape human cognition and behavior. Yet reported associations between EEG features and behavioral measures often fail to replicate, raising fundamental concerns about the robustness of this literature. A key reason is the predominance of small sample sizes (median ∼30 participants), but the consequences for reproducibility have not been systematically evaluated. Here, we provide the first large-scale empirical test of this key issue using the largest available harmonised resting-state EEG dataset, comprising 2,292 participants across healthy individuals and clinical populations. We extracted a comprehensive set of spectral, temporal, complexity, and dynamical features, and assessed their correlations with behavioral, cognitive, and mental health measures. Using extensive resampling across a wide range of sample sizes, we show that small samples produce unstable associations with inflated and inconsistent effect sizes, whereas larger cohorts converge on smaller but reproducible effects. These findings establish sample size as a critical determinant of reliability in EEG research and call for a re-evaluation of how oscillatory dynamics are linked to human cognition. By moving beyond low-powered correlations toward large-scale, reproducible approaches, the field will be better positioned to uncover the true role of neural oscillations in shaping behavior.