A fourfold decrease in residual power moved 8–13 Hz EEG centroids in both directions in models with fixed shapes from 60 adults

We tested a power-weighted mean frequency in posterior EEG spectra from 60 adults with three visits each. The primary model used closed eyes and fixed 8–13 Hz limits. A lower background and a nonnegative residual sum to each measured spectrum. We held both frequency shapes fixed and reduced residual power to one quarter. The mean centroid change was −0.00047 Hz, with a 95% participant bootstrap interval of −0.03460 to +0.03222 Hz. The primary test showed no clear mean shift. Individual model changes had both signs. Five of 60 people had a change greater than 0.25 Hz in magnitude after averaging their visits. An exact identity and a sharp bound show how relative component power can change the total centroid while the residual centroid stays fixed. The package contains the audit tool, proof, all 360 input spectrum files, all planned checks, 20,000 participant draws, figures, source hashes, and reproduction commands. A new offline run matched all 18 result files by SHA-256. Four complete source EEG downloads rebuilt their stored arrays exactly. The split is a mathematical model. It does not identify a neural source or show a change in neural rhythm frequency. The model results do not establish a causal eye-state effect.

EEG · power centroid · alpha frequency · spectral analysis · power spectrum · signal processing · test-retest · estimator audit