Division by mean power lowered one-month agreement of log 8-13 Hz power standard deviation in 60 ds004148 v1.0.1 adults with closed eyes and four-second windows from 280-second records

We tested whether division by mean power changes repeat agreement of within-record alpha power spread. We used 60 adults from OpenNeuro ds004148 version 1.0.1. The primary test used closed eyes at sessions one month apart. It used preserved posterior 8-13 Hz spectra from four-second windows within seconds 10 to 290. Absolute agreement ICC was 0.896 for log standard deviation and 0.476 for log coefficient of variation. The coefficient of variation is standard deviation divided by mean power. The paired ICC difference was 0.419, with a 95% person bootstrap interval of 0.152 to 0.730 from 10,000 draws. All five planned checks gave positive difference intervals. The exact relation log spread = log mean power + log normalized spread accounted for the variance components. The package preserves all 360 input spectrum files, source hashes and retrieval records, 50,400 window power rows, all scores and bootstrap draws, code, figures, and run instructions. Separate formulas checked all 450,000 bootstrap statistics. Four original raw records rebuilt all arrays exactly. A fresh run made all nine numeric result files identical. The result applies to this cohort and processing protocol. It does not test window order, burst timing, a neural cause, or a new independent cohort.

electroencephalography · alpha power · test-retest reliability · intraclass correlation · coefficient of variation · signal processing