A mean visit-change correction had higher observed loss for later normalized posterior 2–30 Hz EEG spectra in 106 ds005385 v1.0.3 test adults

We tested a fixed mean visit-change correction for later EEG spectra in OpenNeuro ds005385 v1.0.3. The test used 101 reference adults and 106 separate test adults. It used pre-task closed-eye records, nine posterior channels, and normalized power from 2 through 30 Hz. Both rules used a fixed weight of 0.5 on the person's earlier square-root spectrum. The correction added half the mean visit change from the person's reference age and sex cell. The mean squared Hellinger loss was 0.0358594 for the comparison and 0.0383726 for the correction. The correction had 7.01 percent higher observed loss. The mean gain was -0.00251315. Its planned basic 95 percent bootstrap interval was [-0.00266010, 0.00066684]. This interval includes zero. It does not establish benefit, harm, or equality. All ten fixed checks also had negative mean gain and basic intervals that include zero. The percentile interval for the main test was [-0.00569314, -0.00236620]. This difference between interval methods limits claims about a population. The package preserves source spectra, hashes, code, all predictions, bootstrap scores, and retrieval instructions. Independent checks reproduced the main arithmetic and one complete raw record pair.

resting-state EEG · longitudinal EEG · spectral prediction · Hellinger loss · visit-change correction · OpenNeuro · posterior spectrum