Linear month forecasts had higher mean squared error than group means for early-to-late alpha contrast change in 60 ds004148 adults

We tested whether a person's early-to-late posterior alpha contrast change could predict its repeat about one month later. We used 60 adults from OpenNeuro ds004148 version 1.0.1. Each eye-state part had 140 planned seconds. Contrast was closed-eye minus open-eye 8–13 Hz log power. Each fixed linear fit excluded its test person. The linear mean squared error was 5.747 dB squared. The training-group mean error was 5.677 dB squared. Gain was -0.070 dB squared. We refitted the forecasts in 20,000 whole-person bootstrap draws. The 95 percent percentile interval was -4.664 to 0.663 dB squared. The basic interval was -0.803 to 4.524 dB squared. Both included zero. Mean contrast change was negative at all three visits. This group pattern did not establish useful prediction for a person. The wide intervals do not prove absence of benefit. Fixed state order and different source processing between eye states limit interpretation. The package preserves 360 unchanged spectrum files, new scores, forecasts, draws, source hashes, code, figures, and reproduction instructions. Separate numerical checks passed. Four complete source EEG files gave exact spectrum matches. A second analysis run gave identical bytes for all 12 outputs. A fresh environment gave numerical differences below 8e-13.

electroencephalography · alpha contrast · test-retest reliability · prediction · bootstrap