Welch tests that counted 20 EEG epochs per person as independent rejected 60.5 percent of random person label allocations for posterior theta-to-alpha ratios in 88 ds004504 v1.0.9 records

This study tested statistical inference for repeated EEG epochs in 88 OpenNeuro ds004504 version 1.0.9 records. Each person had 20 ten-second epochs. The fixed measure was the natural log of the posterior theta-to-alpha power ratio. For each of 20,000 random allocations, 44 people received each label. The label stayed constant within a person. At a two-sided p value below 0.05, Welch tests that counted epochs as independent rejected 60.54 percent of allocations. Tests of one mean per person rejected 5.02 percent. Their mean differences were equal. The median person-to-epoch standard error ratio was 3.83. Person randomization tests rejected 5.22 percent. Epoch rejection rates remained high within each diagnosis group. Intervals describe Monte Carlo error in this fixed cohort. They do not describe new-cohort uncertainty. The package preserves input features, source hashes, code, every label allocation, p values, figures, and instructions. Independent calculations checked all 40,000 primary Welch tests. A fresh source extraction matched saved features. A repeat analysis matched all six scientific output files. This result measures test error. It does not establish clinical diagnosis or disease causation.

pseudoreplication · electroencephalography · Welch test · randomization inference · repeated measures · null control