Temporal halves increased squared disagreement 2.76-fold for EEG theta-to-alpha scores in 95 ds008896 v1.0.1 records

We compared temporal halves with balanced random halves of the same posterior EEG score pools in 95 public OpenNeuro ds008896 v1.0.1 records: 32 people with CLN3 disease and 63 NIH volunteers. Each pool contained the first 28 retained four-second epochs, and each half contained 14 epochs. The score was the mean epoch log10 theta-to-alpha power ratio using a 19-channel mean reference. Temporal mean squared difference was 0.022883; the exact random-partition mean was 0.008292. The excess was 0.014592 squared log10 units (approximate 95% person bootstrap interval 0.007944 to 0.022452), and the ratio was 2.760 (interval 2.027 to 3.517). All six fixed cohort checks had positive excess intervals, and no one-person omission changed the excess sign. A fresh replay checked all 95 source records and 5,558 epochs. The result shows a split-choice effect in these processed records; it does not measure repeatability across visits, establish a neural cause, or support a causal or clinical interpretation. Retained epochs need not be continuous, and source processing and missing covariates limit interpretation.

EEG · theta-to-alpha ratio · temporal split · random partition · epoch order · CLN3 disease · finite-population variance · reproducibility