The 95% error reduction interval included zero for a ds003775 EEG age model in 60 ds005385 adults before cognitive tasks

We tested whether a fixed EEG age rule could work in a separate dataset. We used 108 screened people from ds003775 v1.2.1. We fit age on sex and a posterior 7-13 Hz band maximum. We applied the rule to 60 selected adults from ds005385 v1.0.3 without a new fit. Each of five age bands had equal weight. Before cognitive tasks, mean squared error fell from 254.62 to 233.11 years squared. The reduction was 21.50 years squared, or 8.45%. A two-cohort person bootstrap gave a 95% interval of [-17.82, 53.85] years squared. Mean absolute error was 12.36 years. Mean signed error was -6.33 years. All fixed secondary error reduction intervals also included zero. The source and target signal methods differed. Thus this test cannot separate a cohort effect from a signal method effect. The package preserves 231 derived records, source receipts, raw retrieval recipes, code, predictions, and bootstrap samples. All 12 numerical result files matched a second offline run exactly. One fresh raw record from each dataset matched all derived arrays exactly.

electroencephalography · EEG · age prediction · frequency · alpha band · cross-dataset validation · bootstrap · OpenNeuro