Encoding theta gave −1.14 percent mean maintenance prediction gain in excluded sessions from 14 SEEG patients with ridge penalty 10

We tested if encoding theta power improved prediction of later maintenance theta power in public hippocampal SEEG data. The baseline included letter load, trial position, fixation power, and encoding broadband power. We excluded each test session from model fitting. Both models used ridge penalty 10 without tuning. The analysis used 14 patients, 62 sessions, and 2,561 trials from OpenNeuro ds004752 version 1.0.1. Mean patient prediction gain was -1.14 percent. The 95 percent patient bootstrap interval was -2.58 to +0.11 percent. A positive gain means lower test mean squared error. This fixed model test did not establish a mean prediction gain. The interval assumes independent and exchangeable patients. It holds fitted patient scores fixed. The result does not prove that encoding theta has no relation to later power. The package preserves code, feature data, source records, trial predictions, patient results, figures, and repeat instructions. Fresh complete raw file retrievals at three sample rates reproduced every included feature array exactly. A second analysis run gave byte-identical result files. This is a new analysis of the source cohort, not an independent cohort replication.

intracranial EEG · theta power · verbal working memory · hippocampus · ridge regression · session transfer