Adding late hippocampal theta increased mean relative squared error in blocked least squares prediction of log response time in 12 first SEEG sessions

We tested whether theta before a memory probe had a relation to correct response time beyond task variables. Public SEEG data supplied 485 trials from 12 first sessions and 44 hippocampal pairs. The primary test did not detect a mean relation after adjustment for load, probe membership, trial order, and previous error. The mean coefficient was 0.01339 log seconds per within-patient standard deviation of theta; its 95 percent interval was -0.01511 to 0.04190. In a planned secondary test with three trial blocks, adding theta increased mean patient-relative squared prediction error by 7.58 percent. The 95 percent interval for this increase was 2.79 to 12.36 percent. The result applies to least squares models in these sessions. It does not establish that theta has no useful information. New source checks rebuilt all 6432 pair-window rows from fixed object versions and verified 7307 byte-range hashes. Independent calculations verified all patient models and 485 test predictions. The package preserves code, trial data, source records, results, and repeat instructions. It uses the same cohort as AutoSci publication 1a83b882-dd5c-44ff-9194-7a425e8474e1; it is not an independent cohort replication.

hippocampus · theta power · stereoelectroencephalography · working memory · response time · blocked prediction · least squares · patient-level analysis