Across five RAM verbal-memory tasks, theta decreases generally while HFA increases are task-specific

This analysis asks whether the SEEG encoding subsequent memory effect (SME) is consistent across five RAM-family verbal-memory tasks. A NumPy-only script calculated one metric from archived sample/fs-timed band-power features for FR1 (31 persons), catFR1 (30), PAL1 (18), pyFR (20), and non-stimulation catFR3 lists (5). The 104 task-level observations include people who took part in more than one task. For each person and bipolar pair, the analysis removed list and serial-position means in two passes, z-scored across words, averaged across pairs, and calculated Cohen's d for recalled versus not-recalled words. It pooled task means with a DerSimonian–Laird random-effects model. Theta d was negative in all five tasks; pooled d = -0.087 (95% CI -0.140 to -0.035; Holm-adjusted p = 0.002). HFA d was positive in catFR1 (+0.076) and pyFR (+0.151), but not in FR1, PAL1, or catFR3; pooled d = +0.032 (95% CI -0.037 to +0.101; p = 0.37). HFA task means differed more than chance (Q permutation p = 0.028), and the prediction interval for a new task included both signs (-0.20 to +0.27). The HFA-minus-theta tilt pooled d was +0.089 (p = 0.0007; not in the Holm correction set). A unique-person analysis (89 persons) and leave-one-task-out analyses gave the same main conclusions. Effects were small (absolute task-level d < 0.16), and limits include only five tasks, five catFR3 participants, different source time windows (0–2 s for PAL1; 0–1.6 s for other tasks), overlapping participants, and features that were not re-derived from raw data. Pair selection, referencing, and band estimation followed each source package; sample/fs timing was assumed correct as reported by those packages.

SEEG · subsequent memory effect · free recall · theta · high-frequency activity · random-effects meta-analysis