Archived pyFR SEEG data did not replicate the FR1 cross-word tilt effect

We tested whether a word's encoding tilt, defined as mean z-scored high-frequency activity (70–150 Hz) minus theta power (3–8 Hz) across bipolar pairs, predicts recall of other words in its list after removing serial-position means and the word's own recall. The pre-specified replication test used archived sample/fs-timed 0–1.6 s band power from 17 pyFR SEEG subjects in OpenNeuro ds004865. The primary cross-word measure was S_W = +0.0056 (95% CI −0.016 to +0.027; one-sided p = 0.31; Holm p = 0.62; 8/17 subjects positive), so the FR1 result did not replicate. The own-word memory effect remained present (r0 = +0.063, p = 0.014), and a 23-person cohort independent of FR1 had S_W = +0.0016 (95% CI −0.015 to +0.019; p = 0.43). Power for an effect of the FR1 size was about 0.3. The result is therefore a weak null: a small true effect is not excluded, and pyFR differs from FR1 in list length, word pool, hospitals, and recording systems. Features were archived means across bipolar SEEG pairs without artifact or interictal epileptiform discharge rejection. The recorded source is OpenNeuro ds004865 at https://s3.amazonaws.com/openneuro.org/ds004865/; the exact dataset release used by the source packages is uncertain.

memory · free recall · SEEG · intracranial EEG · high-frequency activity · theta · replication · winner's curse · cross-word prediction · OpenNeuro