Cross-fit tests found no reliable pair-level overlap between serial-position change and subsequent memory effect in 27 FR1 patients

We tested whether bipolar electrode pairs with larger serial-position changes in encoding band power also had larger subsequent memory effects (SMEs) in free recall. We used archived 0–1.6 s word-locked, sample/fs-timed band-power features from 32 pure-SEEG subjects in OpenNeuro ds004789, session 0; 27 met the inclusion rule. In cross-fit analyses, each pair's serial-position slope and recall t statistic (with serial position controlled) came from separate odd/even list halves. Across subjects, the mean Fisher z overlap was −0.043 for high-frequency activity (HFA; 70–150 Hz; 95% CI −0.102 to +0.009; Holm p = 0.14) and −0.057 for theta (3–8 Hz; 95% CI −0.108 to −0.006; Holm p = 0.075). Neither pre-specified hypothesis was supported. Serial-position effects were present in this subset (HFA slope −0.0087 SD/position; theta +0.0130). SME map split-half reliability was low (HFA 0.078; theta 0.111), so a small true relation remains possible. Exploratory, uncorrected analyses found negative overlap for beta (mean z −0.096, p = 0.00015) and low gamma (−0.075, p = 0.00055); these results need confirmation in another cohort. Without cross-fitting, HFA overlap was larger (mean z −0.102, p = 0.014), consistent with shared trial noise inflating overlap when both measures use the same trials. The inference is across subjects; bipolar pairs within a subject are not independent.

free recall · subsequent memory effect · serial position · SEEG · band power · cross-fitting · memory encoding