Theta power variance across words was lower for recalled words in 51 SEEG patients

In a pre-registered analysis of archived log10 band-power features from 51 patients with stereo-electroencephalography (SEEG) during free recall (OpenNeuro ds004789, RAM FR1, session 0), theta power variance across words in the 0–1.6 s encoding window was lower for later-recalled than not-recalled words. After removing list and serial-position means, the subject-level recalled-minus-not-recalled log variance effect, corrected by within-list label permutations, was −0.037 (95% CI −0.054 to −0.021; 37/51 subjects negative; Holm-adjusted p = 0.0002), about 3.6% lower variance. The pre-registered 70–150 Hz high-frequency activity test was not supported (−0.005; 95% CI −0.026 to 0.016; p = 0.69). Alpha and beta effects were also negative; low gamma was not significant. Exploratory hippocampal-pair effects were negative in all bands. A median absolute deviation analysis supported theta and beta effects. Theta variance and mean subsequent-memory effects correlated across subjects (r = 0.31), while the estimated theta variance effect at zero mean effect remained negative (−0.031; bootstrap 95% CI −0.048 to −0.013). These are word-to-word effects measured from a single 1.6 s window, not within-trial variability. The study did not test held-out prediction or establish that variance predicts recall better than the mean.

stereo-electroencephalography · free recall · memory encoding · theta power · word-to-word variability · subsequent memory effect · OpenNeuro ds004789