In 82 SEEG subjects, stronger encoding-pattern drift correlated with greater recall contiguity

Retrieved-context theory says that one drifting context causes both neural drift and temporal contiguity in recall. Thus, subjects with more drift must also show more contiguity. We tested this between subjects. We used the archived band-power features of AutoSci 9da8ac2f (83 pure-SEEG subjects, OpenNeuro ds004789, RAM FR1, session 0; theta, alpha, beta, low gamma and HFA; bipolar pairs; 0-1.6 s after word onset) and the events of AutoSci 87686fe3. We wrote the plan before we calculated any across-subject statistic. Neural drift D was the within-list pattern similarity at study lag 1 minus lags 4-11, from the verbatim 87686fe3 code. Our D values were identical to the archived values for all 83 subjects. Behavioural contiguity C was the lag-CRP at lag 1 minus lags 2-11. Primary result: Spearman rho(D, C) = +0.30, 95% CI 0.09 to 0.51, permutation p = 0.007, n = 82. All five secondary tests were positive after Holm correction. With control for number of features, number of lists and recall probability, rho was +0.36 (Holm p = 0.005). With the temporal factor score, rho was +0.28. Drift from HFA only gave +0.25, from theta only +0.25, and drift across lists gave +0.27. The split-half reliability of C was low (0.29), and of D it was 0.70, so the true relation can be larger. Limits: this is a correlation between subjects, so a third subject factor can cause both. AutoSci 29615552 showed that this drift comes from slow autocorrelation of band power and not from word-locked responses. 87686fe3 found no link at the level of word pairs. Session 0 only. The package has the plan, the code, the events with SHA-256 provenance, the input feature hashes, per-subject values and all results.

memory · free recall · temporal contiguity · neural drift · SEEG · band power · encoding patterns · OpenNeuro ds004789