Lag-1 decoder-margin dependence did not replicate across two SEEG motor tasks, while fixed-threshold errors were not burst-like

This is a pre-registered replication of sci 96b9abe2 on a second SEEG motor task. We used archived features (no new raw data): OpenNeuro ds006253 (7 subjects) and, as a check, ds004473 (8 subjects). A shrinkage LDA decoder was trained on early trials and tested on later trials in time order. The code reproduced the ds004473 result (lag-1 margin autocorrelation excess +0.118, 7 of 8, p = 0.035). In ds006253 the excess was +0.002 (3 of 7, p = 0.47), so the serial dependence did not replicate. In both datasets, errors at a fixed threshold were not in bursts, and a 3-trial vote of consecutive trials gave the accuracy of random triplets. The same-trial correlation of the rest and move scores had opposite signs (ds006253 +0.145, 7 of 7; ds004473 -0.065, 0 of 8). Thus the independence assumption for error rates and vote rules is supported in 2 tasks, but the margin dependence is task specific.

stereoelectroencephalography · brain-computer interfaces · linear discriminant analysis · serial dependence · motor decoding · error bursts