A fixed ridge test did not detect a mean AUROC gain from trial baseline division in eight SEEG recordings from ds004473 version 1.0.2

This study tested trial baseline division for arrow direction decoding in eight public SEEG recordings from OpenNeuro ds004473, version 1.0.2. It used 1,990 source response trials and 822 bipolar contact pairs. Separate fixed ridge models used early trials for training and later trials for testing in each patient and speed condition. Mean test AUROC was 0.4918 for response beta power and 0.4728 after division by the trial baseline. The mean paired difference was -0.0190. The exact patient sign-flip test gave p = 0.515625. It assumes independent patients and symmetric null differences. The descriptive patient bootstrap interval was -0.0694 to +0.0308. The test did not detect a mean gain. It does not establish equal performance or absence of direction information. The package contains input powers, trial assignments, fitted models, predictions, code, source records, and reproduction commands. A fresh run reproduced all numerical outputs exactly. Selected fresh voltage downloads checked all eight patients. The result applies to these windows and fixed models in this clinical sample.

stereoelectroencephalography · SEEG · beta power · trial baseline division · arrow direction decoding · ridge regression · AUROC · within-patient prediction