Recorded response timing reduced mean test error for beta power by 0.47% in selected pairs from eight SEEG recordings under fixed models

We tested whether recorded response times improve predictions of beta power on separate trials. We used OpenNeuro ds004473 version 1.0.2. The data came from eight people with epilepsy. Fixed rules retained 1,875 trials and 822 adjacent bipolar pairs. We compared a cue model with a model that added response timing. Pair choice and model fitting used training trials only. The mean reduction in test squared error was 0.47%. Its participant bootstrap 95% interval was -0.09% to 1.30%. Four of eight participants had lower error. The exact two-sided sign-flip p value was 0.3125. Recorded timing exceeded the mean shuffled-timing gain by 0.79 percentage points. That control kept pair choice fixed and assumed exchangeability within condition and fold. The planned all-pair model increased mean test error by 0.41%. Thus the primary models did not show a consistent gain across people. The result depended on the target and model rules. The package preserves new derived powers, all test predictions, source code, source versions, range hashes, raw samples, and reproduction instructions. Independent readers and calculations passed 221 checks. A second analysis reproduced all numerical outputs exactly. The source paper already reported sharper patterns after response alignment. This study adds a fixed prediction test in the same cohort. It does not isolate a causal movement process.

stereo-EEG · beta power · response timing · held-out prediction · ridge regression · OpenNeuro · permutation control