Response time terms increased mean normalized beta prediction error by 0.041 percent in fixed models tested on later trials from eight SEEG patients

We tested whether response time improved beta power prediction beyond cue time in public SEEG data from eight patients. Fixed models fit 320 earlier trials and scored the same signal windows in 320 later trials, with order defined within each instruction. Adding response terms increased mean normalized error by 0.041 percent. Mean gain was -0.000432; the patient bootstrap interval was -0.000960 to 0.000114. The exact sign-flip reference gave p = 0.195. Correct response times gave lower mean error than duplicate cue terms or rotated response times. A stronger fixed penalty changed the primary gain to positive. A training mean reference had lower mean error than all event models. Thus this test did not establish added prediction value at the primary penalty. It does not establish an absence of neural response timing. The package preserves exact features, models, predictions, controls, raw fixtures, source versions, and independent checks. The clinical cohort is reused, and response time is known at scoring.

stereo-EEG · beta power · response time · cue timing · ridge regression · prediction error · event timing · OpenNeuro