Adding two shaft power means increased mean squared log error by 1.49% in a fixed forward ridge test of eight SEEG patients

We tested whether two power means improve predictions of response time in later stereoelectroencephalography (SEEG) trials. We used public OpenNeuro ds004473 version 1.0.1 and a fixed forward ridge test in eight patients. The beta and gamma means gave each recorded shaft equal weight. The control model used task conditions and recent behavior. The test gave 1,220 predictions from 2,024 eligible trials. Mean squared log error increased by 1.49%. The 95% patient bootstrap interval for error reduction was -3.43% to +0.44%, under independent, exchangeable patient sampling. Error decreased in two patients and increased in six. The exact two-sided sign-flip probability was 0.21094, under independent patient effects and a symmetric null. Thus the test did not establish a benefit or a population loss. All fixed pooled checks had negative mean gain. Mean absolute error in seconds decreased by 4.05 milliseconds, so the result depends on the error measure. This is a secondary analysis of an already studied cohort. It does not show that SEEG lacks information about response time. The package preserves inputs, source versions, code, weights, models, predictions, uncertainty draws, checks, figures, and repeat instructions. Fresh pinned source checks rebuilt all 2,030 epochs exactly. An independent EDF reader reproduced 24 sampled epochs.

stereoelectroencephalography · response time · ridge regression · power features · forward prediction · spatial pooling · patient bootstrap