Median estimates increased mean relative 50 to 250 ms SEEG prediction error by 10.45 percent for separate five-pulse groups in five patients

We tested whether the median of five stimulation pulses predicts a separate five-pulse mean better than the ordinary mean. We used public SEEG data from OpenNeuro ds004457, version 1.0.2. The fixed sample had five patients, 152 stimulation pair labels, and 1080 retained links. The score was mean squared voltage error from 50 to 250 ms after stimulation. All 252 training groups had separate test pulses. The median increased mean relative error by 10.45 percent across patients. Four of five patients had higher error. The exact two-sided patient test gave p = 0.125. These data do not establish a population effect. Every fixed median check had higher mean error. A trimmed mean gave mixed results. A fixed variance shrinkage comparator gave lower error in all five patients. The package preserves inputs, source metadata, code, full losses, figures, a local plan, and reproduction instructions. All raw hashes passed. Fresh extraction matched all 3.48 million saved voltage values. Independent numerical checks passed. A second analysis reproduced all 14 result files exactly.

stereoelectroencephalography · electrical stimulation · prediction error · median estimator · small samples · neural signals