Under independent patient label swaps, pair bootstrap intervals covered the known mean AUC of 0.5 in 44.57 percent of tests with 34 SEEG patients

We tested interval coverage for an equal-patient mean AUC with public SEEG data from OpenNeuro ds004100 version 1.1.3. The fixed inputs had 34 patients and two 60-second clips per patient. The score measured time concentration of 80 to 150 Hz power. Independent reversal of all labels within each patient gave an exact expected mean AUC of 0.5. In 10,000 null cohorts, nominal 95 percent pair bootstrap intervals covered 0.5 in 44.57 percent of tests. Patient intervals covered it in 94.14 percent of tests. Patient and pair intervals covered it in 95.14 percent of tests. Pair resampling supplied only 9.33 percent of the exact null variance. The gap remained with disjoint pairs and a mean-power score. This is a computational null model. It does not estimate clinical error rates. The package preserves measured window powers, trial outputs, bootstrap banks, source versions, code, checks, and repeat instructions.

bootstrap intervals · AUC · SEEG · label permutation · coverage · grouped data