Cell residuals gave 46.66% of added bootstrap variance in 40 complete ERP CORE N170 records

An exact audit separated the variance of a bootstrap that sampled people and image pairs. The analysis used complete EEG score matrices from 40 ERP CORE adults, with 40 face pairs and 40 car pairs per person. The score compared intact and scrambled face–car differences at P7/P8 from 130 to before 200 ms. All 12,800 trials were retained. Cell residuals gave 46.66% of the variance added by image resampling. The exact standard error increased from 0.338 to 0.426 microvolts. This partition is conditional on the fixed matrices and sampling rule and requires no population model. Under a separate independent crossed-effects model, the moment standard error was 0.347 microvolts. At 40 people and 40 images per stratum, the expected bootstrap cell variance factor is 2.900625. The EEG data do not establish that model. The work provides a general NumPy audit, proof, and a complete EEG example. Exact enumeration, 100,000 conditional draws, four 20,000-dataset model checks, two continuous source replays, and a full numeric repeat passed. The established bootstrap method is credited. A previous complete-pool mean was reproduced; the new empirical result is the variance partition. SCI retains reported numerical results and figures but omits source epochs and scripts because the source terms conflict with SCI distribution terms, so the full numerical replay cannot be rerun from this package.

crossed bootstrap · variance decomposition · cell residuals · EEG · N170 · ERP CORE · image resampling