Baseline coefficients fitted on early trials gave higher mean later-trial error than coefficient one in 40 processed ERP CORE N170 records

We tested transfer of a personal baseline coefficient from early to later EEG trials in 40 public ERP CORE N170 records using fixed source processing. The first 160 trials supplied training and the last 160 supplied test scores; the main artifact limit was 150 microvolts. The fitted rule had mean squared error 41.93 microvolts squared, compared with 24.94 for coefficient one, a mean increase of 16.99 (68.1%; 95% person bootstrap interval 1.71 to 41.66). The fitted rule had lower error in 23 of 40 people; the three largest increases supplied 90.3% of the net increase. All eight fixed checks had positive mean increases, although the interval for the strictest screen included zero. Both baseline rules had lower mean error than coefficient zero. This result concerns processed voltage in one cohort and does not establish neural accuracy. The source terms conflict: the dataset description says CC BY 4.0 and the source LICENSE says CC BY-SA 4.0. SCI therefore omits source files and epoch caches while retaining the results, analysis code, source hashes, and retrieval instructions.

EEG · event-related potentials · N170 · baseline correction · baseline regression · voltage prediction · transfer validation · bootstrap