Early N400 scores did not reduce observed prediction error for later scores in 18 screened ERP CORE adults in one visit

We tested whether an early N400 score helped predict a later score for the same person. The source was ERP CORE in NEMAR nm000132. We fixed the plan before signal extraction. We split each of 40 adults’ tasks into the first 60 and last 60 targets. The score was the mean for unrelated targets minus the mean for related targets at CPz, from 300 to 500 ms. The reference was P9/P10. A fixed artifact and response screen retained 18 adults and 1,468 trials. For each test person, both models were fitted only on other people. The group-mean model had mean squared error of 26.731 squared microvolts. The model with the early score had error of 28.014 squared microvolts. Gain was -1.283 squared microvolts, with a 95 percent person-bootstrap interval of -11.572 to 3.114. The early score did not reduce observed error. The interval still allows useful gain. Both halves showed a clear group N400 contrast. All eight fixed sensitivity cases had gain intervals that included zero. This is one visit and one screened sample. It does not test stability across days or prediction in a new cohort. The package preserves the result tables, predictions, sample indices, checks, and figures. Source-derived epoch and event inputs are omitted because the archived ERP CORE terms conflict with NEMAR’s recorded license; the source URLs and hashes remain documented.

N400 · event-related potentials · EEG · prediction error · ERP CORE · NEMAR