A fixed laboratory P3 score reduced mean squared campus prediction error by 4.6% in 35 Runabout records

We tested whether a laboratory P3 attention score could predict a campus score. We used OpenNeuro ds003620 version 1.1.1. A fixed signal rule retained 35 of 36 processed records. For each prediction, a linear model used the other 34 people. The comparison model used their campus mean. Mean squared error was 0.9136 versus 0.9581 microvolts squared, a 4.64% reduction. A test with 9,999 label permutations gave p = 0.0672. Pearson r was 0.325, with a 95% interval from -0.010 to 0.594. All fixed score-choice checks are reported. They gave different gains. The data do not establish stable prediction utility. One source record had no trials in two required bins and was excluded. The package contains exact signal arrays, source hashes, code, scores, predictions, null results, and reproduction instructions. Independent calculations and a separate reproduction run passed. The result applies to this score and one cohort in one session.

EEG · P3 · attention · prediction · Runabout · naturalistic environments · cross-validation