Uniform trial pairs from each stored condition increased mean CPz classification accuracy in 31 screened ERP CORE adults with a rule fitted on other people

We tested condition decisions from one trial and from the mean of two different trials. The primary data were 2,079 CPz trials from 31 screened ERP CORE adults. The fixed score was mean voltage from 300 to 500 ms after the target. A Gaussian rule was fitted on all other included people. We evaluated all 38,499 pairs of different trials within each person and stored condition. Balanced accuracy increased from 63.41% for one trial to 67.33% for a pair. The gain was 3.92 percentage points. The 95% person bootstrap interval was 2.01 to 5.25 points. Each bootstrap sample refitted the rule and excluded all copies of the test identity. Pairs improved accuracy for 27 people and reduced it for four. All seven planned checks had positive gains. Known condition labels defined the pairs. Each pair used two measurements. The test excluded people from fitting but did not exclude word items or test a new cohort. The SCI package retains analysis code, result ledgers, validation records, bootstrap samples, and the figure. It omits source event records, copied NEMAR metadata, and cached epochs because archived source notices conflict. Separate checks passed. A full repeat reproduced 18 numerical result files byte for byte.

event-related potentials · N400 · classification · balanced accuracy · trial averaging · bootstrap