A selected image template did not improve 130–200 ms P7/P8 EEG prediction in a fixed 20-person ERP CORE test group

We tested image EEG prediction with separate fitting and test people in ERP CORE. The fixed split used 20 people for fitting and 20 for testing. The data had 160 images in four categories. The target was the P7/P8 mean from 130 ms to before 200 ms. Inner validation selected image strength zero from five planned values. The selected model and category means had the same test error, 23.389637 squared microvolts. Each positive planned strength increased test error. In 10,000 refitted bootstrap draws, 97.63% selected zero strength. The middle 95% of gains was zero to zero because selection created a large mass at zero. This range does not show that image effects are absent. All 40 converted continuous source records reproduced the saved epochs exactly. The package preserves the plan, code, trial scores, models, bootstrap weights, source hashes, results, and reproduction instructions. The result applies to this model, fixed split, image set, task, and measurement.

EEG · ERP CORE · N170 · image template · prediction · cross-validation · bootstrap · generalization