Build a single-trial EEG, inject artefacts, and watch what averaging does — and does not — remove.
Higher coherence means stronger, faster evidence — the DDM drift rate.
Each trial adds independent noise. Averaging cancels it as 1/√N.
Turn these on to see what the average looks like before cleaning.
Red is a single contaminated trial; blue is the average of 40. Negativity is plotted downward, following the usual convention.
What to watch. Raising evidence compresses every latency and scales the P300 — the signature a diffusion model predicts. Artefacts are trial-locked, so averaging does not remove them: blink, muscle, and hum all survive the average, which is why they must be removed before averaging rather than after.