Event-Related Analysis

Level 3 turns cleaned continuous data into an event-related potential you can defend. It runs in seven lessons from cutting epochs around events to stating precisely what a cluster-based permutation test does and does not license: choosing an epoch window and a baseline, and knowing what the baseline subtraction assumes; averaging, the square-root law that governs it, and the canonical components with the ERP CORE paradigms that elicit them; measuring amplitude and latency without letting noise or a post-hoc window do the work; reading topographies for what they contain rather than for what the interpolator drew; computing the standardized measurement error and turning it into a trial count and a design; looking at the trials themselves through ERP images, single-trial regression and deconvolution; and testing the result.

Two threads run through the level. The first is that every number is a choice — epoch window, baseline interval, measurement method, measurement window, electrode set — and that each of those choices has to be fixed before the effect is visible, or reported as having been made afterwards. The second is measurement precision as a first-class quantity: the standardized measurement error is what connects trial counts, rejection imbalance, the bias of peak measures and the power of the final test, and it belongs in the methods section next to the trial counts. The capstone replicates a published ERP CORE effect with the pipeline from Level 2, unchanged.