The effect of estimation time window length on overlap correction in EEG data

Abstract

Regression Event-Related-Potentials (ERPs) with overlap correction (also referred to, as linear deconvolution, or temporal response functions) are becoming more popular for the analysis of Electroencephalography (EEG) data. A common question for the analyst is, how to specify the length of the estimation windows. Long estimation windows might capture all relevant event-related activity, but might introduce artifacts due to overfit, short estimation windows might not overfit, but also might not capture all (overlapping) activity, and thereby introduce bias. Using a systematic simulation approach, we show that longer rather than shorter time windows should be preferred for typical EEG designs. We further provide an interactive app to visualize various design parameters: https://estimationwindow.ccn2023.s-ccs.de