The Problem with Time: Application of Partial Least Squares Analysis on Time-Frequency Plots to Account for Varying Time Intervals with Applied EEG Data.

When attempting to study neurocognitive mechanisms with electroencephalography (EEG) in applied ecologically valid settings, responses to stimuli may differ in time, which presents challenges to traditional EEG averaging methods. In this proof-of-concept paper, we present a method to normalize time over unequal trial lengths while preserving frequency content. Epochs are converted to time-frequency space where they are resampled to contain an equal number of timepoints representing the proportion of trial complete rather than true time. To validate this method, we used EEG data recorded from 8 novices and 4 experts in veterinary medicine while completing decision-making tasks using two question types: multiple-choice and script concordance questions used in veterinary school exams. The resulting resampled time-frequency data were analyzed with partial least squares (PLS), a multivariate technique that extracts patterns of data that support a contrast between conditions and groups while controlling for Type I error. We found a significant latent variable representing a difference between question types for experts only. Despite within and between subject differences in timing, we found consistent differences between question types in experts in gamma and beta bands that are consistent with changes resulting from increased information load and decision-making. This novel analysis method may be a viable path forward to preserve ecological validity in EEG studies.