A statistical comparison of EEG time- and time–frequency domain representations of error processing

Successful behavior relies on error detection and subsequent remedial adjustment of behavior. Researchers have identified two electrophysiological signatures of error processing: the time-domain error-related negativity (ERN), and the time-frequency domain increased power in the delta/theta frequency bands (~2-8 Hz). The relationship between these two signatures is not entirely clear: on the one hand they occur after the same type of event and with similar latency, but on the other hand, the time-domain ERP component contains only phase-locked activity whereas the time-frequency response additionally contains non-phase-locked dynamics. Here we examined the ERN and error-related delta/theta activity in relation to each other, focusing on within-subject analyses that utilize single-trial data. Using logistic regression, we constructed three statistical models in which the accuracy of each trial was predicted from the ERN, delta/theta power, or both. We found that both the ERN and delta/theta power worked roughly equally well as predictors of single-trial accuracy (~70% accurate prediction). Furthermore, a model including both measures provided a stronger overall prediction compared to either model alone. Based on these findings two conclusions are drawn: first, the phase-locked part of the EEG signal appears to be roughly as predictive of single-trial response accuracy as the non-phase-locked part; second, the single-trial ERP and delta/theta power contain both overlapping and independent information.