The data was pre-filtered in order to isolate the relatively weaker higher frequency theta activity from the lower frequency delta activity for the PCA (cf. Bernat et al., 2011). The theta and delta filter cutoffs were chosen based on visual inspection of the unfiltered averaged TF energy representation across all subjects and stimuli. The separation of the prominent lower frequency component and the higher frequency activity in the N2-P3 time window is evident in the unfiltered averaged TF representation in Figure 2, and occurred at 1.5 Hz. The condition averaged stimulus-locked signals were filtered independently using separate highpass and lowpass 3rd order Butterworth filters set at 1.5 Hz for the high frequency and low frequency data, respectively. Next, the filtered signals were transformed into TF energy representations using the binomial RID variant of Cohen's class of TF transformations using the full epochs to provide sufficient data to resolve low frequency activity (Bernat et al., 2005). PCA was applied to the TF transforms of the theta and delta filtered signals separately. This TF-PCA approach (based on the covariance matrix with a
