As an additional preprocessing step, temporal PCA was employed as an effective filter to reduce or eliminate persistent drifts. All OERP waveforms (41 participants, 49 sites, 4 conditions: 8,036 cases; −250 to 1,750 ms: 401 variables) were submitted to a covariance-based PCA, followed by Varimax rotation of all covariance loadings (Kayser & Tenke, 2003). The time course of the first extracted factor (55% explained variance) was characterized by a monotonic, virtually linear, increase from the baseline to the end of the recording epoch, which accounted for unsystematic drifts across conditions, sites, and participants. By virtue of the linear decomposition, this unsystematic variance was removed from the data by reconstructing the surface potentials from all but the first factor loadings and corresponding factor scores and the grand mean (i.e., summed factor loadings multiplied by corresponding factor scores plus grand mean waveform; cf. Sinai & Pratt, 2002; Tenke et al., 2011). The drift-corrected ERPs were re-referenced to linked mastoids (TP9/10) for comparison to prior OERP studies using linked ear lobes or mastoids as reference.
