Temporal PCA is one of the most frequently used multivariate decomposition approaches for ERP data and has been repeatedly shown to be superior to more traditional ERP measures, for instance, revealing more robust F statistics and better reliabilities (i.e., internal consistency and temporal stability) when directly compared with integrated time windows or baseline-to-peak measures (e.g., Beauducel, Debener, Brocke, & Kayser, 2000; Kayser et al., 1997; Kayser, Tenke, & Bruder, 1998). Often-cited limitations, such as misallocation of variance because of latency jitter, are not restricted to the use of temporal PCA but also affect traditional component measures and more severely (e.g., Beauducel & Debener, 2003; Chapman & McCrary, 1995; Donchin & Heffley, 1978; Wood & McCarthy, 1984). With careful consideration, temporal PCA can provide a concise and unbiased summary of ERP/CSD activity (Kayser & Tenke, 2003, 2006a) associated with generator patterns underlying stimulus processing, even for slow and long-lasting components (e.g., Kayser et al., 2006), and could therefore provide an answer to the question of relative statistical independence between putative olfactory components (Lorig, 2000). Moreover, because the extracted CSD factors are independent of the recording reference, they have an unambiguous component polarity and topography.
