Determination of the number and waveshapes of event related potential components using comparative factor analysis.

This paper applies new factor analysis methods to determine empirically the number and waveshapes of ERP components. It addresses several well-known problems in previous applications of factor analysis or principal components analysis (PCA) to ERP data, such as the assumption of simple structure by varimax rotation, low noise levels required by PCA, response latency variation, and background EEG (correlated noise). ERPs are modeled as consisting of nonstationary (response) components, stationary (random-phased EEG) processes, and independent noise. The analysis indicated that there was one main response component for the datasets analyzed, and that the EEG processes accounted for a major portion of the variance. Response components corresponded to the first and second derivatives of the dataset mean. This finding is due to response latency variation. Once the number and waveshapes of ERP components have been determined, source localization methods can be applied with much greater accuracy.