Doubts on the adequacy of the principal component varimax analysis for the identification of event-related brain potential components: a commentary on Glaser and Ruchkin, and Donchin and Heffley.

Glaser and Ruchkin (1976) as well as Donchin and Heffley (1978) have proposed principal component varimax analysis (PCVA) for identification of event-related potential (ERP) components. In the present commentary, the assumptions on the ERP data structure of this procedure are considered. They are formalized and compared with a simpler and more restrictive hypothesized ERP data structure containing autoregressive properties. Referring to the principle of parsimony, the comparison of the two models shows the superiority of the hypothesized autoregressive structure: to fit a set of ERP data equally well, the autoregressive structure does so with considerably more parsimonious assumptions, i.e. with fewer parameters to be estimated, than the structure induced by PCVA. It is concluded that an appropriate model for the ERP data structure should take the autocorrelated nature of the ERP into account.