A procedure for using multi-electrode information in the analysis of components of the event-related potential: vector filter

This paper describes a procedure, Vector filter, based on a multiple regression model, which decomposes the event-related brain potential into components on the basis of scalp distribution. It is assumed that the voltage values observed at several electrode sites at any point in time are given by the linear combination of a set of components and background noise and that the scalp distribution of each component is invariant and known. Each component's scalp distribution is expressed by a set of weights, one for each electrode. The amplitude of the component, at any point in time, is then derived using a least squares criterion. Unlike other component decomposition procedures, Vector filter can be applied when the latency of a component varies as a function of trial, condition, or subject population. We review two problems in the use of the procedure: the selection of the set of components, and the derivation of the scalp distribution of each of the components. Two applications of the procedure are discussed: identification of the component structure of an event-related potential waveform, and filtering of a waveform for a particular component.