Dynamic Factor Analysis in the Frequency Domain: Causal Modeling of Multivariate Psychophysiological Time Series

Factor analysis of a multivariate time series by means of frequency domain or spectral techniques does not yield a causal model of the time series. We propose a solution to this long-standing problem, namely unitary rotation to minimum phase-lag in the frequency domain. The method is illustrated with applications to simulated and real multivariate time series. The latter applications involve topographic analysis of the multichannel EEG, including a comparison with time-varying dipole modeling of brain potentials.