Adaptive Granger causality measures for multi-trial data – A simulation study

Introduction: Event-related paradigms violate the stationarity assumption needed for fitting the required multivariate autoregressive model (MVAR) for Granger causality based measures. Instead, they require fitting of a time-variant, adaptive MVAR. Two potential solutions making use of multiple realizations of the same process (“multi-trial data”) have recently been proposed: a sliding-window approach [1] and a time-continuous fit via Kalman filtering [2]. A third, non-parametric, solution factorizes wavelet cross-spectral densities (CSD), thereby avoiding MVARs [3]. We assess the methods’ relative performance w…