Ability of Granger Causality Analysis to Detect Indirect Links: A Simulation Study

Abstract

Granger causality (G-causality) has emerged as a useful tool to investigate the influence that one system can exert over another system, but challenges remain when applying it to biological data. Specifically, it is not clear if G-causality can distinguish between direct and indirect influences. In this study time domain G-causality connectivity analysis was performed on simulated electroencephalographic cerebral signals. Conditional multivariate autoregressive model was applied to 19 virtual time series (nodes) to identify the effects of direct and indirect links while varying one of the following variables: the length of the time series, the lags between interacting nodes, the connection strength of the links, and the noise. Simulated data revealed that weak indirect influences are not identified by G-causality analysis when applied on covariance stationary, non-correlated electrophysiological time series.