Causality analysis of multivariate neural data

With the rapid advances in multielectrode recording and brain-imaging techniques that have occurred in recent years, multichannel data sets have become increasingly obtainable. While standard signalprocessing techniques such as crosscorrelations in the time domain and coherence in the frequency domain remain the main statistics for assessing interactions among these multichannel data, it is increasingly felt that these symmetric interdependence measures are no longer su®cient for many intended applications, and further partitioning of relationships among a set of simultaneously recorded signals is needed to parcel out the functional connectivity of complex neural networks. Recent work has begun to explore a class of techniques called Granger causality as a potentially useful addition to the current analytical repertoire in the attempt to add directionality to neural interactions.