The above described methods are univariate methods since they are designed to assess connectivity between pairs of univariate signals, e.g., MEG signals at sensor level. When dealing with brain activity estimated from MEG signals, e.g., according to one of the strategies described in section Source space MEG, the most general situation results into three brain activity components (also termed dipole orientations or dipole moments) at each location of the discretized brain. To apply univariate connectivity methods on these three-dimensional activity components, it is necessary to perform a dimensionality reduction to one-dimensional signals which might result in a loss of information and in a subsequent inaccuracy in the estimation of connectivity. Conversely, multivariate connectivity methods are specifically designed to take into account the whole information contained in the multidimensional nature of the input signals, i.e., they do not require a dimensionality reduction of the input signals. In the following subparagraph, a review of multivariate methods for MEG connectivity is provided.
