Figure 2.

Basic concepts of group analyses using temporal concatenation and multilevel decomposition in combination with ICA or SOBI. For temporal concatenation, data aggregation yields a horizontally elongated matrix on which the demixing matrix W can be estimated, assuming the same mixing process for all subjects. This, however, is not the case with multilevel decomposition since single-subject as well as group-level decomposition prior to ICA/SOBI not only reduce the number of variables, but also allow for some variability of the latent structure across subjects. Note that usually only a subset of the c*n vertically concatenated components (c = number of channels/components, n = number of subjects) enter final decomposition via ICA or SOBI.