Dear Alex,
A short correction: In the current version of the code, the modes are not based on SVD of channel data as I had previously written, but on SVD of the lead-fields. There is, however, an option to do CVA with both data and lead-fields to find the spatial subspace that best represents the data (see spm_dcm_eeg_channelmodes). If you want to use CVA to inform the SVD of the lead-fields by the data you can specify DCM.options.CVA = 1 (in SPM12b see spm_dcm_erp, lines 178-182).
As for using DCM.H and DCM.R to calculate the proportion of variance explained, I should have noted that DCM.H+DCM.R will give you data after dimensionality reduction. However, you can still use your channel data to calculate model fits, or see how much of the observed data is preserved after dimensionality reduction.
Apologies for the confusion,
Best wishes,
Ryszard
|