> As I understand it, for each voxel the program computes the correlation across the
> time domain, and then sums across the space domain.
Yes, that's right.
> If this is the case, then the temporal discontinuities would create jumps in the data,
> thus increasing the shared variance.
The correlation between the time series at any two voxels is basically (assuming the time series have been standardized for simplification)
sum(X
i Y
i) / (n-1).
In other words, the correlation coefficient roughly measures the overall synchronization between the two voxels, and it does not assume any temporal continuity in the time series.
Gang
Edited 1 time(s). Last edit at 01/21/2015 01:34PM by Gang.