We can think of the analysis task as being to separate the measured signal into (1) the signal variations we are interested in (the signal), (2) the signal variations we can explain but are not interest in (the confounds or covariates), and (3) the signal variation that we cannot explain (the error).  e signal intensity time-course from a single voxel can be a sum of all three of these sources of variation, or it might not contain the signal we are interested in. Prior to doing statistical testing, ideally we would want to remove the e˜ects of confounds as much as possible, by means of the preprocessing, discussed below.