Hi Jay, In the paired t-test, these subject-specific EVs should definitely be included, because the test is strictly whether timepoint1 differs from timepoint2 for each given subject, and as such, the means for each subject must be discounted, which is what these EVs do. Another way of seeing this is to take the full line corresponding to one of the timepoints (say, t1) for a given subject, and subtracting it from both t1 itself, and also t2 (doing that for both sides of the equation). Doing this does not change the model, and leaves it with, in the left side, the difference in the measured values, and in the right side a constant in the main EV (this will be equal to 2 if it was coded as +1 and -1), and a zero in the place where the subject-specific EV was in the original, making it empty. Such empty EVs have no effect and are as if they didn't exist. It will also leave a row full of zeroes in both sides, which is as if it didn't exist. Repeating the same procedure for all subjects, leaves you with a one-sample t-test on the differences, with the exact same degrees of freedom as the original, and giving the same final result (using sign-flippings). That said, yes, it is still possible to remove these subject specific EVs from the original model and keep the shuffling within-subject (within-block). But then the hypothesis is a different one: it is that timepoint 2 is different than timepoint 1 across all subjects (i.e., when all are pooled), as opposed to within-subject. The shuffling within-block ensures that exchangeability is preserved. It is still a valid and exact test, but it tells something different about the data. And it can no longer be called a paired t-test. All the best, Anderson On 17 September 2015 at 12:59, SUBSCRIBE FSL Jay <[log in to unmask]> wrote: > Hello Anderson, > > I wanted to know your opinion to understand the design for longitudinal > analyses (paired two-sample t-test) mentioned at ( > http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Randomise/UserGuide#Two-Sample_Paired_T-test_.28Paired_Two-Group_Difference.29 > ) > > Can you please tell me the role of the subject specific regressors (used > to model the individual subject means)? What would be the difference if i > did not model for the individual subject means and still use the design.grp > file to pair them? > > BW > Jay >