Hi, Anderson,

Thanks so much for so insightful comments.

I now get to know the real definition "group" in FSL. This opens a new question: how to differ the real "experimental group" in the FSL, then?  Is there any a good method for this?

Thank you very much,

Sim Luck

On Tue, Feb 21, 2017 at 4:26 AM, Anderson M. Winkler <[log in to unmask]> wrote:

Hi Sim Luck,

On 20 February 2017 at 22:59, Sim Luck <[log in to unmask]> wrote:

Hi, FSL experts,

I am a beginner with FSL.  Previously I asked one question on the independent t-test with FSL (e.g., patients, healthy controls). Two experts suggested me simply using the "1" as "Group" for both patients and healthy controls. I agree as if I used "1" for one group and "2" as the other group, I did not observe any effects between the groups. I was thinking that this FSL approach is different from the common behavioral data analysis. 

The information from the "Group" setting is stored in a file named something as "design.grp", and this file takes different roles depending whether the analysis is used in FEAT, in PALM, or in randomise. In all three cases, the "Group" does not refer to "experimental group". Instead, it refers to "variance group" (VG) in FEAT, to "exchangeability block" (EB) in randomise, and to either VG or EB in PALM, depending on whether it's supplied with the options "-vg" or "-eb".

In FEAT, unless there is reason to suspect (or assume) that the variances for the groups are different, all subjects should belong to group "1". If variance groups are defined, then the design must be "separable", that is, the design must be such that, if the observations are sorted by group, the design can be constructed by direct sum (i.e., block-diagonal concatenation) of the design matrices for each group separately.

In randomise, the same "Group" setting can be supplied with the option "-e design.grp", defining exchangeability blocks. Observations within a block can only be permuted with other observations within that same block. If the option --permuteBlocks is also supplied, then the EBs must be of the same size, and the blocks as a whole are permuted.

In PALM, using "-eb design.grp" has the same effect as "-e design.grp" in randomise. Further using "-whole" is equivalent to using "--permuteBlocks" in randomise. However, it's also possible to use "-whole -within", meaning that the blocks as a whole are shuffled, and further, observations within block can be shuffled. In PALM the file supplied with the option "-eb" can have multiple columns, indicating multi-level exchangeability blocks, which are useful in designs with more complex dependence between observations. Using "-vg design.grp" causes PALM to use the v- or G-statistic, which are replacements for the t- and F-statistics respectively for the cases of heterogeneous variances.

In both FEAT and PALM, using "group" as VG will only make difference if such variance groups aren't balanced.

 

Now I want to do the paired t-test (the same patients were tested two times, before the treatment, and after the treatment). My quick thought is that of course, I should use the "1" as group for both sessions ("before" session and "after" sessions). But my problem is that I did not find any effect of treatment - this does not make sense to me. So I want to confirm with the FSL experts to see if I am wrong on anything. 

For the paired t-test in FEAT, you can leave the "Group" column as all ones. The same test in PALM or in randomise requires one exchangeability block per subject, so you'd use the "Group" column to define the EBs, and supply them with "-e design.grp" (randomise) or "-eb design.grp" (PALM).

Hope this clarifies the matter!

All the best,

Anderson

Here I attach some details in 2 slides - see the attachment. Could any experts advise me, if I am incorrect on anything?

Thank you very much,

Simi Luck