Hi Yuchuan,

Please see below:


On 26 August 2015 at 17:21, Yuchuan <[log in to unmask]> wrote:
Dear Experts,

I have a very similar case with Romain (I have two timepoint measurements for each subject and I'm interested in the difference between these two timepoints). The GLM experimental design for one-group paired t test on the website is a little different with what Romain made. The Group column in Romain's design is "1 2 3 ..... 1 2 3 ....." while on the website is "1 1 1 .... 1 1 1 ....". Will this kind of design make a difference for the following randomise test?


Using the same design as in the GLM manual for the paired t-test (that is, without subtractions) but using randomise, then the "design.grp" file needs to be supplied, and this file should be coded as 1, 2, 3, ..., 1, 2, 3, .... These are the exchangeability blocks, being one per subject, and permutations will happen within-block (within-subject) only.

The case for 1, 1, 1, ... would be for running in FEAT, but with TBSS, you'd use randomise instead.

 

And just want to make sure I understand correctly, as long as the experimental design matrix are set up like this, and the design.grp to randomise with -e is included, I can directly use the "all_FA_skeletonised.nii.gz" generated by TBSS as the input of randomise. I don't need to manually subtract measurements for each subjest such as

"fslmaths subject1_rA -sub subject1_rB sub1_diff
fslmaths subject2_rA -sub subject2_rB sub2_diff
(repeat for subjects 3-8)
fslmerge -t runA_minus_runB sub1_diff sub2_diff sub3_diff sub4_diff sub5_diff sub6_diff sub7_diff sub8_diff"

The reference webpage link: http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Single-Group_Paired_Difference_.28Paired_T-Test.29



Exactly, no need for the subtractions as long as the design is the full paired t-test, and the exchangeability blocks are defined as above. However, if you want, you can do the subtractions and use sign-flippings in a 1-sample t-test. The result will be the same.

All the best,

Anderson


 
I hope I make my question clear.

Thanks for your help.
Best,
Yuchuan


==============The following is the email of Romain's design and Anderson's reply =========================
Hi Romain,

This design is fine as long as you supplied the design.grp file to randomise with the option -e. There's no need to run again doing subtractions, as it'd be equivalent to what was already done.

All the best,

Anderson


2015-01-21 14:09 GMT+00:00 Romain Quentin <[log in to unmask]>:
Dear experts,

I'm analysing FA images with TBSS. I have two scan per participants (run1 and run2).
I did a single-group paired difference with a design.mat like that.

Group EV1 EV2 EV3 EV4 EV5 ....
1          1      1    0     0     0
2          1      0    1     0     0
3          1      0    0     1     0
.           .       .     .      .     .
.           .       .     .      .     .
1         -1     1    0     0     0
2         -1     0    1     0     0
3         -1     0    0     1     0
.           .       .     .      .     .
.           .       .     .      .     .

On the Glm wiki, you advice to make the substraction manually with fslmaths and run one-sample Ttest.
May I split my all_FA-skeletonised.nii.gz with all images (s1run1, S2run1..., s1run2, s2run2) and substract s1run1 with s1run2, s2run1 with s2run2, ... etc.
Or it is better to re-start from original FA images?
I hope to be clear. Thanks a lot for your help.
Best,
Romain