Hi Alan,

It seems to me that your hypothesis is different than the original one. 
The original was asking for group differences, whereas yours is asking, 
in a single group, if time (or age) would have an effect.

For yours, I'd suggest to have a single column with the age of the 
subjects at the time of the acquisition of the image, or the acquisition 
timings (as in the example), plus the subject-specific regressors 
(intercepts). When calling randomise, supply the .grp file (one group 
per subject), and don't use the option --permuteBlocks. The reason is to 
allow shuffling within-subject (so as to break the potential 
relationship between age/timing and striatum shape), rather than across 
subjects, which would keep that relationship unaltered (but which would 
be good to compare groups).

 From the description, I think you can keep all as it you've already set 
up, just dropping the --permuteBlocks option.

One thing I didn't understand, though: you have 70 or 42 subjects? And 
you you have the two timepoints for all subjects?

Anyway, hope this helps!

All the best,

Anderson



Am 26.02.14 12:47, schrieb Alain Imaging:
> Dear list,
>
> I took inspiration from Anderson's mail below, in order to perform a longitudinal analysis on shape data obtained using first.
> I have 70 subjects with 2 images (at 2 years of distance) each. I wanted to see if age have an influence on the shape of the striatum.
> So, the matrix I fed in randomise was very similar to the one proposed by Anderson in the mail below, with the difference that I have only one column for time (and of course 42 columns for the subjects' intercepts). Another difference, but I think it should not affect the analysis, is that I have my subjects divided by wave of testing, that means that I have Subject1T1,Subject2T1,...Subject70T1,Subject1T2,Subject2T2,...,Subject70T2. Of course I created the .grp file accordingly.
> The contrasts in the .con file were simply [1 0 0 ..] and [-1 0 0...]
> So, I made randomise turn with the call
>
> $ randomise -i ReconMNI.nii.gz -m ReconMNI_mask.nii.gz -o Randomise_Native -d design.mat -e group.grp -t T.con --permuteBlocks -T -D -N
>
> With my great surprise, I found no significant correlation, neither positive nor negative, between age and striatum shape. This is not only in contrast with the literature, but also with cross-sectional correlation that I performed on the very same data.
>
> Is this the real result of factoring out the within subjects variance, or have I committed a mistake in the design matrix/randomise call ?
>
> Any answer and tip is much appreciated
>
> Alain
>
>> Dear Jones,
>> Here is how I would do: First, use as the time marker a number that closer to what your hypothesis is, e.g., if the visits at 0, 6 and 15 months are to assess the >efficacy of a treatment applied before or at t=0, these time indicators are probably fine. If the visits are to assess, e.g., the effect of age, then perhaps you could >replace these values for the actual age of the subjects. You can use years with decimal places, or express the age in months, or maybe even in weeks if it's for >preterm or newborns. The design matrix can be arranged as:
>> (Data)    (Grp)    (Time,controls)    (Time,patients)    (Intercept,subj1)    (Intercept,subj2)    (Intercept,subj3)    (Intercept,subj4)
>> (Subj1,Visit1)    1    a11    0    1    0    0    0
>> (Subj1,Visit2)    1    a12    0    1    0    0    0
>> (Subj1,Visit3)    1    a13    0    1    0    0    0
>> (Subj2,Visit1)    2    a21    0    0    1    0    0
>> (Subj2,Visit2)    2    a22    0    0    1    0    0
>> (Subj2,Visit3)    2    a23    0    0    1    0    0
>> (Subj3,Visit1)    3    0    a31    0    0    1    0
>> (Subj3,Visit2)    3    0    a32    0    0    1    0
>> (Subj3,Visit3)    3    0    a33    0    0    1    0
>> (Subj4,Visit1)    4    0    a41    0    0    0    1
>> (Subj4,Visit2)    4    0    a42    0    0    0    1
>> (Subj4,Visit3)    4    0    a43    0    0    0    1
>> The timings for each subject and visit is indicated as "a", which could be the 0-6-15 months as you have or the age at scan. The contrasts to compare the two >groups are [1 -1 0 0 0 0] and [-1 1 0 0 0 0]. The contrasts to test if the slopes for each group are positive (increase in FA over time) or negative (decrease in FA) are >[1 0 0 0 0 0], [-1 0 0 0 0 0], [0 1 0 0 0 0] and [0 -1 0 0 0 0].
>> Define a file with the exchangeability groups as in the column "Grp" above, i.e., one group per subject. When running randomise, include the option >--permuteBlocks, so that the subjects are permuted while keeping the order of the visits.
>> All the best,
>> Anderson