Hi Graham,
Please, see a suggested design for your study here:
https://dl.dropboxusercontent.com/u/2785709/outbox/mailinglist/design_graham.ods
Also, please see below:
On 12 January 2016 at 11:35, Graham David <[log in to unmask]> wrote:
> Dear FSL Experts,
>
> I have a 2 x 3 mixed design. I have 2 sessions, so TIME will be a repeated
> measure. I have 3 groups, so GROUP will be a between subjects measure. I
> will use the model described in this post
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;a3eedcc7.1501 (also
> attached).
>
> I'm struggling a bit with the contrasts. Please can you advise if I have
> got this right and help with the blanks?
>
> Effect of TIME
> [1 0 0]
> [0 1 0]
> [0 0 1]
>
Yes, this is fine.
> F-test combining these 3.
>
Yes, this is fine but it will test for any effect of time (any direction)
for any of the groups, so it's a quite unspecific contrast. Instead,
perhaps consider:
[1 1 1]
[-1 -1 -1]
>
> Effect of GROUP
> ????
>
This requires a different design, given the repeated measurements. Please
see the "design2" in the spreadsheet above. This will compare the groups
regarding the average of their timepoints. For the comparison of the time
differences, this would be the interaction as below.
>
> Interaction 1.
> A > B
> [1 -1 0]
>
> Interaction 2.
> A > C
> [1 0 -1]
>
> Interaction 3.
> B > A
> [-1 1 0]
>
> Interaction 4.
> B > C
> [0 1 -1]
>
> Interaction 5.
> C > A
> [-1 0 1]
>
> Interaction 6.
> C > B
> [0 -1 1]
>
Yes, all fine, for timepoint2 > timepoint1, as coded in design1 in the file
above.
>
> I have also read that I need to use --permute blocks for some contrasts.
> Can you advise which?
>
For design1, use within-block permutation ("-e design.grp").
For design2, use whole-block permutation ("-e design.grp --permuteBlocks").
In case you instead use PALM, you can correct for all these contrasts by
running synchronised permutations with both designs in the same run and the
options "-eb design.grp -whole -within".
Hope this helps.
All the best,
Anderson
>
> Kind regards,
>
> Graham.
>
>
>
>
