Dear FSL experts,
This may come across as a basic question but I was wondering how correlation between FA and for e.g. a behavioural measure across groups should be done within the design and contrast matrices.
I have read previous posts that suggest collapsing the groups into a single EV and putting '1's throughout, then demeaning the behavioural measure and putting it as a second EV. For the contrasts, it would be 1 and -1 on a separate line.
However, what if I were to retain the groups as separate EVs and utilise a design similar to the 'Two-Group Difference Adjusted for Covariate' as seen in the FSLWiki? Specifically this would mean my contrasts would look something like -
group 1 > group 2 1 -1 0
group 2 > group 1 -1 1 0
positive beh effect 0 0 1
negative beh effect 0 0 -1
And then I would examine the last 2 tests to see if the behavioural measure would result in a +ve/-ve effect on FA across the groups.
What is the difference between these methods? I tried both and the results were vastly different so I am certain I am misunderstanding something here!
Thanks very much!
|