 |
 |
Dear Anderson,Thank you so much, this is really precious help!One small last thing: in order to compare C1, C2 and C3 would it be valid to addEV1 EV2 EV32 -1 -1-1 2 -1-1 -1 2to my list of contrasts or would these not work appropriately in my group-level PALM analysis?Thank you very much again!Best wishes,Lucie2018-08-02 23:56 GMT+02:00 Anderson M. Winkler <[log in to unmask]>:Dear Lucie,Please see below:Dear all,
Even if this thread already dates back some time I would like to ask a follow-up question on the different ways of tackling repeated-measures designs in FSL’s PALM:
1) Would the following approach be valid if used in PALM and specifying exchangeability blocks per participant (i.e. shuffling only within data of the same participant but not across different participants)?
As an example, a design where I have 3 participants, for each of which I obtained data in the 4 conditions of my fixed factor.
design ANOVA_rm.mat
eb GROUP cond1 cond2 cond3 sub01 sub02 sub03
1 1 1 0 0 1 0 0
2 1 1 0 0 0 1 0
3 1 1 0 0 0 0 1
1 1 0 1 0 1 0 0
2 1 0 1 0 0 1 0
3 1 0 1 0 0 0 1
1 1 0 0 1 1 0 0
2 1 0 0 1 0 1 0
3 1 0 0 1 0 0 1
1 1 0 0 0 1 0 0
2 1 0 0 0 0 1 0
3 1 0 0 0 0 0 1
…
As far as the EBs are concerned, this is fine. Design is also ok, but see below.Contrasts ANOVA_rm.con
Title EV1 EV2 EV3 EV4 EV5 EV6
Cond1 1 0 0 0 0 0
Cond2 0 1 0 0 0 0
Cond3 0 0 1 0 0 0
Cond1>2 1 -1 0 0 0 0
Cond1>3 1 0 -1 0 0 0
…
The design is fine but contrasts 1, 2 and 3 don't really measure the effect of cond1, cond2 or cond3, but instead, the value of cond1-cond4, cond2-cond4, and cond3-cond4, respectively.palm -i group1234.nii -d ANOVA_rm.mat -t ANOVA_rm.con -f ANOVA_rm.fts -Cstat mass -T -C 3.1 -Cnpc 3.1 -npcmethod Fisher -npccon -corrcon -logp -fdr -zstat -o out/ANOVA_rm -eb.csv -within
I think there is no real point in combining contrasts using NPC (or even with NPC, using Fisher). Instead, correction across contrasts (-corrcon) is what really matters, and it obviates the need for an F-test (or any other "omnibus" inference) across contrasts. So, I'd drop NPC here.2) Using exchangeability blocks to allow shuffling only within subjects controls for repeated measures here. Hence, would this approach be valid? Or will all differential contrasts (i.e. contrast 4 and 5 in this example) need to be assessed via the calculation of simple difference copes (fslmaths) and a simple t-test? What about the correction for multiple comparisons across all contrasts in that case?The inference is within subjects (between conditions), so shuffling within subjects is what is needed. There is an assumption, though: compound symmetry. It may or may not hold.
3) Can the PALM input files consist of concatenated inter-subject correlation 3D maps (nifti format), one per participant pair (which would be used equivalently to ‘participant’) and condition?Yes, this is fine. However, correlation maps are bounded between -1 and 1 (or 0 and 1), causing variance to shrink towards the extremes, A simple workaround is to do a logit, probit, or Fisher's r-to-z transformation.Hope this helps!All the best,Anderson
I would be really grateful for your help!
Lucie
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1