Thanks a lot for your reply Anderson.
I have another question regarding the use of covariates in within and between subjects designs in PALM:
As far as I understood, age and gender do not need to be considered as covariates because they do not change between the measures, but what about intervention order?
Can I create a variable with intervention order to see whether it has effects in my designs, even when they are not significant between groups?
And motion measures such as FD? I am working with resting state data and I cleanes a lot before. There are not differences in FD between groups. Should I still include the mean FD per subject and intervention?
My last question is about the place where these nuisance variables should be: Should I include them in the within or between subjects design? If I include them in both, my design will be rank deficient, right?
I am looking forward to your useful comments
Thanks a LOT!
Yacila
Von: FSL - FMRIB's Software Library [mailto:[log in to unmask]] Im Auftrag von Anderson M. Winkler
Gesendet: Samstag, 21. Januar 2017 14:54
An: [log in to unmask]
Betreff: Re: [FSL] Linear or quadratic effects with permutations testing?
Hi Yacila,
Please, see below:
On 20 January 2017 at 10:06, Yacila Isabela Deza Araujo <yacila_isabela.deza_araujo@
tu-dresden.de > wrote:Dear FSL experts,
(especially Anderson),
I have two groups and three conditions. Anderson already provided me with a design and contrast template that can be used in Palm.
My main question is: Is it possible to have linear contrast (or quadratic) with permutation testing such as randomise or PALM? My conditions are antagonist, placebo and agonist, so I expected some linear effect more than a comparison between conditions.
Do you mean a quadratic contrast with these 3 conditions only? With only 3 values (even if many subjects), this will be prone to overfitting. Testing the pairwise differences between the 3 conditions, and these between the groups (as in the design you have) is probably more informative and less error prone.
If you want something linear as in the order A>B>C, then testing A>C leads to the same result (and you have this contrast already, so nothing to change).
My second question is whether is valid to investigate main intervention effects in the total sample, even when I know I have two groups . (It is a genetic marker, so I also expect differences)
Yes, it's ok, provided there are no group differences. Otherwise the test is still valid (i.e., "statistically valid") although it may not be informative.
All the best,
Anderson
Thanks a lot for the attention!
Yacila