Print

Print
Hi Isik,

Please see below:


On 16 February 2017 at 14:00, user user980 <[log in to unmask]> wrote:
Hi again,

Thank you for the design. I have one more question. when I test the age effect in one condition only, will only the age regressor (EV 17-18) in last design change such that  I fit the slope only for that condition? 

To have the age effect in just one condition, as opposed to the average of the three conditions, you'd have to change the last design to "knock out" the other two conditions; this could be done by changing the subject specific EVs: for each subject, where currently there is a matrix block as:

-1 -1
1 0
0 1

You'd replace for:

0 0
1 0
0 1

or

1 0
0 0
0 1

or

1 0
0 1
0 0

depending whether you want, respectively, condition A, B or C. Perhaps it'd be simpler, though, to just ignore the other 2 conditions and run a design like this one, from the manual: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two_Groups_with_continuous_covariate_interaction
This is something similar to what you propose below.
 

And actually what if I feed each condition as different modalities in Palm and compute the age effect per condition, how do these two models differ? I am assuming the full model is a bit more relaxed since it allows permutations within condition...

For the between-subject comparisons, it can be done separately for the 3 conditions, as each were a different modality, and correcting across with -corrmod. The between-subjects design shown in the spreadsheet is for the average of the 3 conditions, so the answer will be different.
If you take the "knock out" approach above, and correct for the contrasts of the (then) 3 different designs using -corrcon, the result should be the same as doing as separate modalities and using -corrmod.

All the best,

Anderson

 

thanks
isik

On Feb 16, 2017 6:09 AM, "Anderson M. Winkler" <[log in to unmask]> wrote:
Hi Isik,

Thanks for sending in .ods format. For within-subject effects, given that age doesn't change between the three conditions, it doesn't have to be included in the model. The subject-specific EVs already take care of anything that is subject-specific, such as age. So, age can be omitted.

Unless in the scenario in which age (or some other variable that is constant within subject) could be affect the differences between the conditions, that is, an interaction age by condition, which is what you'd like to test. Then it's different.


There are 4 different designs inside:
- design1: within-subject effects, no age
- design2: between-subject effects, no age
- design3: within-subject effects, with age and interactions
- design4: between-subject effects, with age and interactions

Answering the earlier questions:

1) The interactions make demeaning a bit confusing. It's needed for some, not needed for others, but where not needed, it doesn't affect. So, to make it simpler, do the mean centering.
2) You can if you want. Just test the contrast [1 1 0 0 0 ... ]
3) Looks correct: should be one EB per subject for this design.

For the call to PALM: need to add "-t design.con". Also, I recommend using -logp, as the figures later will look more informative, with better contrast. Significant values are those above -log10(0.05) = 1.3.

Optionally you can accelerate the run with "-approx tail -n 500 -nouncorrected".

Hope this helps!

All the best,

Anderson


On 15 February 2017 at 14:49, Isik Karahanoglu <[log in to unmask]> wrote:
Oh sorry about that, I attach the .ods.