On Sat, Oct 5, 2013 at 6:18 PM, Aser A <[log in to unmask]> wrote:
Dear Donald and all,

Sorry for these naive questions. I would like to confirm and ask some questions regarding my previous message.
Just to summarize I scanned 8 subjects each in 2 sessions. Each session has three conditions A, B, & C.
Session one using the right hand and session 2 using left hand.


I would like to answer these questions :

1 - The differences between the conditions : for example A1 - A2, B1 - B2, & C1 - C2   then A2 - A1, B2 - B1, & C2 - C1

2- The activated regions that are common to both A1 & A2 ( I think this is called as the interaction ).

This is called a conjunction. It's not possible to compute it from this model. The reason its not possible is that A1>0 and A2>0 is a between-subject effect and the model can only test within-subject effects. I'd suggest a one-sample t-test for each condition, then threshold the maps, then binarize them, then add them together. Voxels with a value of 2 will have A1 and A2>0.
 

My questions are : If I use a flexible factorial design, should enter under the section specify subject 8 subjects with their 6 con images ( for right and left )?

Then in the condition section, should I enter 3 because of the 3 conditions that I have?

I would choose specify all. Then enter all the images at one time. Then you need to create a matrix with 4 columns. Column 1 will be all 1s. Column 2 is the condition number, column 3 is the side, column 4 is subject number.
** assumes, 1 2 3 is factor 1, and A/B are factor 2 and subject is factor 3.
 


After that, what should I enter in the section : main effect and interaction ?!

main effects: 1 2 3
interaction: 1 2
** assumes, 1 2 3 is factor 1, and A/B are factor 2 and subject is factor 3.
 

Thanks in advance and really appreciate your help


Aser














On Wed, Oct 2, 2013 at 3:45 AM, MCLAREN, Donald <[log in to unmask]> wrote:
On Tue, Oct 1, 2013 at 4:14 PM, fMRI <[log in to unmask]> wrote:
Hi all,

I am confused about which test or model I should select.

I have scanned 8 subjects. Each subject was scanned in two sessions, left and right hand. Each session consists of three conditions: A, B & C. So basically I have from each subject six conditions using both hands.

In the first level, i defined each condition in a separate regressor. Then I contrast each condition versus rest. So I have in each session : A>rest,      B>rest,    C>rest.

So I end up by three contrast images per session or six per the two sessions.

I am interested in comparing right versus left. So I would like to compare first A from session 1 versus A from session 2 and so on.

I thought first to use two sample t-tests by taking the con from each subject in session 1 and condition A and compare that to session 2. And then do the same for Conditions B & C.

Is this the right way to do it?

No. You only have one sample. To compare two conditions from the same subjects you need to use a paired t-test or create difference images for the two sessions and use a one-sample T-test. The results will be the same.
 

Or are there better options such as ANOVA? Can anyone please explain what should I select an how I should do this ?

Flexible factorial (see my previous posts) can be used to compare 2 or more conditions. This is basically a repeated measures ANOVA.
 

Another consideration that it might be of interest is that I may need then to compare A1 versus B2 or C2 or B2+C2.

I'm not quite sure why you would want to compare condition A from the left hand to Conditions B and C with the right hand. Nevertheless, for A1 vs B2 or C2, its paired t-test as above. If you want A1 vs B2+C2, I'd would advise you to compare A1 to B2+C2 instead as two conditions will usually be higher than 1. For A1 vs (B2+C2)/2 [which doesn't make sense due to different hands], you'd create  a contrast image for (B2+C2)/2 then use a paired t-test. You could also use a flexible factorial with the contrast of 1 -1/2 -1/2 as well.
 

I believe if I do want to answer this question I cannot do two sample t test. So what would be the better approach?

Thanks in advance!


Aser