Hi Donald, Helmut,

Thanks Donald for your response. I have some questions below:

With 2 groups, I'd use 2 two-sample t-tests. 

So a two-sample t-test for the task effect (2 samples = young and old), and later, in a separate analysis, another two-sample t-test for the behavioural effect (2 samples = young and old), right?

For this analysis, you can only enter 1 contrast per subject per two-sample t-test.

For the second-level, I thought that when setting up the contrast in the contrast manager, you enter 1 contrast for the whole group analysis, not per subject? Or are you talking about the 'scans' that you input under 'Design' (ie the con*.nii images that are per subject, and come from the first-level analysis?)

For contrast creation. You want to use the same approach as any other model. Start with the null hypothesis. Make the null hypothesis equal to 0. Apply the coefficients to the model columns. Tasks not listed get a coefficient of 0. This becomes the contrast.

I'm quite new to this, first time doing an analysis, so bear with me :-)

So I suppose my 'columns' here are the groups?

Contrast becomes

[1 -1]

So, by setting up a 2-sample t-test, inputting the subject-wise con*.nii images from the first-level analysis (only .con*.nii images for the task effect), and setting up the contrast [1 -1], I am testing how the 2 groups differ with respect to task effect.

Then, I would repeat same process, save for the con*.nii's this time would be the con*.nii's from the first level analysis of the behavioural effect.

Thanks for your help!

Joelle

On Wed, Jun 10, 2015 at 7:50 PM, MCLAREN, Donald <[log in to unmask]> wrote:

I would do the group (2nd level) analysis separately for the task effect and behavioral effect. With 2 groups, I'd use 2 two-sample t-tests. For this analysis, you can only enter 1 contrast per subject per two-sample t-test.

For contrast creation. You want to use the same approach as any other model. Start with the null hypothesis. Make the null hypothesis equal to 0. Apply the coefficients to the model columns. Tasks not listed get a coefficient of 0. This becomes the contrast.

Ho: G1=G2

becomes

Ho:G1-G2=0

Contrast becomes

[1 -1]

Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Postdoctoral Research Fellow, GRECC, Bedford VA
Website: http://www.martinos.org/~mclaren
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at (773)
406-2464 or email.

On Wed, Jun 10, 2015 at 8:31 AM, Joelle Zimmermann <[log in to unmask]> wrote:

Hi Donald and Hulmut, and anybody else who may have suggestions...

I've mastered the first-level analysis, and now going to the second-level. I have an idea about how to set it up, but was hoping I can run the idea by you to verify. 

I'm interested in looking at the effect of the task (ie rest versus task) as well as the effect of behaviour - and comparing young and old subjects. Maybe I should look at the effect of task, and the effect of behaviour, in 2 separate contrasts.

In the first-level analysis, I have 9 columns in my design matrix:

1) the task effect (ie task vs rest)

2) the behaviour (put in as a PM)

3-8) the motion parameters from realignment

9) the constant

In this first-level analysis, I set up 2 separate t-contrasts:

1)    1 0 0 0 0 0 0 0 0  # tests effect of the task (ie con0001.nii)

2)    0 1 0 0 0 0 0 0 0 # tests effect of behaviour (ie con0002.nii)

So my idea for the second-level analysis:

1) Must first do a first-level analysis (as described above) for each subject

2) Second-level analysis Factorial Design Specification:

Do a two sample t-test? 

Forward the .con.nii images into the group model (by choosing these under Design>Scans)

3) Now, after setting up the Factorial Design Specification, and estimating, how can I set up the contrasts?

Would you recommend F contrast for the effect of task, and then separately another F contrast for the effect of behaviour? 

What contrast values would make sense?

If you have comments or ideas to any of my points, that would be much appreciated :)

Thanks,

Joelle

On Wed, Jun 10, 2015 at 10:16 AM, Joelle Zimmermann <[log in to unmask]> wrote:

Hi Donald - perfect thanks. That's what I needed to know. Right, I don't actually want to test which areas of the brain are correlated with motion, I just want to control for the motion.

Joelle

On Tue, Jun 9, 2015 at 6:27 PM, MCLAREN, Donald <[log in to unmask]> wrote:

Joelle,

You asked "I think that for sure I am interested in including the realignment motion parameters in my model. 

How can I set up a contrast that tests for these AND for the task/rest design for instance?

Or a contrast that tests for these AND the behavioural performance?"

Helmut provide the F-test as only the F-test would test for this. As he pointed out, its usually useless (and furthermore can't be taken to the second level).

In your case, I doubt you actually want to test for the areas of the brain that are correlated with motion; rather you simply want to control for the motion. There are three parts to the GLM:

(1) Setting up the model;

(2) Estimating the model; and

(3) Creating contrasts.

A step later in the model has no influence on the earlier stages as the process is done sequentially. When you estimate a model with motion parameters in the design, you are already accounting for their effect in the brain. When you do the contrast [1 0 0 0 ..] to test for the effect of the task, you are testing the effect of the task after controlling for the effects of motion.

Hope this helps.

Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Postdoctoral Research Fellow, GRECC, Bedford VA
Website: http://www.martinos.org/~mclaren
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at (773)
406-2464 or email.

On Tue, Jun 9, 2015 at 11:05 AM, Joelle Zimmermann <[log in to unmask]> wrote:

Hi Helmut,

Thanks so much! This is very helpful.

Why do you say that usually the contrast should be useless? Do you mean this F-contrast that you describe, that looks at the effect of all regressors (task design +  behavioural PM + all motion params) together?

If so, is there another way to include the motion params in the design that makes more sense? 

If I include the motion params as regressors, but dont include them in my contrast, their contribution won't show up in my result I think..

Thanks,

Joelle

On Tue, Jun 9, 2015 at 4:46 PM, H. Nebl <[log in to unmask]> wrote:

Dear Joelle,

This is an effects of interest contrast, you would go with an F contrast with as many rows as you have predictors (ignoring the constant, so eight), and for each of the rows you would weight one of the columns with "1" and the rest with "0", resulting in
[1 0 0 0 0 0 0 0 0;
0 1 0 0 0 0 0 0 0;
0 0 1 0 0 0 0 0 0;
0 0 0 1 0 0 0 0 0;
0 0 0 0 1 0 0 0 0;
0 0 0 0 0 1 0 0 0;
0 0 0 0 0 0 1 0 0;
0 0 0 0 0 0 0 1 0]
I think there was also a simpler way to create effects of interest contrast, but I don't remember. Usually the contrast should be useless though.

Best

Helmut