Hi Donald,

I'm just going back through these emails, and wanted to double check one part of this question. Apologies in advance for my confusion. The question was:

And then create a contrast at this second-level [0 1]? The '1' tests the Covariate. You can see my design matrix and contrasts in the screenshot attached.

I'm wondering how much it makes sense to put the contrast over the covariate (ie second column) rather than the average (ie the first column)? I am interested in how behavioural performance (the covariate) affects the overall pattern of activation. Just FYI, from the first-level, it was the overall pattern of activation (ie onset and duration) .con images that were forwarded to this second level analysis.

Thanks,

Joelle

On Thu, Aug 13, 2015 at 3:42 PM, Joelle Zimmermann <[log in to unmask]> wrote:

Thanks Donald. Can I input the Covariate as a vector with one number per subject, so the vector = # subjects? I presume the order then is important, so the order of subjects in the Covariate vector must be the same as the order in which I input subject's .con scans into the second-level model...

And then create a contrast at this second-level [0 1]? The '1' tests the Covariate. You can see my design matrix and contrasts in the screenshot attached.

On Thu, Aug 13, 2015 at 3:27 PM, MCLAREN, Donald <[log in to unmask]> wrote:

Create a contrast of Trial10-Trial1 for each subject. Use this in a one-sample t-test and add a covariate. The covariate should be the performance difference of Trial10-Trial1.

Best Regards, 

Donald McLaren, PhD

On Thu, Aug 13, 2015 at 3:24 PM, Joelle Zimmermann <[log in to unmask]> wrote:

Thanks, I tried this now. I'm a bit disappointed I'm not seeing much activation related to behaviour when I look at it this way. Could you perhaps expand a bit on your other suggestion on our other email chain, where you suggest looking at Trial 1 and Trial 10 (which indeed makes sense in my case), computing the difference in performance of Trial 1 and Trial 10 for each subject. And then regress the differences of behavior onto the differences in Trial response.

It's something I'd want to do but not quite sure how to tackle it.

On Thu, Aug 13, 2015 at 3:12 PM, MCLAREN, Donald <[log in to unmask]> wrote:

Yes. That should be correct. I am assuming that the PM is the second column of the design matrix.

Best Regards, 

Donald McLaren, PhD

On Thu, Aug 13, 2015 at 2:04 PM, Joelle Zimmermann <[log in to unmask]> wrote:

Thanks Donald :-)

Just to clarify - set up a contrast on the first level for the PM (ie in this case 0 1 0 0 0 0.. ), and then forward the subject's resulting .con images into a second-level.

On Thu, Aug 13, 2015 at 1:46 PM, MCLAREN, Donald <[log in to unmask]> wrote:

Use a contrast for the PM modulator and then test if the PM is different than 0 at the group level using a 1-sample t-test.

Best Regards, 

Donald McLaren, PhD

On Thu, Aug 13, 2015 at 1:42 PM, Joelle Zimmermann <[log in to unmask]> wrote:

Hi Experts,

I'm setting up a second-level analysis, where I plan to examine how activation across trials is related to behaviour across trials. I was hoping someone would have some suggestions of how to practically implement this in SPM.

I have a task, with 10trials (with rest periods in between), and one behavioural measure per trial.

For each subject, I did a first-level analysis with one condition with 10 onsets, and a parametric modulator where I put in behavioural performance. So first column of the design matrix is task versus rest, the second column is behavioural performance, and the next 6 are motion parameters.

Now I want to set up a second-level model. Does anyone have ideas of how I can look at how behaviour and activation are related at this level? And also how I can implement this design in the GUI?

Thanks,

Joelle