Print

Print
I very appreciate your help,  Helmut.

But I still have quite a lot of confusion...
1. Using "One-way ANOVA - within subject" definition I add "subject" field (scans, conditions): do I treat this "subject" field  a subject (by adding 13 times, number of my subjects) or as contrast (by adding 5 times, number of my contrasts)? Then I use F-contrast? I tried, but it seems to be invalid.
2. I did not quite understand your easiest solution with single-subject contrast. I already have a 1-st level t-contrast, which are reflected by my con* files. So, what are the contrasts [1], [0 1] etc that you propose?

Thanks again.



On Thu, Sep 12, 2013 at 12:42 PM, H. Nebl <[log in to unmask]> wrote:
Dear John,


in your case (repeated measurements) you have to go with "One-way ANOVA - within subject" instead of "One-way ANOVA". As you can see your current design matrix has no subject factor, so you treat the data as if you had scanned five different groups of subjects.

With "One-way ANOVA - within subject" you should be able to set up a valid F contrast "condition" and valid comparisons between different conditions like [-1 1]. Alternatively, 1) use GLM_flex or 2) build all the relvant contrasts on single-subject level and conduct a bunch of one-sample t-tests on second level.

However, a conjunction analysis within "One-way ANOVA - within subject" is still going to be invalid due to issues with the error term (for details see some of the messages by Donald McLaren). The easiest solution, set up five contrasts [1], [0 1], ... on single-subject level. Conduct five one-sample t-tests on second level. Those voxels showing up significantly in all five one-sample t-tests can be understood as the conjunction.

Oh, and hopefully I didn't make any mistakes, otherwise Donald has to correct me.


Bestm,

Helmut