Hi Catarina - 

On 5 September 2011 13:11, catarina saiote <[log in to unmask]> wrote:

I would like to set up a feat analysis and after reading the examples and going through previous posts I'm still not sure how to set this up.
I have 3 groups (10 subjects in each group) and 3 separate scans (for each subject) and I've run individual first-level analyses on all 90 scans. I wanted to look at group effect, scan effect and, if possible, the interaction between group and scan. 

It seems to me that I need:
2 columns to model group differences,
2 columns to model scan differences,
3x9 columns to model differences between subjects in each group

1 f-test across columns 1 and 2 to get group effect
1 f-test across columns 3 and 4 to get scan effect.

Does this sound correct?

This seems OK.  

Second question: how can I model the interaction?

You would have to multiply all the different group and scan differences columns to create a new set of evs I think - complex!

Third, I also want to set up contrasts to look at the specific differences between groups and scans (group1-group2, scan1-scan2, etc.). The example for the "tripled" t-test in the website sounds like I want to do, at least with the scan differences where there's repeated measures, except that I am not modelling subject means, so I'm not sure how that affects the rest. And I'm also not sure the line of thinking is the same in the group differences... Can I have some help with this?

You might have to be clearer on you exact model - don't you have an ev group1-group2 for which  you can generate your desired contrast?


Or, is there a better way to find this effects/differences?

What exactly are your differences across scans and you key hypotheses? It might be difficult to interpret such a complex model.  It could be easier to generate scan differences in fixed effect second level analyses and look for mean and group effects/interactions in separate third level models.


Thanks in advance for your help,