Hi,

Thanks for sending the design matrix.  I assume you're referring to cope 2?  Since you took out the subject-specific regressor for the last subject, cope 2 is now modeling something pertaining to the mean of that last subject.  I'm a bit foggy right now and can't figure the exact interpretation, but I'm sure it isn't meaningful or related to what you ran in the first model, so I wouldn't use this model for any of the contrasts.

To be clear, use this for the interaction

http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#ANOVA:_2-groups.2C_2-levels_per_subject_.282-way_Mixed_Effect_ANOVA.29

To test the group effect, do what is described in that link as well.  Sounds like you already ran it.

Hope that helps,

Jeanette

On Fri, Aug 15, 2014 at 11:22 AM, Leslie Cao <[log in to unmask]> wrote:

Thank you for your replies. My second matrix looks like this.

On Fri, Aug 15, 2014 at 12:07 PM, Jeanette Mumford <[log in to unmask]> wrote:

Hi Wen,

It is a bit difficult to understand from your description, but I think your second model is testing a different hypothesis than you think.  Can you send snapshots of the design matrices and contrasts that you're comparing from the 2 models?  

Thanks,
Jeanette

On Thu, Aug 14, 2014 at 4:18 PM, Leslie Cao <[log in to unmask]> wrote:

Dear FSL experts,

I have a question about ANOVA group effect. We have two groups and each subject has two sessions. First I followed the instructions on FSL wiki ANOVA: 2-groups, 2-levels per subject (2-way Mixed Effect ANOVA), so I got session effect and interaction from first step and  averaged the 2 measures within-subject and ran a 2 group mean comparison on these data in a 3 level analysis approach. Another method I just ran one ANOVA with group main effect modeled, to avoid rank-deficiency with the EVs, I dropped one subject-specific mean in each group. I was expecting similar results from both methods.The interaction and session effects are pretty similar, however, the group difference I got using first approach and group main effect in the second ANOVA are totally different. The second group effect is much stronger but appear to be suspicious. So is there a explanation that the second approach might be giving wrong group main effects?

Thanks,

Wen