Hi,

Sounds like group is a between-subject effect, so you actually cannot test it in this model as it doesn't really make sense to. Definitely keep you subject-specific means in the model because you must control for repeated measures. Use that model to test for an interaction between group and time.

Also, if your interaction effect is significant, testing the main "group" effect doesn't really make sense.

If you want to test the group effect, it requires a different model. Basically, you'd need to average your 2 measure per subject first and then run a 2-sample t-test in a different model. Again, this doesn't make sense to test if you have an interaction effect as a significant interaction means that the group effect changes according to your other factor, so that's your information on group difference. So, I'd only do this if the interaction was not significant. For example, say the means are G1T1=5, G1T2=3, G2T1=4, G2T2=6. Granted this is an extreme case, but you can see there's an interesting interaction (G1 decrease over time and G2 has a increase over time), but the main group effect would be 0.

If you are an R user email me individually and I can send you some code that matches up a repeated measure ANOVA result with models like what we use for our fMRI data (linear regression models).

Cheers,
Jeanette

On Sun, Oct 23, 2011 at 1:49 AM, Jon Brock <[log in to unmask]> wrote:

Dear Eugene & Donald,

Thanks for your responses.

Sorry, after requesting clarity I inadvertently introduced some confusion. Group 1 and Group 2 contain different subjects (Subject1 Group 1 is not the same as Subject1 Group 2).

It sounds like my design allows me to test the main effect of time and the group x time interaction but not the main effect of group - if I replace the EVs holding each subject's mean effect with a single column of ones would the main effect of group contrast then be valid?

Pos effect of time 1 - I thought this showed scan 1 > scan 2 ;and pos effect of time 2, scan 2 > scan1.

Many thanks
Jon