Thanks Donald. I avoided including an EV for each of the three groups
as the wiki suggested this would cause a rank-deficiency with the EVs
modeling the subject-specific means (which is the error msg I get when I
add them). Do you have any additional thoughts on this, i.e. is this a
warning that can be ignored?
I think I needed to provide a little more information in my original
post. For the contrasts, I was under the impression that EV2 and EV3 in
my design already specified the interaction between groups 1&2 and
condition (EV2) and between groups 1&3 and condition (EV3):
EV1(c) EV2(gXc_g1&g2) EV3 (gXc_g1&2)
g1c1 1 1 1
g1c2 -1 -1 -1
g1c1 1 1 1
g1c2 -1 -1 -1
g2c1 1 -1 0
g2c2 -1 1 0
g2c1 1 -1 0
g2c2 -1 1 0
g3c1 1 0 -1
g3c2 -1 0 1
g3c1 1 0 -1
g3c2 -1 0 1
c1>c2 1 0 0
intg1&2 0 1 0
intg1&3 0 0 1
If you have any more thoughts or suggestions they are appreciated.
On 21/11/2013 3:36 PM, MCLAREN, Donald wrote:
>
>
> On Thu, Nov 21, 2013 at 8:31 AM, Greg Reckless
> <[log in to unmask] <mailto:[log in to unmask]>> wrote:
>
> Dear FSL list,
>
> I have a 2(condition) x 3(group) mixed-design study. I have created
> a model and contrasts (attached .png file) based on the 2-way Mixed
> Effect ANOVA described on the FSL Wiki
> (http://fsl.fmrib.ox.ac.uk/__fsl/fslwiki/GLM#ANOVA:_2-__groups.2C_2-levels_per___subject_.282-way_Mixed_Effect___ANOVA.29
> <http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#ANOVA:_2-groups.2C_2-levels_per_subject_.282-way_Mixed_Effect_ANOVA.29>).
> For simplicity I have only included 2 subjects per group. Two
> COPEs from the first-level, representing the two levels of
> condition, were entered for each subject in the higher-level model.
>
> My questions are as follows:
> 1) Is this model correct and is it the most appropriate for a 2x3
> mixed design?
>
>
> Yes. Although I would add 2 columns for group for completeness.
>
>
> 2) Are my contrasts correct for testing the main effect of condition
> (Lvl.1>Lvl.2 and Lvl.2>Lvl.1 and the interaction effect between
> condition and group?
>
>
> Interaction contrasts aren't interactions. The interaction would be 0 0
> 1 -1 0 0 0 ...
> The contrast 0 0 1 0 0 0 ... tests if Lvl.1>Lvl.2 is significant in
> group 1.
>
>
> Thank you for your input!
> Greg
>
>
|