Dear Genevieve,
So you've got one within subject factor with two levels and one between
subject factor with two levels. If you use a flexible factorial design,
you should only include 'main effect of subject' and 'interaction group
x condition'. Then you can test for the main effect of condition or the
group x condition interaction. To test for the main effect of group, you
would create another model without the main effect of subjects (and
could use a full factorial design for that).
A recommended alternative would be to use partitioned error models
where, for example, to test for the group x condition interaction, you
would compute the differential effect of condition within subject with a
contrast [1 -1] then enter these images (one per subject for each group)
in a two-sample t-test and use a contrast [1 -1] again.
All your factors have two levels so contrasts for F-tests for main
effects and interaction will have a single row, meaning that you can use
t-tests of F-tests depending on whether you are interested in the
direction of the effect (or have specific hypothesis regarding this).
Best regards,
Guillaume.
On 28/03/17 12:21, Genevieve Allaire-Duquette wrote:
> Hi,
>
> I'm defining the contrats in a flexible factorial design.
>
> Factor 1: subject (n1=21 n2=18)
> Factor 2: groups (ng=2)
> Factor 3: conditions (nc =2)
>
> I've included in my model:
>
> Main effect subjects
> Main effect group
> Main effect condition
> Interaction group X condition
>
> When I define the contrasts for main effets and interaction, should I use t-constrats of f-contrats?
>
> Thanks in advance!
> Genevieve
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|