Paul
Did you set up the F contrast by excluding certain columns rather than
typing in the contrast weights. If so then SPM seems to set up the
weights incorrectly for flex factorial designs. (When I say
incorrectly, SPM is probably doing what it is supposed to but it may
not be what you want.). First you will considerably simplify the
making of contrasts if you just include the interaction term in your
design. The main effects you can pull out of the interaction. It looks
like you have an unbalanced design, by the way, but that isn't causing
this problem.
Why don't you try estimating it again just including the subject
effects and the interaction. So you will have 58 subject columns and
then 6 columns for your interaction (3 for group 1 and 3 for group 2)
Then the contrast for the main effect of group would be something like
this (i'm not sure if I have the subject numbers correct.
ones(1,14)/14 -ones(1,44)/44 ones(1,3)/3 -ones(1,3)/3
Now when you say the F-values are strongly negative is that because
you directly plotted this contrast or you tried plotting it using
effects of interest or some other F contrast?
-----
Darren Gitelman
On Fri, Nov 14, 2008 at 6:18 AM, Dr P. Wright <[log in to unmask]> wrote:
> Dear SPMers,
>
> I am running a group X task ANOVA using the guidelines by Glascher and
> Gitelman for flexible factorial designs. The design matrix includes main
> effects of subject, group and task, and the interaction of group and task,
> and the contrasts match those in the guidelines (design matrix and
> guidelines attached).
>
> For the F contrast for main effect of group the SPM_F image shows negative
> values only. This does not affect other contrasts (e.g. main effect of
> task). There is a cluster of strongly negative F values in a sensible
> location, and this same cluster has positive F scores when I rerun the ANOVA
> using full factorial (SPM_F images for flexible and full factorial
> attached).
>
> I want to stick with the flexible design, since it models subject effect
> explicitly. Can anyone advise how to fix this?
>
> Best wishes,
>
> Paul Wright
>
> --
> Paul Wright PhD
> Centre for Speech, Language, and the Brain
>
> Dept. of Experimental Psychology Tel: +44 (0)1223 766559 University of
> Cambridge Fax: +44 (0)1223 766452 Downing Street
> email:[log in to unmask] Cambridge http://csl.psychol.cam.ac.uk CB3
> 2EB
>
|