Dear all,
We are setting up a repeated measures analysis using the flexible factorial
in SPM5. We have 2 groups (n=18 and n=26) and one condition with 4 levels.
Initially we specified the main effect of subject and the interaction effect
of group and condition for our design matrix. However, this creates a huge
design matrix with many regressors. We are now having a discussion on
whether this amount of regressors might not lead to overfitting since we
have 52 regressors for 176 datapoints. Is there a way to determine whether
we are overfitting? Or a good ratio between regressors and datapoints below
which you should stay?
One of the reasons this discussion started is because the group effect is
now insanely big, with the whole brain covered and clusters with infinite
z-values. When we define the deisgn matrix based on group, condition and
group*condition as main effects and interactions of interests, the main
effect of group looks reasonable again.
I hope someone could help us with this discussion.
All the best
Harma
|