> To my surprise, the 3dANOVA2 output shows much more
> activation than the single factor, fixed effects ANOVA in all
> conditions and comparisons. Is this normal or should the
> activation be less? Does it say something about our data
> that this result comes up consistently?
With your design type there should be no surprise in that 3dANOVA2 gives you more activation than 3dANOVA. Basically one-way ANOVA is more appropriate for a design with which you can compare a condition/task among several groups of subjects, similar to a two-sample
t test for two groups of subjects.
If you insist on using one-way ANOVA, the following theoretical proof can convince you why it would generate less significant results than 3dANOVA2 (e.g., one-way ANOVA within-subject).
For a two-way ANOVA (AxB, A fixed and B random), the
F test for the main effect of factor A is: F = MSA/MSAB
If you treat the design as a one-way ANOVA with repeated samples, the
F test for the main effect of factor A is: F* = MSA*/MSE*
Careful algebra operations would show that
MSA = MSA*,
MSB + MSAB = MSE* (as SSE = 0).
Thus more statistical power of the first model can be seen from the following fact
F < F* unless MSAB = 0 (that is, no interaction between factor A and subjects, which is rarely true in FMRI experiments).
The same logic applies to any contrasts since the denominators are the same.
Gang