Dear Orhan,
1) Full factorial is not appropriate for within-subject designs. Turn to Flexible factorial, GLM Flex, or build all the relevant contrasts on single-subjet level and forward them into one-sample t-tests each. This might explain different findings for your ANOVA and your ROI analyses.
2) As you relied on con images from differential contrasts (d11-g1), (d12-g1), (d21-g2), (d22-g2), did you also subtract estimates/percent signal change for g1, g2 from d11, d12, d21, d22 in your ROI analysis? In case you used Marsbar to extract values from single-subject models you would typically get one value for each of the six conditions, thus you would still have to subtract g1, g2 to get results reflecting those of the ANOVA.
3) Just in case, your ROI analysis would be double-dipping (detecting some effects for ANOVA, then looking at the sig. clusters again on ROI level). For illustrative purpose the approach should be okay though.
Best,
Helmut
|