Dear Professor,
I am using CONN to perform a one-way ANOVA between three groups of subjects. When I performed an F-test (FWE control settings: voxel p<0.001 uncorrected; cluster p<0.05, FDR corrected), I obtained a significantly different region in the brain. However, when I tried to perform post-hoc comparisons using the same method,(The contrast matrix I set is [1 -1 0])I obtained three regions of difference.(the pictures are as below) The similar phenomenon sometimes results that the different region of post-hoc comparisons are larger than that of f test. This seems to indicate that I am getting some false-positive results. Should I adjust the p-value to avoid these false positives?if so ,how should I adjust?
looking foward to your reply
Dear Junxiao Ma
That is perfectly fine, typically you would simply restrict your post hoc analyses to the regions where you found a significant effect in the main F-test to avoid these issues. You can do that, for example, in the results explorer window by selecting the 'plot effects' button (the one with a bar-plot icon), which will show aggregate statistics within each cluster, or alternatively you may run again voxel-level analyses for your post hoc tests as you are doing here but then use masking (e.g. by selecting the 'SPM display' button and selecting the masking option there) to restrict your analyses to only suprathreshold areas from your original F-test.
Hope this helps
Alfonso
Originally posted by Junxiao Ma:
Dear Professor,
I am using CONN to perform a one-way ANOVA between three groups of subjects. When I performed an F-test (FWE control settings: voxel p<0.001 uncorrected; cluster p<0.05, FDR corrected), I obtained a significantly different region in the brain. However, when I tried to perform post-hoc comparisons using the same method,(The contrast matrix I set is [1 -1 0])I obtained three regions of difference.(the pictures are as below) The similar phenomenon sometimes results that the different region of post-hoc comparisons are larger than that of f test. This seems to indicate that I am getting some false-positive results. Should I adjust the p-value to avoid these false positives?if so ,how should I adjust?
looking foward to your reply