Hi,
You can find other postings on the list about this, but I'll give a very brief explanation.
The tests in GLM are separate and so the t-tests are not post-hoc.
If you restrict the t-test findings to the significant areas of the F-test results then they are equivalent to post-hoc test.
The reason that you can get some results for the t-tests but none for the F-test is that the F-test will have a slightly different threshold (due to incorporating more than one contrast and also because it is a two-tailed test, whereas the t-tests are one-tailed). Results like this indicate that you have something close to threshold, but not quite reaching significance for your F-test/ANOVA. Also, strictly speaking, you should not treat your t-test results as significant without having adjusted them for multiple comparisons across the number of independent contrasts (which will almost certainly make them non-significant, as this will be a harsher correction than the adjusted threshold in the F-test).
If you want a longer explanation of these issues then look at some of the previous postings on this list.
All the best,
Mark
On 1 May 2014, at 04:57, Gong-Jun JI <[log in to unmask]> wrote:
> Dear FSL experts
>
> I performed a one-way four level ANOVA by using GLM_gui and Randmise command.
>
> I design the matrix and contrast as the GLM manual:
> F
> c1 1 0 0 -1 1
> c2 0 1 0 -1 1
> c3 0 0 1 -1 1
>
> I selected both F and c1, c2, c3 (I understand this as t tests). And got one *fstat map and three *tstat maps.
>
> There are not significant results in f-map. However, the tstat map did show significant results. Thus, I am confused to the relation between fstat and tstat. I thought the tstat is a post-hoc of fstat. If the F-test is non-significant, how the t-test could be significant? or what the meaning of this t-test?
>
> Any comment would be appreciated.
>
> Gong-Jun JI
|