The design matrix and inputs look ok.

There is no right or wrong primary threshold. An F threshold between 5-10 would be reasoable. An F of 20 is probably too high and if you are finding that all edges are signifciant at F=20, then something is probably wrong. 

The NBS manual provides details about post hoc testing. 

Originally posted by Xiang YAN:

Dear all

I’m new to NBS and SPM, and I'm currently working on NBS correction for a connectivity matrix. The experimental design involves three within-subject conditions.

I calculated a 64 × 64 connectivity matrix for each condition.
The final data matrix is 64 × 64 × 93, corresponding to 3 conditions × 31 subjects.

Question 1 : Is my input below correct for ANOVA (F-test)?  

Question 2(Main) : How should I determine the primary threshold in NBS, and which value is appropriate to report in my paper? In my case, even after setting the threshold to 20, NBS still reports one large significant network consisting of 1,136 edges out of a total of 2,016. Since we used the Schaefer parcellation for ROI definition, how should I choose the appropriate test statistic threshold (F or t)? 

Question 3: After performing the F-test, should I extract the significant network (edges and nodes) by masking the data matrix (non-significant data point to 0), and then conduct post-hoc t-tests? 

design matrix is : (which match the data matrix)
1    0    0  
0    1    0
0    0    1
1    0    0
0    1    0
0    0    1
1    0    0
0    1    0
0    0    1
1    0    0
0    1    0
0    0    1
1    0    0
0    1    0
0    0    1
1    0    0
0    1    0
0    0    1
1    0    0
0    1    0
0    0    1
1    0    0
0    1    0
0    0    1
...

contrast:
[1,1,1]


Exchange block :
1
1
1
2
2
2
3
3
3
4
4
4
5
5
5
6
6
6
7
7
7
8
8
8
9
9
9