Dear Andrew

Thank you very much for your prompt reply.

I have a follow-up question.

We are analyzing within-subject data across three conditions. I am wondering how NBS handles NaN  data, especially when it is converted to 0. If a particular connectivity measure is 0 for all subjects, it won't be calculated. However, how does NBS deal with cases where connectivity is 0 for only some subjects? Does NBS accommodate missing values (like Skillings-Mack test and Wilcoxon signrank on post-hoc)?

In our data, certain subjects have data for only one or two conditions(out of 3 condition) because the conditions were sorted based on in-task thought probes. Hypothetically, we would prefer not to exclude these subjects. However, if this situation negatively impacts the NBS analysis, exclusion might be our only option. What are your thoughts on this?

Originally posted by Andrew Zalesky:

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