Dear Andrew; thanks for keeping NBS well supervised!
I' am trying to disclose groups differences between static functional networks connectivity FNC (as computed from Gift toolbox, which includes output matrices accounting not for pairwise but for bold correlations between different independent components-subnetworks that spawn multiples nodes).
My question is about how appropriate is NBS clustering method to compute differences between different networks that represents average bold correlations across several nodes... unfortunately could not find related information in the forum.
I have 3 groups with 11x11 matrices (accounting for 11 different independent networks ie. DMN, LANG, SALIENCE etc..)

If this is possible... should use NBS, FDR?  what should be the appropriate matrix design for an ANOVA-ANCOVA.. will this be ok?

grp  sex age

100 1 35
100 2 30
010 1 33
010 1 30
001 1 33
001 2 45
 
and a T test contrast looking for connections with greater connectivity in the firs group?

1 0 0?

Thanks a lot for your time and help.

Juan P

Hi Juan, 

NBS can be performed on a set of 11 x 11 matrices, although for small networks of this order, its may be feasible and better to use Bonferroni correction, particularly if your sample size is large. 

When you say greater connectivity in the first group, do you mean greater connectivity relative to the other two groups? If so, the first three columns of your design matrix can be replaced with a single column in which group 1 subjects are coded with 1 and group 2 and 3 subjects are coded with -1. 

A t-test can then be used with the contrast of [1 0 0 0.....]

The current design matrix is suited to an F-test that asks whether there is a differences in the means of the three groups. 

Andrew
Originally posted by Juan Pablo Princich:

Thanks!
Will this t test work like an anova for 3 groups that way?
In this case will need to reorder groups to test group 2  >  G1+G3 .. and the same for G3..?

cheers
JP

Yes - need to reorder for each of the three t-tests. 

Originally posted by Juan Pablo Princich: