Hello everyone,

I'm using Network-Based Statistics (NBS) to study changes in resting-state functional connectivity, and I have a conceptual question about the interpretation of the dependent variable. 

NBS uses the signed functional connectivity value (e.g., Fisher z-transformed correlations) as the dependent variable in the edge-wise GLM. Consequently, a positive regression coefficient can arise either because:

• a positive correlation becomes stronger (e.g., 0.2 → 0.6), or


• a negative correlation becomes weaker (e.g., -0.6 → -0.2).

Both produce the same sign of the regression coefficient, so they can be grouped into the same significant NBS component. My concern is not whether this is statistically valid. Rather, I'm wondering whether the dependent variable itself is appropriate for the biological question. If positive and negative functional connectivity reflect different functional mechanisms, then an NBS component may group together edges representing qualitatively different processes (strengthening positive coupling vs. weakening anticorrelations). In that case, the component would be statistically coherent but not biologically.

So my questions are:

1. Have people addressed this issue by analyzing positive and negative edges separately, or by using another representation of functional connectivity?
   
2. If not, is the standard practice simply to interpret the sign of the individual edges post hoc after identifying the significant component?
   
3. Is this considered a genuine limitation when interpreting NBS components, or is it generally accepted that signed functional connectivity can be treated as a single continuous variable for this purpose?
   

Thanks you in advance!

Best regards, 

Iván.