Hi Andrew,

Many thanks for your answers!

1.) Please allow me to clarify:
We compare fMRI-based connectivity from two experimental conditions. We have a within-participant design, hence we'd use a paired t-test. Does the same argument of yours apply to the paired t-test?
In "classic" task activation designs, however, we would first contrast the conditions at the individual level and then combine the contrast images at the group level using a one-sample t-test. Hence our questions on the one sample t-test.
Would you advise to use the paired sample t-test for NBS?

2.) When comparing the two conditions, we noticed that we find differences in both directions (i.e. we have connections where condition A>B, and we have connections with AB or B>A). And the permutations randomly flip the condition labels (A vs. B). And inference is done on the size of the components alone. And not on their topological position.
When looking at the relatively sparse A>B directions, the sheer number of effects in the B>A direction seem to cause a large component in the permuted networks. In other words: We are under the impression that we find a meaningful empirical component at expected locations (which is rather small), but the permutations show even larger components (at different locations). Which of course interferes with the NBS inference.

Does this make any sense to you? And do you think that our concern is justified?

We are aware that this should not present a barrier for statistical inferences per se, but we are suspecting that this might influence our results negatively.
For instance, in the case of the one-sample t-test you stated elsewhere in the forum, that if someone aims to conduct a one-sample t-test, this someone should be cautious about having more negative than positive connectivity values —> https://www.nitrc.org/forum/forum.php?th...

3.) Regarding the thresholding:
Applying light proportional thresholding (i.e. removing the most negatively signed connectivity differences), we do get NBS results that we deem meaningful. Applying an absolute threshold and zeroing all negatively signed connectivity differences, we also find NBS results that we deem meaningful. But such thresholding would only be justified if the misbalance between positive and negative connectivity differences was actually a concern. Hence our questions.
Best,
Ole