Dear Alfonso,
Following your suggestions in previous threads and impressed by recent debates regarding parametric vs non-parametric inferences (Eklund et al), I re-run my MVPA (5 components, 21 subjects, testing for condition A vs condition B) using non-parametric approach.

I have a couple of questions regarding the non-parametric approach that could, possibly, be also interested for other CONN users:

1. Does the permutation test in CONN estimate the null distribution for one side? I am a bit confused as the option 'two-sided' appears as default (please, see the screenshot attached). Please, correct me if I am wrong. The permutation approach will flip my labels (2Λ21 in my case), calculate a range of t-tests and then using max stats estimate the 5% threshold (this threshold we will apply then to our original map of t-tests). If this is what CONN does, what the 'two-sided' option means?

2. On the top of the screen, CONN displays F-stats for non-parametric. What are these degree of freedom? I am not sure how to interpret the F-value (please, see the screenshot).

3. Whether CONN uses smoothing for the variance of t-tests? 

Thank you

alla