Dear Alfonso,

I hope this message finds you well. I am currently working on a study using the CONN toolbox with 246 user-defined ROIs to conduct ROI-to-ROI analysis. For the 2nd level analysis, I am using the 'customize (advanced Family-Wise Error control setting)', along with the MVPA omnibus test for cluster-level inference.

In this study, I am comparing two groups: Group A (33 subjects) and Group B (24 subjects), for a total of 57 participants.

I came across your previous response to Christopher Hyatt's question(https://www.nitrc.org/forum/message.php?msg_id=37343), where you mentioned that the MVPA omnibus test employs the same SVD-based procedure as fc-MVPA and is often suitable for cases where the number of ROIs is relatively large (e.g., hundreds or more). Based on this explanation, it seems that using the MVPA omnibus test with 246 ROIs should be appropriate given the relatively large number of ROIs in my study.

However, I would like to confirm whether my current approach—setting the connection-level threshold to p<0.01 and the cluster-level threshold to p<0.05 (cluster-level p-uncorrected, MVPA omnibus test)—is valid and suitable for reporting results in this context.

Could you kindly provide clarification or guidance regarding whether this analysis setup is statistically sound for my dataset? Additionally, do you recommend any alternative approaches for datasets with a similar number of ROIs and a two-group comparison if improvements could be made?

Thank you in advance for your time and support.

Best regards,
Junhyeok Jang