Dear CONN experts,

I am currently working on a second-level ROI-to-ROI analysis and have some questions regarding the statistical outputs, specifically about the relationship between connection thresholds and the resulting statistics. I would be very grateful for some clarification.

Here are my observations:

1. 
   

   Constant F-statistic with varying connection thresholds:

   
   

   In the second-level results window, I have selected cluster-level p-uncorrected (MVPA omnibus test) for correction. I've noticed that the cluster-level statistic, for example, F(3, 88) = 5.46, remains identical whether I set the connection threshold to p < 0.05 or p < 0.005.

   
   
   
2. 
   

   Constant F-statistic with different contrast types:

   
   

   Following on from the first point, changing the contrast for the connection threshold from two-sided to positive contrast also has no effect on the reported cluster-level F-statistic, which stays at F(3, 88) = 5.46.

   
   
   
3. 
   

   Change in statistic type when plotting effects:

   
   

   However, when I click on "plot effects" to visualize the results for this cluster, the displayed statistic changes. It is no longer an F-statistic but a T-statistic, for example, T(90). Furthermore, the visualization and the T-statistic do change when I adjust the connection threshold (p < 0.05 vs. p < 0.005) or the contrast type (two-sided vs. positive).

   
   
   

This leads to my questions:

• How can I calculate the cluster-level F-statistic based on different initial connection thresholds? If I want to test how the cluster-level F-statistic itself changes with different formation thresholds (e.g., p < 0.05 vs. p < 0.005), how can this be achieved? It seems I need to adjust the initial "cluster-forming threshold" rather than the "connection threshold" used for visualization. Where can I find this setting for the ROI-to-ROI analysis? My understanding is that in SPM, the initial voxel-level threshold is fundamental to the subsequent cluster-level statistics. It seems that the "connection threshold" in the CONN results window acts differently.


• 
  

  Why does the cluster-level F-statistic from the MVPA omnibus test remain unchanged by the "connection threshold," while the post-hoc "plot effects" T-statistic is sensitive to it? 

  
  

Thank you very much for your time and for developing such a powerful toolbox. Any insights you can provide would be extremely helpful.

Best regards,

Xiaochen