Hello,

I am seeking clarification regarding an apparent discrepancy between the direction of condition-specific effects observed in CONN and those obtained from extracted connectivity values analyzed externally. Specifically, for visualization purposes, extracted connectivity values were plotted to aid interpretation of significant interaction effects observed at the CONN second level. To this end, I am extracting the Fisher z-transformed correlation coefficients for each connection from the first-level results folder and using these values to visualize condition-specific connectivity patterns and perform follow-up analyses. Despite this, the direction of the condition effects does not align with what is implied by the CONN second-level contrasts.

More specifically, I am conducting a within-subject resting-state fMRI study in which each participant undergoes four scans under different pharmacological conditions. At the CONN second-level, I performed a whole-brain network analysis with a full factorial interaction contrast across the four conditions (contrast: [-1 1 1 -1]), while controlling for age, sex, and BMI. The interaction effect observed in CONN is reproducible when I analyze the extracted values in SPSS, so the presence of the interaction itself is not the issue.

However, the problem arises when examining the direction of pairwise comparisons between conditions. Based on the CONN second-level results and contrasts, I expect that connectivity between two regions should be stronger in Condition 1 compared to Condition 2. To aid interpretation of this interaction, condition-specific connectivity estimates were extracted and examined. In SPSS, using the extracted connectivity values, the pairwise comparison instead suggests the opposite pattern (Condition 2 > Condition 1). I have carefully verified that the correct values were extracted for each subject and each condition.

My questions are:

1. 
   

   Why might the direction of pairwise condition effects differ between CONN second-level results and analyses of extracted first-level z-transformed connectivity values, even when the interaction effect itself is consistent?

   
   
   
2. 
   

   Are the values in the first-level results folder the appropriate estimates to use when examining condition-specific connectivity differences, or does CONN apply additional centering, scaling, or model-based transformations at the second level that could affect the interpretation of simple effects?

   
   
   
3. 
   

   Is there a recommended way to extract condition-specific connectivity estimates directly from the second-level model in CONN, such that covariate adjustment and contrast coding are preserved when visualizing or exporting the data?

   
   
   

Any guidance on how to correctly interpret or extract connectivity values so that the direction of condition effects is consistent with CONN’s second-level results would be greatly appreciated.

Thank you for your time and help.