Hi Yang,
-glfCode type_short 'type : 1*ME -1*MI task : 1*short' \
-glfCode type_middle 'type : 1*ME -1*MI task : 1*middle' \
-glfCode type_long 'type : 1*ME -1*MI task : 1*long' \
To obtain the directionality of the comparisons, it would be more informative to implement these three lines using -gltCode instead of -glfCode.
What do the first three sub-bricks mean?
As revealed by their labels, the first three sub-bricks are the statistics for the two main effects and their interaction.
I've found that the sub-brick #0 can set a p-value, but the sub-brick #21 shows a p-value of [N/A].Why is this happening？
For some reason, the degrees of freedom for sub-bricks #21-26 were not assigned in your output. Update your AFNI, and see if the issue would be resolved.
I see that Gang mentioned a soft cluster threshold way, it doesn't make adjustments to the results?Alternatively, you may simply adopt a soft cluster threshold (e.g., 20 voxels at the p-value of 0.01), and use a highlight-but-not-hide approach as suggested in this recent paper.
My perspective may deviate substantially from the mainstream. In line with George Box's famous assertion, I contend that all models are approximations, but it is crucial to identify where a model deviates from accuracy. The traditional mass univariate modeling approach (as seen in, for instance, 3dLMEr) entirely overlooks the hierarchical structure of the brain, leading to an overly conservative family-wise error (FWE) rate through multiple testing adjustment. Put simply, blindly accepting the FWE adjustment at face value is unwarranted.
Gang Chen