I ran GLM analysis on resting-state datasets that include one EEG regressor. I need to find the correlated maps at the group-level. So, I applied a one-sample t-test using 3dttest++ on the stat output from induvial GLM analysis. Then, I had to estimate cluster wise correction for the correlated maps. For this, I tried two approaches to estimating ACF.
- I used errts from individual subjects to estimate the ACF parameters and then averaged them for all subjects and then plugged them into 3dClustSim function.
- I generated a residual dataset from 3dttest++. Then, I estimated ACF from there and used 3dClustSim.
The issue is when using the second approach, I tend to get really small estimated clusters size ~ 14 voxels at p=0.005, for example, as compared to ~130 voxels from the first approach.
What is the right approach for estimating the correlated maps at the group-level?
Is cluster correction the right approach to be applied for controlling for the false positive, while not being too conservative?
Really appreciate your support!