Dear freesurfer experts,
I have a mixed 2x2 design with group (approx. 18 subjects per group) and timepoint (t0, t1) as factors. I analyzed the thickness data with the longitudinal stream and the spatiotemporal LME as a random intercept model, age as covariate. However, the model does not converge perfectly, although I varied the size of the homogenous covariance regions between 1-2 (parameter k, Bernal-Rusiel et al., 2013). In addition, when I apply a voxelwise-LME, more clusters survive the FDR correction. It seems that the V-LME is more sensitive than the SP-LME in my case (clusters at an uncorrected threshold are more or less the same). I would love to understand why. Could anyone explain this to me? And would it be statistically valid to apply the V-LME to my data?
Best,
Ann-Kathrin