I ran the linear mixed model matlab code on DTI-data (FA) from tbss (700 subjects and ca. 130 000 voxels). I adapted the code a little bit to accommodate the data format. I have a longitudinal design with 3 time points and two groups, the model is intercept (random factor) + group (patients, controls) + centered age + sex + a continuous clinical variable. Centered age = age - mean age for all observations (across all subjects, groups, and time points).

The scripts run without crashing but lme_fit_FS.m converges in only ca. 30% of the voxels. lm_F produces F and p-values that I can use. However, I'm uncertain what to make of the messages about "Algorithm does not converge after 50 iterations!!!" Does that invalidate the analyses? Or can I use the resulting pvals even where lme_fit_FS did not converge? 

Also, I'm  not used to such a high rate of non-convergence. Any idea what may be going wrong? The FA-data looks fine (between 0 and 1; mean = 0.49; SD = 0.15).

Thank you!