Voxelwise statistics of FA and MD data were carried out in AFNI using independent sample t-tests. Type I error control for multiple comparisons was achieved with a combination of individual voxel probability and cluster size thresholding, requiring that any significant group difference was comprised of 54 contiguous voxels, each differing at p <.05. Monte Carlo simulations determined that this yielded a brain-wise probability of a false positive at p <.01. Thirteen such FA clusters and 2 MD clusters represented significant differences between users and controls (Bava et al., 2009). Clusters showing significant group differences in FA or MD were correlated with neuropsychological test scores using Pearson's r correlation coefficients (α=.05). Significant correlations were subjected to hierarchical regressions to assess the unique influence of FA on neuropsychological performance for each group, with mean FA and group status entered on step 1 and their interaction term on step 2. If the interaction was nonsignificant (p>.05), results from the first step were interpreted.
