First of all, are you using a GLM-based statistic (e.g. t-test) or the non-parametric Brunner-Munzel test? There are differences in the fundamental assumptions of NPM and VLSM. Each is a valid way of testing for differences based on these assumptions. While I am happy that NPM has proved robust and popular, I no longer maintain it. You may want to try my NiiStat. Alternatively, I am a huge advocate for Anderson Winkler's PALM. I only created NPM/NiiStat because PALM did not exist at that time.