Pascale,
3dMVM does not account for heteroskedasticity because the impact is usually not substantial. Especially when the number of subjects in the two groups is roughly equal, the 3dMVM (as well as the typical two-sample t-test) approach is robust to the presence of unequal variances. However, if this is really a concern, for a test comparing *two* groups, you can use 3dttest++ -unpooled to perform Welch's t-test: [
en.wikipedia.org]
Gang
Edited 2 time(s). Last edit at 02/28/2017 04:48PM by Gang.