What are the differences between
-model_outliers (Model outlier betas with a Laplace distribution of subject-specific error) in afni 3dMEMA
automatic outlier de-weighting (FSL FLAME-Mixed Effects) that detect outlier datapoints (for each voxel)) [Woolrich M (2008), NeuroImage], and automatically de-weighted them in the multi-subject statistics.
The difference between 3dMEMA and the FLAME module in FSL lies in their approaches to outlier handling. 3dMEMA employs a Laplacian distribution to accommodate potential outliers, while FLAME utilizes a mixture of two Gaussian distributions, one for the primary population and the other for outliers. In case you try out both approaches, it would be interesting to see how the two compare to each other in handling outliers.