behinger
commented
3 years ago I replicated @jaromilfrossard findings using a Pluto notebook. It definitely depends on the number of trials, thus our sd/n-p might be a good route.
Further, the permutation() function should also expose all options of permute! (e.g. scaling factor)
Residuals are only exposed in permute! but not in permutation()
https://github.com/palday/MixedModelsPermutations.jl/blob/a03eff074edaa28cf6658971ded71fc0556e90af/src/permutation.jl#L103
General Problem is that residuals are underestimated in permutations right now
Jaromil also writes: "First trying to take the residuals that come from the mixed model estimation. Second reflating only the residuals. I think the first one are more "elegant" as we do not use reflating factor, but may not work."