Consider changing the default to `by = NULL` or `by = "margin"` in `adonis2()`
Closed this issue · 2 comments
The default permutation test in adonis2() is by = "terms" for a sequential test of model terms where order of terms can matter. I appreciate that this was the default (and only option in adonis()?) but it really isn't very useful in many situations that ecologists will be using adonis2() for (observational data).
by = "margin" would be better if you want a test of model terms by default. by = NULL would be more consistent with the anova() methods for rda(), cca(), etc.
I think we should change this to do the statistically better thing at the expense of backwards "compatability" with adonis().
Personally I think that by = "terms" is statistically more correct than by = "margin": single terms sum up to total in sequential tests, but not in marginal tests. I only added that alternative to get rid of whining. That said, I agree with changing the default, but I'd prefer by = NULL.
This is now implemented in the master branch and in the cran-2.6-6 branches.