I currently have a data set where the response is compositional with many components.
I am considering lumping some of the components together. I believe this will make it easier to spot the potential effect that some independent variables may have on the components. The specific effect on each component may be insignificant, but if we lump the effects together, it may become more clear.
Is this reasoning correct, or does the CODA theory work just fine with $k >> 2$ components than with less?
