Dear Jesse.
The most logical explanation is that in the fitted ARD model the rate of
transition away from the rare state is >> than the fitted transition
rates to it (or even that the ML value of these rates is zero).
Conditioning on this fitted model make it highly probably that the root
is
My biggest issue with this setup, as you give it, is that it isn't
clear how you'll make this into a test - is there some summary
statistic of the null morphospace you'll be generating many times, and
then comparing that distribution to the observed value of the
observed, real data? I would