Closed jwdink closed 8 years ago
Seems cool. See /Misc/GCA-based-divergences.R
It has some ambiguities/concerns: do you get CIs from only models w/all significant coefficients? E.g., ot1 was significant, but not ot2,ot3, etc. Only fitting ot1 (even when the curve is clearly curved) seems like it could lead to wild misestimation in time-of-divergence, although obviously would need simulations to confirm.
For now, not enough motivation to add this (not clear it adds anything over clusters). Might circle back in future.
This could be a theoretically and practically good way to compute divergences