- Still writing. Working on the comparisons to AIC section. Things to think about adding:
- Explicitly show p values on bootstrap
- What are we calling this thing anyway? After Hulsenbeck & Bull, 1996: ``likelihood heterogeneity test
- All-different seems to have some stability/convergence problems, perhaps a less aggressive example would be better? --Carl Boettiger 19:46, 5 October 2010 (EDT) Stability issues resulted from different fitting methods in initial fit vs. bootstraps. Using nelder-mead (with increased error tolerance) gives consistent, convergent results (that also result in better fits than the box-limited method of before).
- revisit intro to add map section
- Avoid confusion between simulating from a model and assuming that a model is "correct"
- What to do about OU.1? use the improved fit info...
- How about power examples for multiple peaks?
provide power test between models (return the power directly). So far only do the full power curve on the tree over alpha for OU1 vs BM.
More power tests
Power between OU.2 and OU.LP -- a close call? The observed ratio (dark red) seems on the border between the ratio's expected under one case and under the other. 1000 replicates.