Hi,
I've been working on some simulations to estimate type I and II error rates
for my models, since these models are attempting to infer hidden states and
processes. I'm especially interested in knowing how much power, if any, I
have to detect a hidden precursor, when present.
While my simulations work for most of my models, in one model the
precursors are evolving at extremely low rates. As a consequence, several
of the simulated datasets produce no data at the tips (i.e. if precursor
doesn't evolve, then the trait it facilitates doesn't evolve, either!).
When trying to re-fit the simulated data sets, my script gets jammed on
these "empty" datasets. Can't fit a corHMM model to no data, can you?
I can think of a few possible routes of resolve this. Open to other ideas,
or critiques if any of these is a dangerous path to tread!
1. I could write a conditional statement within a for loop to refit only
simulated datasets if they have D > 0 data at the tips.
2. Could I assume an LR of 0 for any models fitted to "empty" datasets? I
am assessing the simulation outputs based on likelihood ratio statistics.
Presumably, as the as data at the tips approaches 0, so should the ratio
between likelihood scores (i.e discriminatory power is depreciating)?
Ultimately, I'm concerned with biasing the LR distribution if I go about
this the wrong way. Looking forward to your input!
Thank you!
Michael Foisy
MSc. Student, Rodd & Mahler Labs
University of Toronto
[[alternative HTML version deleted]]
_______________________________________________
R-sig-phylo mailing list - R-sig-phylo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
