Thanks for clarifying this, Joe.
My function for ancestral character estimation under the threshold model
(ancThresh) allows multistate data - but the order of the states along
the liability axis needs to be specified a priori by the user.** The
relative positions of the thresholds are sampled (along with the
liabilities of tips & nodes) from their joint posterior probability
distribution using MCMC. This method is still undergoing peer review,
but it seems to work fairly well in simulations - so this also might be
a viable approach for estimating the correlation between multistate
threshold characters.
(**In my manuscript about ancThresh I suggest that an information
criterion like DIC could be used to identify the correct order - but a
single true order must exist.)
As for unordered multistate characters, I'm not sure how the threshold
model could be used. Several people have made suggestions about this to
me but I'm not sure I understand them.
All the best, Liam
Liam J. Revell, Assistant Professor of Biology
University of Massachusetts Boston
web: http://faculty.umb.edu/liam.revell/
email: liam.rev...@umb.edu
blog: http://blog.phytools.org
On 4/9/2013 7:22 PM, Joe Felsenstein wrote:
New Bio wrote:
Thanks, Liam. good to know. I think extending the tools to analyze traits
of ordered/unordered multistates can be very useful. There are many
interesting traits such as oxygent requirement (anaerobic, facultative,
aerobic) of microbes, which is ordered multistates, and habitats (water,
air, soil), which is unordered.
For approaches like the threshold model, it is not simple to see how to
handle multistate characters. Should we assume one scale? If so, how
far from the 0/1 threshold should we place the 1/2 threshold? That
becomes an additional parameter to be estimated.
Or should we have an additional axis, so that 0/1 is on the x axis,
while [01]/2 is on another axis? And then what do we do about
state 3 if it also exists? That way lies madness ... or perhaps a
good Ph.D. thesis topic.
(This is in some way related to Transformation Series Analysis,
which was an issue with parsimony methods. One imagined one's
states arranged in a character "transformation series" which could
even be a tree, the Character State Tree. Then one wanted to
infer the phylogeny and at the same time also the CST, using
parsimony as the criterion. In a sense what I am raising is the
likelihood and modeling equivalent problem.)
Joe
----
Joe Felsenstein j...@gs.washington.edu <mailto:j...@gs.washington.edu>
Department of Genome Sciences and Department of Biology,
University of Washington, Box 355065, Seattle, WA 98195-5065 USA
_______________________________________________
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/
