Dear Rafael.
I believe the standard errors are computed from (negative inverse of)
the Hessian matrix - which is a matrix containing the second-order
partial derivatives (or some numerical approximation of them) from the
likelihood surface.
These values are measurements of the curvature of the likelihood
surface. If the likelihood surface is highly curved (downwards, that is
negatively) then this means that the ML solution is much more likely
than other nearby possibilities, and the negative inverse of this value
is a small quantity - indicating little variance (i.e., uncertainty) in
the estimated parameter. Conversely, a large standard error (the square
of which is the variance) indicates that the likelihood surface is very
flat (that is, it has a very small negative curvature) around the ML
solution.
In your particular case broadly overlapping CIs for the parameter
estimates (which can be computed as theta+-1.96*SE) of theta probably
mean that the 'adaptive peaks' of different regimes can't be
distinguished one from the other; whereas a CI for alpha that included
zero (for instance) might suggest that a BM model probably better fits
the data.
All the best, Liam
Liam J. Revell, Associate Professor of Biology
University of Massachusetts Boston
& Profesor Asociado, Programa de BiologĂa
Universidad del Rosario
web: http://faculty.umb.edu/liam.revell/
On 4/4/2018 2:30 PM, Rafael S Marcondes wrote:
Dear all,
I'm writing (again!) to ask for help interpreting standard errors of
parameter estimates in OUwie models.
I'm using OUwie to examine how the evolution of bird plumage color
varies across habitat types (my selective regimes) in a tree of 229
tips. I was hoping to be able to make inferences based on OUMV and OUMVA
models, but I was getting nonsensical theta estimates from those. So
I've basically given up on them for now.
But even looking at theta estimates from OUM models, I'm getting really
large standard errors, often overlapping the estimates from other
selective regimes. So I was wondering what that means exactly. How are
these erros calculated? How much do high errors it limit the biological
inferences I can make? I'm more interested in the relative thetas across
regimes than on the exact values (testing the prediction that birds in
darker habitats tend to adapt to darker plumages than birds in more
illuminated habitats).
I have attached a table averaging parameter estimates and errors from
models fitted across a posterior distribution of 100 simmaps for four
traits; and one exemplar fitted model from one trait in one of those
simmaps.
Thanks a lot for any help,
*--
*
*Rafael Sobral Marcondes*
PhD Candidate (Systematics, Ecology and Evolution/Ornithology)
Museum of Natural Science <http://sites01.lsu.edu/wp/mns/>
Louisiana State University
119 Foster Hall
Baton Rouge, LA 70803, USA
Twitter: @brown_birds <https://twitter.com/brown_birds>
_______________________________________________
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/
_______________________________________________
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/