Hi Franz,
Could this be a separation issue? Even if you cannot trust the output,
looking at the estimates is informative. It can tell you where the
region of high likelihood might be, and if the convergence problem comes
from alpha going to an extreme, or from the estimated probabilities
going to 0 or 1 (that would indicate a separation issue).
I don't recommend increasing the btol value.
You could also try the "logistic_MPLE" method, rather than
"logistic_IG10". In my experience, maximizing the penalized likelihood
directly (MPLE) converges better and has better accuracy.
Cecile.
On 07/06/2015 03:49 AM, [email protected] wrote:
Hello,
after a discussion with Brian (see [R-sig-phylo] corDISC binary coding
problems, 04/15) I used the method from Ives and Garland (2010),
which is implemented in the R package phylolm.
The problem is, that the likelihood estimator does not converge.
Here is my code:
fit <- phyloglm(rot.fill ~ Acrogymnospermae + Magnoliophyta,
data=dat, phy=data$phy, method = "logistic_IG10", btol=20,
log.alpha.bound = 4, start.beta=NULL, start.alpha=NULL)
And the warning message:
In phyloglm(rot.fill ~ Acrogymnospermae + Magnoliophyta, data = dat, :
phyloglm failed to converge.
So the function produces an output but the results are not reliable since the
global maximum is very likely not reached...
I set the btol value to 20 by try and error-increasing (which I don’t really
understand since I thought the bounds converge to 0 and 1 at about btol=10 ??)
Now I have to find suitable start.beta and start.alpha values. But I have no
clue how to approach them. Can anyone help?
Why doesn’t phyloglm has a option for iterating starting values?
All the Best,
Franz
Franz Krah
Personal Webpage: http://franzkrah.github.io <http://franzkrah.github.io/>
University: http://www.biodiv.wzw.tum.de/index.php?id=18
<http://www.biodiv.wzw.tum.de/index.php?id=18>
[[alternative HTML version deleted]]
_______________________________________________
R-sig-phylo mailing list - [email protected]
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/[email protected]/
--
Cecile Ane
Departments of Statistics and of Botany
University of Wisconsin - Madison
www.stat.wisc.edu/~ane/
CALS statistical consulting lab:
www.cals.wisc.edu/calslab/stat_consulting.php
_______________________________________________
R-sig-phylo mailing list - [email protected]
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/[email protected]/
