[Mesquitelist] BiSSE and hypothesis testing

[Peter E. Midford]

Stacey,
               We have checked BiSSE against a birth-death model  
(removing the q's), but
haven't specifically compared with Pagel's model, which is  
implemented differently.
Since Mesquite has an implementation of Pagel's model (though only  
the 4 and 8
parameter versions), you could try it, but I doubt the likelihoods  
you get will be
easily comparable.  BiSSE uses a system of differential equations,  
while our Pagel
model uses matrix exponentiation to calculate likelihoods.  I don't  
know what Pagel's
Discrete used, but they clearly used a different likelihood optimizer  
than we use.  Since
Diverse and Correl (where our Pagel implementation is found) include  
simulations of
their models, you might be able to evaluate likelihoods using  
simulated distributions.

Hope this helps, and thanks for trying BiSSE,

Peter


On Aug 28, 2007, at 8:55, sds21 at duke.edu wrote:

> Hi all,
> I am using the BiSSE model in v.2.0beta3 to estimate transition  
> rates taking the
> character's effect on speciation (lambda) and extinction (mu) into  
> account.  I
> would like to test whether including these additional parameters  
> (lambdas, mus)
> produces a better fit than simpler Pagel style transition rate  
> model with just
> q01 and q10. So I have two questions:
> 1.  If lambda and mu are taken out, does the BiSSE reduce to  
> Pagel's model?
> (i.e. is Pagel's model nested in BiSSE allowing likelihood ratio  
> tests?)
> 2.  If so, is there a way to do this in Mesquite (compute  
> likelihoods and q's
> for a bunch of trees and single binary character)?
> Thanks in advance for your help!
> Stacey
>
> _______________________________________________
> Mesquitelist mailing list
> Mesquitelist at mesquiteproject.org
> http://mesquiteproject.org/mailman/listinfo/mesquitelist

--
Peter Midford
Mesquite Developer




-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mesquiteproject.org/pipermail/mesquitelist/attachments/20070828/391677f5/attachment.html