Tuesday, May 18, 2010
Maddison et al. 2007
I think it is neat that this is the first paper to incorporate extinction rates into a model of character evolution and diversification (that I have read). Is this why Rick Ree said last week that models prior to 2007 should probably not be used??? If so, this paper seems to be an impressive step forward. I am wondering about delta t. What are the effects of increasing and decreasing the time slice? Why does the model shrink the time interval as it approaches the root? Do high degrees of homoplasy in characters alter the performance of the model? Also, I am wondering why the option of state change and speciation is not included in figures 2 and 3? Following last week's discussion of key innovations, that strikes me as an important combination.
Subscribe to:
Post Comments (Atom)
I think the main concern with earlier models is that they have the potential to confuse different rates of transition between character states with different rates of diversification for different character states.
ReplyDeleteHowever, I do think it is interesting that the BiSSE model must include extinction and that there is no option for a pure birth model. I don't think that pure birth models are plausible, but on the other hand it is hard to estimate extinction and origination separately on molecular phylogenies, so why not just estimate net diversification rate under each character state and use fewer parameters? My guess for why this is not done is because the birth-death process produces different branch length distributions depending on the extinction rate, and you can't compute likelihoods for a given net diversification rate in this model without knowing something about both speciation and extinction.
I don't think that the model shrinks delta t as it approaches the root, when they refer to "shrinking delta t" they mean that they are turning it into a derivative rather than having it divided into finitely many segments. Of course in practice when they implement their model they have to divide it into small discrete segments. As far as I can tell they do this by splitting each branch into a fixed number of segments, the number of which is partially determined by its length.
Finally, I think the reason why the option of simultaneous speciation and character state change is not included is because the time window delta t is assumed to be sufficiently small that the change of both occurring within the window is effectively zero. Key innovations are thought to affect the speciation rate, but not necessarily cause speciation instantaneously. However, I guess this might be a poor assumption if trait changes are concentrated at speciation events.