Re: Prolacertiformes and Protorosauria

[Mike Habib <habib@jhmi.edu>]

On May 19, 2009, at 7:27 PM, David Peters wrote:

> The data set in question is 14x larger than any prior published data  
> set covering the same taxonomic spectrum. With it I can reduce the  
> dataset to the size of the smaller ones and often (not always) get  
> their topology.

I'm not sure if you mean 14x by taxon number, character number, or the  
product of the two (i.e. total matrix size), but in any case, having a  
much larger matrix has its advantages.  However, it is worth noting  
that increased taxon sampling does not necessarily provide a better  
answer - Scotland et al. (2003) discussed this property, and  
simulation tests have demonstrated this somewhat counter-intuitive  
result.  That's not to say that you don't have an interesting result:  
you have shown an important taxon sampling effect.  The question that  
remains is: what exactly about the inclusion sets is driving the  
topology changes? Can get some measure of confidence between the two?   
The larger set is more practically useful *if* the topology is stronger.

> Adding taxa to see if the tree shifts has also been a good test of  
> tree topology. At some point in the addition process, tree shifts  
> become less and less possible because you're reaching a saturation  
> point where the weight of the rest of the tree is becoming harder to  
> shake.

That saturation effect is not altogether unexpected, and it is  
interesting - I'm not sure what it really says about topology  
confidence, though.  I'd have to see some simulation results to have a  
better feel for the empirical ramifications.  I do like the  
sensitivity analysis approach, though, for what that's worth.  I'd be  
quite curious to hear what the theoretical phylogenetics crowd has to  
say about these effects.

> If the lack of confidence and lack of resolution are the result of  
> taxon exclusion, then PAUP offers a clue that some or all of those  
> excluded taxa are better off  included. All I've ever asked for is  
> inclusion. If someone excludes pertinent and previously taxa to  
> protect a precious hypothesis, then they're doing a disservice to  
> their own studies.

If that is the reason for exclusion, I agree.  But if an author  
excludes a taxon because they don't feel confident in coding it - it  
is fragmentary, or they have not had sufficient access to the  
specimen, for example - then it might be justified.

> One reason is that it often happens that a taxon can occupy two  
> equally parsimonious places in the tree, _no matter how robust the  
> rest of the tree is_. (Especially when it's incompletely known and/ 
> or there are just too few characters in the matrix.)
>
> >>>>>> Should be right next to the same sister taxa, though. They're  
> never widely separated (given the above parameters). Important  
> distinction left unsaid. And this is due to the fact that some  
> sisters are really different, or not much is known of one or the  
> other or both.

They don't have to remain with the same sister taxon - it is quite  
plausible (and reasonably common) for a taxon to move substantially  
throughout the tree, without any change in total calculated parsimony  
(i.e. total tree length is constant).

Cheers,

--Mike


Michael Habib, M.S.
PhD. Candidate
Center for Functional Anatomy and Evolution
Johns Hopkins School of Medicine
1830 E. Monument Street
Baltimore, MD 21205
(443) 280-0181
habib@jhmi.edu
>