[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]
Re: SICB Report
John V Jackson wrote:
> DinoGeorge says: The key problem with cladistic analysis is that it is
> >inherently< unfalsifiable.
> That may be so for an individual cladogram, but the effectiveness of the
> basic system may be tested by simulation. Over Christmas, it occurred to me
> that various imaginary scenarios of trees of virtual creatures could be run,
> all busily evolving their characters at controllable rates, and the results
> checked. We could see under what frequencies of character reversals etc the
> actual correct tree would be found, for any cladistic process. With
> increasing numbers of feasible simulations, the envelope of possibilities
> will be increasingly adequately sampled this is called Monte Carlo
> simulation, and in fact Peter Wagner (probably a friend of Chris Brochus)
> advocated them in the Nature discussion Is the fossil record adequate?
> at the end of last year.
Simulations have been a staple of experimental phylogenetics for many
years. In fact, I HIGHLY recommend reading John Wiens' paper on
incompleteness in the most recent Systematic Biology. (And read the
paper - not just the abstract.) And along with John Merck and Colin
Sumrall, I've been doing some simulation work myself to address the
impact of taxon lability and incompleteness.
Overall, these simulations show that parsimony usually gets the correct
tree, or very nearly the correct tree. Some of George's statements
about being 18 percent wrong or something are not entirely correct - the
simulations identify the conditions under which the tree will be
correctly recovered, under which it will be slightly misled, and so on.
They also identify the nodes that are incorrectly recovered, and the
ways in which we can recognize them in real data sets.
> In fact, since most cladistic processes assume independent characters, they
> will not be expected to handle dependent ones.
Ah, but they frequently do - especially in molecular analyses, where
different base pair biases can lead to specific a priori weighting
schemes that account for transformational dependency. And this has been
simulated very well.
Of course, when it comes to
> morphology, independence certainly cannot be relied upon. Consider for
> example the necessity for a bipedal creature to balance short-tailed forms
> usually need ground contact further forward hence relatively immobile
> femurs hence longer legs. Thats one weve been reminded of recently.
> And of course all the (other) flying/ex-flying features are not independent
I think you're confusing "functional correlation" and "phylogenetic
dependence." Two characters can be functionally correlated (or even
flat-out linked), and yet be considered independent for phylogenetic
purposes. All that's required is that they transform at different
> The sixth contributor to Is the fossil record adequate was Dr Peter Wagner
> (Field Museum of Natural History, Chicago, USA) who wrote on:
> Why phylogenetic hypotheses need testing by stratigraphic data
> He warns about problems with cladistics: Unsurprisingly, simulations
> confirm that parsimony analysis is less accurate as homoplasy increases and
> global hierarchy decreases.
> Basically, if I interpret him right, he restates the often heard complaint
> that the homologues cladisitcs is based on deteriorate in usefulness over
> lengthening time spans.
Yes, but something to bear in mind - Pete is approaching this problem as
an expert on Paleozoic gastropods. These animals have very few
characters and a great deal of time on their hands - the situation he
faces is not entirely comparable with that faced by arthropod or
vertebrate specialists, where we have a much larger pool of character
information and not as much time, at least within specific lineages.
I am also not convinced by his arguments on "character saturation."
Pete's been arguing that character space gets saturated over time, and
that the same states keep appearing over and over. The problem? He's
basing his conclusions on data sets from the literature, and on data
sets restricted to hard-part characters - there are two non-random
aspects that could bias his results quite strongly.