[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]

Re: Prolacertiformes and Protorosauria

On May 17, 2009, at 9:05 AM, David Peters wrote:

Yes, there is some intuition that goes into judging results.

When I do phylogenetic reconstruction (or when I review the reconstructions that others have published), I try very hard not to judge by intuition - if we do that, then we might as well build the trees "by eye", exactly as was done pre-cladistics. Part of what is so useful with matrix-based phylogenetic reconstruction is it allows you to find well-supported hypotheses that you would have otherwise overlooked because they are counter-intuitive. Intuition has it's place, but I'm not sure this is it.

ie. placing mesosaurs with pareiasaurs = bad.

Ah, but how do we know it is "bad"? Perhaps we get a quick intuitive twinge that tells us something might be amiss, but if you actually want to add some weight to such an assertion, you have to demonstrate that something isn't right more objectively (I'm not saying that mesosaurs and pareiasaurs must be related, incidentally). For example, perhaps the taxa in question almost never come out together in any trees that anyone runs, with various character sets, except for one dataset with a very limited number of taxa or characters included. Then we might be able to, with some degree of objectivity, indicate that the "bad" clade is indeed likely to be a fluke, because alternative pairings are supported in other trees that seem to be more robust in their methodology. Or perhaps consistency index or branch support metrics suggest that a given topology is weak. Stratigraphic indices might do the same. There are a lot of data-driven ways to demonstrate that a given clade composition is questionable. It's important to knock them down with data, though, rather than our own sensibilities, alone.

More importantly, there is some experience at work here too. Yes, I've done the larger, more inclusive study, that indicates the breaks are real and taxon exclusion is the culprit.

Alright, that might be more informative - the question I have here is how does the larger, more inclusive study indicate that the breaks are real? I presume are you saying that you get a different topology with your dataset than the one published. No problems there - the question then becomes: why is your tree more robust? Just because it has more taxa doesn't mean it's more accurate. It *might* be, but other indicators are important. It might be case that we have no objective reason to prefer one tree or the other at this time, in which case we simply have two alternative topologies, each with some support from data. That's fine, and happens all the time, but to really push one topology as "better" requires more than an intuitive sense through experience (though intuition and experience don't hurt).

As for your 'by definition' comment, yes. True. Even so, 'by default' is also at work here. The authors were working from too large a gamut and too small of an inclusion group to make sense. There are better sister taxa out there. The larger study would have revealed this.

I see what you are arguing, but consider this: if we are to judge trees entirely by how many taxa they include, does that then mean we cannot trust any phylogeny for, say, Amniota that does not include every amniote known? Clearly there are thresholds in there somewhere that we recognize - so, stepping back and being a little less extreme: what percentage of the group do you need to be confident? Keep in mind that for any large fossil vertebrate lineage, you might be working with 1%-10% or so of the taxa that actually existed.



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