[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]
Re: SICB Report
> I advocated this some time ago on the dinosaur list, and I have since learned
> that such simulations have already been performed here and there in various
> ways. In particular, this is the basis of my remark that about 10-18% of
> cladograms with four leaf nodes are incorrectly reconstructed. This percentage
> seems to be irreducible; different kinds of cladistic analyses yield different
> kinds of wrong cladograms, but generally in the same ratio.
Not correct. Different evolutonary conditions will most definitely
impact the degree to which a strict parsimony analysis recovers the
correct tree, or parts of it. This is something that we've seen in our
simulations, and that Huelsenbeck and Hillis' work showed.
> only way to find these is to compare them to the known control cladograms;
> there is no algorithm that will tell you in advance that a cladogram is
> incorrectly constructed (if there were, it would become part of the analysis,
> of course). I was surprised to learn that the percentage was so high.
Also not entirely correct. I agree that we can never *know* whether we
have the One True Tree or not, but we can go beyond simply saying
"Here's my tree - deal with it":
1. We can measure internodal support. Many morphologists (myself
included) have problems with bootstrapping, but there are other ways of
quantifying the robustness of a particular node. And this goes beyond
merely flagging weak nodes - see below.
2. We can quantify the degree to which one topology overturns another
with a given data set - for example, Templeton's Wilcoxon signed-rank
test, or likelihood ratio tests with molecular data. Or, for that
matter, some of Pete Wagner's stratigraphic tests, even though I am
skeptical of their utility for most vertebrate groups.
3. We can see how different sources of information (molecules,
morphology, behavior, biogeography, etc.) converge on one tree. And
surprise! most of the time, different data sets will give more or less
the same result. Even if we may never know The Truth, I would be
surprised if multiple genes and morphological systems preserved the same
signal merely out of chance. This brings us back to point 1 - usually,
when two or more data sets disagree, the nodes of disagreement will have
very low bootstrap/bremer/whatever support.
Something I've noticed - morphological data sets for a particular group
usually agree with one another. Theropods are an exception - this is
something I've had to adjust to, as in my previous life I was involved
with a community that was basically producing the same tree. And I
think this has more to do with differences in taxon sampling than