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

Re: Popper and Palaeontology (was: Re: "running" elephants - locomotary analoges)

John Conway wrote (quoting me):

> >     I have never been happy with this characterization of parsimony,
> > and I think a number of others share my displeasure. [...]

> I don't think your example really supports your point. [...] the universe
> [...] is "newtonian enough", just as people say that parsimony
> analysis presupposes evolution to be "parsimonious enough".

    I was objecting to the assertion that an assumption of parsimony methods
is that evolution is parsimonious. In that context, I feel that my example
more than supports my point, as does yours (in the case of newtonian
mechanics); the model need not be a perfect fit to reality to be better than
other models. As to whether evolution is "parsimonious enough," you are
obliged to explain what you mean by the assertion that "evolution is
parsimonious" before you qualify it with the word "enough." Which was my
second point.

> But it is meant to model the >pattern< of evolution - otherwise no-one
> would bother doing it. By pattern of evolution, I mean the arrangement
> of the real characters in respect to clades.

    Are characters real? Shall we dance?

    Actually, I hate to weasel, but I really don't have time to dance now.

"Weaseling out of things is what separates us from the animals... all except
the weasel." {H. J. Simpson)

> And why would you want to minimise assumed homoplasies, if you don't
> think they are minimal (i.e. unlikely)?

    No one every asserted that they believe homoplaisy is UNLIKELY... I
suspect that, as an estimator of phylogeny, parsimony is robust to a large
proportion (ergo, likelihood) of homoplaisy. You should specify whether you
are concerned specifically with the case in which homoplaisy is rampant, or
the case in which homoplaisy is MORE LIKELY than synapomorphy. The former
case appears to describe the situation in many datasets, and there is a case
to be made for the utility of parsimony methods under those circumstances.
Down the latter route lies madness. Seriously, there are some places science
can't go... if homoplaisy is more likely than synapomorphy, and you have NO
other data (e.g., branch length estimates), I think you can see that the
odds of getting a useful result are pretty low. The answer, under such
circumstances, is not to abandon parsimony, but to abandon attempts to
reconstruct phylogeny alltogether. There are, in fact, caes where certain
datasets ARE uniformative, because too much change has occurred and either
no signal can be found in the data, or the data are simply incomparable. In
rapidly evolving gene regions, multiple substitutions at particular
nucleotide positions can effectively erase the phylogenetic signal, and
insertion/ deletion events can render the sequence unalignable, among widely
divergent species. Under such circumstances, we cannot expect any method to
produce the correct tree, at least not with any confidence.

[out of order]
> That's to say, the pattern
> of evolution must be parsimonious enough to be approximately recovered
> by a parsimonious arrangement of data. I don't see how the parsimonious
> arrangement of data could possibly reflect a non-parsimonious real
> arrangement. [...]
> Although the statement that "Evolution is [...]* at least [...]*
> parsimonious enough" was not mine, I think it can be clarified by
> replacing "evolution" with "pattern of evolution" (as above). [...]
    [* indicates original brackets]

    Are you trying to say that, for parsimony analysis to be valid, the true
tree must be the parsimony tree for all the data? I don't think anyone is
under the illusion that this is the case, and I don't think anyone would
insist it must be. Again, this the newtonian mechanics issue: we cannot ask
for a method that WILL reproduce the correct tree, given all the data, we
can only ask for a method that brings us closer to the true tree than all
other competing methods. Since, until recently, there was NO accepted method
other than parsimony for discrete morphological data, parsimony wins by
default. If you want to dethrone it, you need to come up with a new method.

> >     I don't miss the Clade Wars. :)
> And I missed the Clade Wars. :)

    Like all wars, they will ultimately be laid to rest not by the triumph
of one ideolgy, but by the death of the last remaining practitioners of the
others. Until then, bitter misunderstandings and echos of old battles
resound in classrooms, lecture halls, and now list servers, while the
victors push on in the march of progress. In phylogenetics, as in space, all
warriors are cold warriors. Or something like that...