[Fwd: Re: Probabilities on Phylogenetic Trees]

Tom DiBenedetto tdib at UMICH.EDU
Tue Sep 16 12:41:04 CDT 1997


 Richard Zander wrote:

>>Tom DiBenedetto wrote:
>> A parsimony analysis deals simply with the distribution of character
>> states (the groupings implied by the state-distributions of each character), and
>> chooses the pattern which is most consistent with all of those groupings. It
>> makes no reference to process.

>Sure it does. You state below that there is no process other than
>descent...that's the process, and it is modeled simply and with a lot of
>assumptions when the model's pattern is imposed on the data.

No, when I say that parsimony is not a procedure which makes
reference to process, I do not mean that it deals with a pattern whch
has no causal basis. Yes, descent is a process, and parsimony can be
said to make reference to it in the sense that descent is the
explanation for the pattern. But this is very different from a model,
and cladistics does not attempt to model descent. There is no such
thing as the "model's pattern", so no such thing is "imposed on the
data" (except the notion of a hierarchy in general - but the issue
here is WHICH hierarchy is correct). Most of the work of a cladistic
systematist revolves around dealing with individual
characters,,,studying the character in various taxa to determine
whether they are the same, whether they are homologies,,,and if so,
then to explain their presence by invoking descent. Parsimony is
merely a logical standard by which one sorts out contradictory
evidence.

>Good. One of my queries on this thread was IF statistical phylogenetic
>analysis was valid, if if if,

I agree with the capitalization and iteration of that "if"....

> then can one assign posterior
>probabilities to the results of maximum parsimony analysis, as is done
>in maximum likelihood analysis. What assumptions must be made (rate of
>evol? what?).

yeah, I suppose, all those things which can not be known until the
phylogeny is in hand,,,hence the problem...

>Ah. Clearly you *don't* see theory imposed on data to wrench it into a
>pattern as I see happening. How you can say a step longer is an ad hoc
>explanation is beyond me.

Hmmm, perhaps you dont appreciate what an ad hoc statement is. It is
an explanation for a specific phenomenon which is not embedded in a
general explanation,,,an exception specially drawn to explain a
phenomenon which doesnt fit in the general explanation. Hair is found
in over 4000 taxa, and our assertion that it is a homology is a
general explanation for all of these phenomena,,,,these 4000
instances are really all one thing, and it arose one time in the
ancestral mammal. If one of our less parsimonious trees would (e.g.)
put bats with birds, then we would need an extra step (actually many
more, but for the purpose of this explanation let us just consider
hair),,,an extra step = an ad hoc explanation for why hair is found
in this group of birds,,,it must have arisen two times, not once. Ad
hoc explanations are what we would have if we never formulate
generalizations,,,every phenomenon has its own specially drawn
explanation. The basic goal of science is to establish valid
generalizations, to efficiently characterize regularities,,,,to find
the explanations which minimize the need for special explanations for
the "exceptions". An extra step is very clearly a paradigmatic "ad
hoc" explanation.

>  The reason scientists like simple/razor explanations is that these can
>be tested right away, most easily.

No that is not the reason. The reason scientists like the simplest
explanation is because they represent the general goal of science,,to
make statements about the world which have broader validity than
merely the immediate phenomenon in front of your eyes.

> Systematics cannot test its simplest
>explanations (other than by total evidence), and therefore we emphasize,
>wrongly, simplicity as an end in itself.

First off, what is wrong with total evidence such that  you consider
it something to be brushed off so easily? You are right in the sense
that cladograms are tested by ever more evidence from other character
systems, but that is a virtue, not a problem.
Simplicity IS a virtue in itself, in any scientific
explanation,,,consider the standards used in any process
investigation as well. Would you prefer e=mc2 or a special
explanation for every physical phenomena we encounter?

> My point is that we are not
>reconstructing anything, we are imposing theory on data. To the extent
>that the theory is assumed to be generally right, that's great.

What theory do you mean? Descent? Ok, we are trying to impose a
descent explanation on empirical character evidence. By so doing, we
are reconstructing the pattern of lineage branching through which
this descent process has operated. That is the point.

>> Given the results of process modelling to date, I sense that there
>> should probably be an expectation that the model will be different in
>> just about every different taxon (and for every different sequence).
>
>I should hope so.

Well good then,,,does this impress upon you the fact that using a
model drawn from one instance to establish the pattern in another
instance is to go about things in a bass ackwards way? Because the
second pattern you then "discover" is based on a model extrapolation
which you admit is probably  inappropriate? And that you have now
lost your chance to develop an appropriate model because you have a
biased pattern, biased by the very parameters you wish to learn
about?




More information about the Taxacom mailing list