Re: Underlying basis of classification (Was: Re Dinobirds)

[chris brochu <cbrochu@fmppr.fmnh.org>]

Philidor11@aol.com wrote:
> 
[shortened]

> 
> <<This is why we usually call trees "recovered results" or "estimates" rather
> than hypotheses - and it's also why so much time and paper are being spent on
> measures of support and robustness for different kinds of trees.>>
> Another possible word, if you want one, is 'plausibility'.  The primary
> meaning of the noun seems to fit.

This wouldn't really fit here, because to most people, "plausibility"
carries a process connotation, and we're dealing fundamentally with
patterns.  I would refer to a theory explaining a pattern in terms of
plausibility, but not the pattern itself.

Example:  Trees putting birds within Theropoda are more robust than
those putting them within Crocodylomorpha.  The theory that flight arose
from ground-dwelling bipeds is more plausible, based on the patterns
available, than one in which flight arose from semiaquatic quadrupeds.



> 
> I said:  <<Unless someone has insisted on the existence of avialae, new
> analysis has produced characters which apparently were present before the
> prior, feather-based analysis and which come to the same result.  This is
> important because it implies that there is not much difference among the best
> results of the cladistic analyses.>>
> You said: <<I'm not sure what you mean here.>>
> I've never seen the output of a cladistic analysis computer program.  I was
> dealing with the issue of whether the results I do see are significantly
> better than rejected hypotheses (those not used).  

They may or may not be - and there are quantitative ways of testing
this.  I've posted some of the references on this before.



Though I had to draw my
> conclusion with analyses that produced the same results (avialae,
> ornithiscians), these alternate sets of characters which both had acceptable
> results told me that a cladistic analysis could also produce contradictory
> results which were both acceptable as far as the program is concerned. 

As far as a given data set?  Agreed - but the best explanation is that
which accounts for *all* of the data.  When different morphological data
sets clash, one of three things is happening:

1.  Different morphological interpretations of the same fossils.
2.  Different taxon sampling.
3.  Different character sampling.

(1) is a general problem, and is "solvable" only by explicitly figuring
all characters coded and with rigorous attention to actual fossils
rather than the literature.  (2) and (3), though, are not intrinsic to
cladistics, but human limitations - is it the fault of phylogenetics or
a budget limitation that, for example, certain South American fossils
might be excluded from an analysis performed by a North American?  The
solution is to include all relevant taxa and characters.

>From my experience, most differences between dinosaur cladograms can be
attributed to 2 and 3.



 That
> in turn meant that the program is not capable of saying that as far as it is
> concerned there is only one possible good answer for any given animal or set
> of animals.  Or I guess I could have just asked.

Correct - it can only give you the best set of trees for a given data
set.  



> 
> <<We don't normally "look at all possible common ancestors."  Rather, we look
> at all possible hierarchies of relationship - or more accurately still, we
> let our computer algorithms do it for us.  That ancestors "exist" on the
> nodes is understood, but
> the search is for the hierarchy.>>
> I was struggling with the fact that the ancestor would not 'exist' without
> the diagnostic character set, but supposedly the characters did not define
> the ancestor.  This really is an apparent contradiction.  


Why are you a member of your family?  Is it because you have a
characteristic nose shape, or third nipple, or extra digit?  Or is it
because your parents were part of that family?

I am not a Brochu because of any physical traits I possess - in fact, my
sisters and I are very dissimilar in outward appearance.  I have that
name for definitional reasons - my father had it.

The program gets
> around this problem by creating ancestors and diagnostic character sets at
> the same time and then accepts or rejects based on its algorithms. 

But it doesn't create ancestors!  It creates hierarchies.  We interpret
branching points as ancestors, but the algorithms are simply creating a
hierarchy that we humans like to read as a branching diagram.  We could
just as easily depict it with paragraphs or Venn diagrams.



 In terms
> of what you're saying, the ancestor is implicit in the hierarchies of
> relationship; if there is a node there must be an ancestor.  You'd produce
> the same result by saying there is a set of all possible ancestors, an
> ancestor for animals in every combination; the program just eliminates the
> implausible ones.

No, this would not be the same thing.


> 
> <<Morphology and molecules should support very similar trees, and most of the
> time, they do.  Stratigraphy and biogeography should likewise preserve
> phylogenetic signals, and having a tree match stratigraphy (or come close to
> doing so) is regarded as a kind of congruence.  But having biotic
> (morphology, molecules) and nonbiotic
> data disagree is not (or should not) be viewed as strong evidence against the
> tree, as the nonbiotic data are not preserving phylogenetic signals alone,
> and we don't know how strong the phylogenetic component is relative to
> others.>>
> What bothers me about this is say that a cladistic 'hypothesis' is
> contradicted by subsequent data, finding the 'black swan' in terms of my
> Popper-Tart thread discussion with Mr. Holtz.  There seem to be good reasons
> for saying that the contradictory data should not be decisive.  Are there any
> firm rules for saying that the amount of evidence against a 'hypothesis' is
> sufficient to reject the 'hypothesis'?  Or can you keep on finding black
> swans for a long time without rejecting the hypothesis 'all swans are white'?

It depends on the specific case being discussed.  There are ways of
putting numerical confidence on a tree, though their applicability to
morphological situations is debatable.


[shortened]

> 
> <<The key to testing hypotheses is having repeatable observations.  In the
> experimental world, we generate these by replicating the process in the
> laboratory.  But we historical scientists can't do that, since the
> "experiment" has already been run.  Where philosophers sometimes trip is with
> the concept of "repeatable" - it's not the *process* that must be repeated,
> but the *observation* - and as long as we continue to resample from nature,
> we can repeat our observations.>>
> Well, as noted in the arbitrary paleontology thread, when cold fusion and
> sheep cloning were being examined, other scientists did not look at the lab
> notes or even the sheep; they tried to follow the recipe from scratch. 

Looking over the notes would not be "replicating an observation" - it
would be double-checking a previous observation.  And I guarantee they
did that as rigorously as they did replicate the process.




The
> 'resample' you're talking about does allow checking the process with new
> material, though this is still different from the experimental world's use of
> repetition, and I don't think you mean re-doing the same thing the initial
> investigator did on the same material.

No - though sometimes it can help to recheck the same material.  Anyone
who's compared a fossil with an older description of same knows what I
mean.



chris

-- 
----------------------
Christopher A. Brochu
Department of Geology
Field Museum of Natural History
Roosevelt Road at Lake Shore Drive
Chicago, IL 60605

voice: 312-665-7633  (NEW)
fax: 312-665-7641 (NEW)
electronic:  cbrochu@fmppr.fmnh.org