Who is the postivist?
James Francis Lyons-Weiler
weiler at ERS.UNR.EDU
Mon Dec 8 07:46:54 CST 1997
On Sun, 7 Dec 1997, Tom DiBenedetto wrote:
> It was not parsimony per se which made cladistics non-authoritarian,
> but you could say, in a sense, that its formalism helped it to be so.
> The point was simply that character evidence should be explicitly
> laid out, with detailed definitions, and an algorithmic approach to
> combining evidence. This formalism did not presuppose any particular
> stance on the relevant issues; for instance, it doesnt preclude
> weighting characters; it merely asserts that one must lay out all the
> evidence that was used to arrive at a conclusion, justify weighting
> some characters more than others if that is what you do,,,,basically
> make the whole procedure transparant. This is very much in line with
> the notion of science as a critical process, open to any person or
> any idea which can arrive at a more compelling explanation of the
So, let me get this right - if I select a page at random
from any book, and go the 14th line and find the third
letter of the first word, and if it is a vowel, I will
encode a similarity as a synapomorphy, if not then
I discard it as not a good hypothesis of homology,
then I'm doing Hennigian cladistics - simply because
it's explicit? There is more to it - clearly my explicit
facetious approach does not try to mirror (model) evolution,
whereas Hennig had a model of what evolution does to
character distributions - and he apparently consistently
felt confident that evolution rarely confuses us, but more
often illuminates. That would be a very nice reality to
> > But just because
> > one group of people agree that parsimony is the best way
> > to perform a phylogenetic estimation problem doesn't make
> > the trees correct.
> ah,,,yeah,,,so? Who ever claimed that? I merely assert that the most
> parsimonious tree is the logically ordered set of corroborated
> hypotheses. I dont see any method which does better.
Every time you state that parsimony renders the best
hypothesis of organismal relationships and stories of
homology, the question of accuracy is not far behind.
And there are some fairly well known conditions in which
other methods outperform parsimony. Parsimony outperforms
itself when the data are subjected to the critical examination
for long edges, for instance.
> > Hennig moved the field to become
> > "more scientific", but others have helped to finish the
> > job to move phylogenetic science - beyond parsimony.
> Ah yes, let us reject some of our homology hypotheses even when there
> is no reason to do so! What a wonderful idea! It should prove to
> everyone that we are passionate falsificationists!
I'm not sure what you mean by rejecting hypotheses without
reason. If you mean rejecting the auxiliary hypothesis,
that suggestion leads to stating relationships on the
basis of mere confirming instances until we are lucky
to find evidence to the contrary. However, as I have
indicated earlier, this leads to the awkward illogical
inference that part A is falsified if we study part A first,
then discover trait B, but if we study trait B first, then
find part A, B is falsfied. This timing effect (for which
I have an ms from some years ago) is fairly prevalent in
evolutionary science, but it need not be. The real
source of "confidence" is not the number of unfettered
synapomorphy estimates, but rather the number we find
for a group beyond which we expect by chance alone. If
there is a remarkble number, given the most critical test
we can devise (and parsimony won't do here), then
we can consider the result worth talking about - but even then
there is the major assumption that the significant amount
of covarying changes (some literature on this, actually),
reflects the kind of evolutionary processes we think
useful for classifying organsism and for making statements
about homology. If we reject the hypothesis through
significance testing, then we have seen the type of evidence
to the contrary that results from a justified, strong
> A synapomorphy does not support a hypothesis of homology; it IS the
> hypothesis, left standing after the test of congruence. Neither does
> it "support" the hypothesis of monophyly, for it IS the hypothesis of
> monophyly, left standing after the test of congruence.
> possess it). After the test of congruence, they remain hypotheses,
> only now a bit more corroborated. A cladogram merely represents the
> logical ordering of individual corroborated hypotheses of
> homology/monophyly. Dont get hung up in casual language.
I'm not hung up - I just think we can do better.
I think you've oversimplified things again. Hypotheses
of monophyly exist independently of the evidence used
to test them. For instance, I might pose the conjecture
that primates are monophyletic. The hypothesis
is there, and evidence is waiting to be collected
as test statements.
> Amen; may I direct you again to "Realism and the Aim of Science" in
> which Popper goes on at length about the subtle differences between
> the two terms, and why he prefers the latter to the former.
> Breifly, "confirmation" is a term which arises from the
> verificationist approach; those who try to validate hypotheses by
> searching out confirming instances. "Corroboration" is meant to refer
> to confirming instances which arise in the course of critical
> testing, serious attempts to refute the hypothesis. Hypotheses gain
> credence as a function of the testing they have endured. A hypothesis
> of monophyly (a character) is tested against all other hypotheses of
> monophyly (characters); the test being the expectation of congruence.
This is where I differ - the past is so vast, complex
processes about - corroboration should come through
congruence for the very opposition reason - that we
don't expect it. But that requires some statement
of the amount of congruence one might reasonably
expect; hence the null hypothesis.
> Confirming instances (congruent distributions) are thus rightly
> referred to as sources of corroboration.
I don't see it that way.
> who was debating with whom?
This section should be cited as Lyons-Weiler, in prep:
Eldredge and Cracraft also presented the argument that most of the
misunderstanding between individuals who assert that cladograms can at
once be constructed and tested with parsimony and those who claim that
they cannot is entirely semantic. They argue that the use of the term
"falsify" can have two meanings in hypothetico-deductive science, and
that the interpretation they (and others) wished to convey is that
homoplasies do not absolutely falsify trees as hypotheses (where to
falsify is equated with to disprove, to cause the rejection of, or to
refute totally), but rather that they are used to falsify trees only to
the extent that the hypothesis is made weaker. This defense was later
recapitulated by Farris (1983) and Ax(1987). As Eldrgedge's and
Cracraft's use the term falsification in the weak sense suggests,
inferences that rely on parsimony alone are seated squarely in the realm
of inductivism, not hypothetico-deductivism. The identification of a
synapomorphy based on its transformation on a tree that was selected, in
part, because it was not a homoplasy, is clearly circular. The
inductive steps include the assumption that a tree of minimum length
provides a picture of the more general history of diversification, and
that specific conditions that require few steps applies generally to
every history of diversification.
I can't speak for E& C's current position on this - I have
encountered numerous authors who will admit that after a
decade or so, they can't agree with their past positions.
? > > > And yet you state without
reservation > > as if it were entirely agreed upon that parsimony
> > corroborates. As an inductive inference, it simply
> > cannot.
> No James, the test of congruence leads to corroboration. Parsimony is
> the principle behind the test, behind most scientific tests......am I
> repeating myself here? Parsimony is not an inductive inference;
We generalize from a few synapomorphies to the history
of the entire organisms - thai is induction.
> than actually exists in the data. It is important ot have powerful
> tests, but accepting falsification when it is not indicated is just
I'm not sure what you're referring to.
> > The difference between molecular data and morphological data still eludes
> On one level, they are not different; they are both sources of
> heritable characters, and thus neither should be ignored. This is a
> *very* important point.
> Once we accept that both are valid sources of evidence, we can note
> differences between them. Morphological structures are complex and
> they have ontogenies,
So goes the current model of morphologcial evolution.
>, In general,
morhological character evolution
> tends not to erase the previous state - a stapes is recognizable as a
> modified hyomandibula which in turn is recognizable as a modified
> branchial arch part.
I see your point for these characters, but I must call
you on the phrase "in general".
evolution tends to erase information -
> an "A" is not recognizable as a modifed "T". This difference tends to
> make sequence data far more problematical.
It depends - and the informativeness varies from
case to case - from organism to organism and from
locus to locus.
> >> Once again, and very briefly, cladistic parsimony utilizes a
> >> parsimony criterion in the
> >> implementation of the test of congruence. We do not use parsimony as
> >> a test in itself.
> > That's like saying that final exams are a test for a subject
> > course, but that the exam is not a test. ??????????
> Not really.....its like saying that the exam (test of congruence)
> will be graded (parsimony criterion), and the exam is not the same
> thing as the principle of grading.
Now you're confusing the criterion with the test.
> > I prefer "estimation"
> As well you should, given your line of work. Estimation has meaning
> in the realm of statistics; when one engages in a statistical
> analysis of sequence data, estimation is the proper term. When one
> does cladistics, which is a very different approach, then
> reconstruction is a far more appropriate term
How can you say something is a reconstruction when it
is probably wrong? It seems more forthcoming and accurate
to refer to the process called "attempts at reconstruction"
and estimate. Reconstruction alone is misleading to the
> > As we develop new and more critical tests of hypotheses of
> > homology (and there are many that have and well continue
> > to be developed), we should be moving farther along the
> > asymptote to truth.
> Well fine!, I have NO problem with that. The more the better. I have
> always thought that if the stats folks would translate thier max-like
> trees into an explicit set of conclusions regarding the homology of
> the various sites, that we might begin to speak the same language.
> Then we could begin a debate on what makes a legitmate test of
I wonder why they haven't?
> > For instance, parsimony
> > can be mislead by long branch attraction, and
> > no model of evolution is needed for that to be true (sparse
> > taxon sampling can lead to the same problem and guarantee
> > the wrong tree, and we are limited in our taxon sampling
> > via extinction),
> Sparse taxon sampling is usually one of those new problems spawned by
> generation of roving sequencers who have no committment to the study
> of a particular group. The folks who are really interested in
> organisms and their diversity have ususally gone to great lengths to
> make sure they include every taxon available.
Lineage extinction alone ensures that sometimes morphological
data are corrupted by long edges - the morphologists'
best estimates at understanding homology can thus also
> And although long branches can conceivably decieve parsimony in some
> cases, it need not do so. Sister taxa can attract each other as well.
Yes they can - but what do we want to base our understanding
of evolution on - signal, or noise?
> Furthermore, long branches can equally decieve every other method.
They can - but if the long edges can be identified, then
other methods can deal with them explicitly (some only
with molecular data). Neighbor joining, for instance,
has shown a tendancy to be less diverted.
> The fear of long branches has already spawned some ridiculous claims;
> sister-group relationships are being denied because of the mere
> possibility of long branches - even when the relationships are
> heavily corroborated by many different character systems including
> complex morphologies.
It's good to hear that this problem, now two decades on,
is being taken seriously in your world.
> And of course, the existence of conditions which might lead to a long
> branch situation can hardly indicate a preference for another tree
I'm not sure what you are saying. When long edges exist,
we are guaranteed to have the wrong tree...
> > new tests for the identification of
> > long edge taxa will prevent errors that would otherwise
> > have been made.
> just be careful that you can isolate real instances
done - for both mol and morph data.
> > No philosophical justification of cladistic parsimony
> > will ever change the informativeness of a single data set -
> > that is predetermined by whatever processes of evolution have
> > already occured to the characters, to the biologcial prowess
> > of the investiagtor, and to a large extent by
> > taxon sampling.
> So, to run throught your list backward: sample your taxa thoroughly
> (this is not an argument you will probably have to make to cladists);
> study biology (this is not an argument you will probably have to make
> to cladists); and dont restrict yourself to single character systems,
> to avoid cases where some evolutionary process has destroyed
> information (this is not an argument you will probably have to make
> to cladists).
The latter point seems to be lost on some.
> > More to the point of the thread, I consider it positivistic
> > (read: like positivism) to presume that parsimony (or any other
> > method of inference) will lead asymptotically to the truth when it
> > has been known for so long that it can be positively mislead under
> > some circumstances, and that adding more data will only
> > make things worse. To continue the myth of the sufficiency of
> > parsimony is to say that it cannot be improved.
> The improvements in our field will not come from abandoning the
> preference for the result which represents the logical ordering of
> those hypotheses which have survived critical testing.
> Your comments here seem to contradict what you said a few paragraphs
> above. You claimed there that we would move along the asymptote to
> truth by further testing homology hypotheses. That is perfectly
> consistent with parsimony - for parsimony is (one last time) the
> logical ordering of tested homology hypotheses. Test 'em more,,,sure,
> but eventually they will be combined in a matrix and submitted to a
> congruence test to form the overall phylogenetic hypothesis.
Don't you see the difference between hypothesis testing
and hypothesis formulation?
More information about the Taxacom