Positivism vs Realism
tdib at UMICH.EDU
Fri Dec 12 17:17:32 CST 1997
James Francis Lyons-Weiler wrote:
> Parsimony is a criterion, but the belief that
> the criterion will lead to meaningful estimates of the past
> requires a bolus of faith I'd rather not choke on.
A bolus of faith??? It has nothing to do with faith, and everything
to do with implementing a critical test of congruence.
> The LEAST
> positivistic position, in my view, is the one that allows that
> more than we might expect could have happened -
> and this allows
> for numerous intrusions into the data that can mislead parsimony
> in particular, and other methods as well (some to a lesser
Huh? Parsimony, in the implementation of the congruence test, is
merely testing the set of homology hypotheses against the expectation
of congruence. It is not attempting to model evolution. It is helping
us find that part of the data which can be interpreted as present by
descent. If you need ever more analogies,,think of a regression or a
principle components analysis. We draw the line which best summarizes
the data; then we deal with the residuals.These statistical tools
make use of a parsimony criterion in precisely the same manner as the
test of congruence; find the line which minimizes the residuals. Then
you develop smaller scale hypotheses (ad hocs) to account for the
> Congruence is expected sometimes, but not always -
> descent with modification only partly explains the distribution
> of states among taxa, and it does not always occur in a clean
yes james, you are not telling us anything new here. THat is the
point of the the tests we do; to distinguish those parts of the data
which need special explanation; i.e. those which cannot be explained
> (1) Popper indicated that the degree of corroboration is
> inversely related to how expected a result is. If we include
> too much "background knowledge" in the formulation of the
> hypothesis (i.e., winnow away unreasonable hypotheses of
> homology on other grounds, say, development), then the degree of
> corroboration afforded by the existence of congruence among
> the characters (individual sets of hypotheses of homology)
> is decreased, and approaced zero if we include all of the
> background knowledge that is in fact accurate and relevant
> (in which case we have a tautology).
as in model-based systems, which explicitly seek to maximize
> Bold hypotheses receive
> a higher dose of corroboration because they are not expected
> to survive a truly critical test; i.e., they represent
> hypotheses which may run counter to expectation.
Once again, you make this almost sound like a psychological
criterion, how much we are surprised by a result, because you fail to
understand what Popper was driving at with his reference to
expectation and surprise. THese are a function of information
content. You can achieve a high degree of corroboration when a
hypothesis is unexpected RELATIVE TO BACKGROUND KNOWLEDGE, i.e. when
the hypothesis has high information content relative to the
underlying assumptions. It doesnt matter a hoot what you personally
> In the case
> of cladistics, if we have done our biology right, our set
> of hypotheses of homology is expected to be congruent. This
> expectation diminishes the level of corroboration for those that
> are found to be congruent RELATIVE TO the level that which should
> be achieved under a truly critical test.
Lets parse James's logic here.
1. We make a hypothesis of homology.
2. It inherintly brings with it an expectation of congruence.
3. We test our hypothesis for congruence; it passes the test.
4. Therefore, since we hoped/expected it to pass, the test is not
really a test. and the hypothesis is not corroborated (or only
extremely weakly corroborated).
Now this wonderful logic can be generalized.
1. We propose a hypothesis.
2. We deduce a set of expectations/predictions from the hypothesis\
3. We test the hypothesis against the real world (observe, experiment
4. Even if it passes the test, it realy isnt corroborated, because we
hoped/expected that it would pass the test, therefore we cant really
pretend we are surprised.
Now the end of this train of logic is, of course
5. since we expect the hypothesis to pass, our surprise might be
maximized if in fact it fails the test,,,so in James's world, perhaps
a hypothesis is maximally corroborated when it is falsified!
OK,,,enough fun,,,,lets see what is missing here.
As I said above,,,the "amount of surprise" that Popper was referring
to was not a personal psychological state, it was a casual way of
making refernce to the relationship bvetween the hypothesis and
Congruence reprents a minimal level of background knoweldge; it is a
direct deduction from the notion of life as having evolved a single
time through a process of descent with modification. James is correct
that even this amount of background knowedge reduces the level of
corroboration, and he seems to demand we abandon even that in order
to have a "truly critical test". But by abandoning congruence we are
left with nothing. THe irony here is that all other methods seem
determined to load up on background knowledge,,,whether it be by
character weighting, or, at an extreme, by accepting a model of
evolution as background knoweldge.
This is why the statistical methods can be seen as engaged in a
retreat from the scientific method - by building ever more factors
into the background knowledge of their analysis, they end up saying
nothing - basically simply deducing the phylogenetic consequences of
the model, with no testing involved whatsoever.
This may have an important place in the research program of
attempting to model genetic evolution, if they then took this
phylogeny and tested it (e.g. against the best reconstruction
available from all of the evidence),,but taken by itself it is
nothing but the phylogeny which would have been true if model X is
true - something we know it aint.
> When hypotheses
> of homology are revised to account for those found to be
> discordant, either by calling them homoplasy or by
> re-examining the material, then Popperian corroboration is
> conveniently dismissed -
huh? When something is labeled homoplasy and we are forced to accept
an ad hod explanation for its presence in the data, we are
acknowledging that it has failed to be corroborated by out test - it
is a falsification. How can this be seen as dismissing Popperian
> i.e., it is applied to the
> hypotheses of congruence homologies, but it is not allocated
> to the emergent hypotheses of discordant homologies, and
> yet they represent the archetype of bold hypotheses because
> they were unexpected.
An ad hoc hypothesis is the archtype of bold hypotheis......buddy you
are getting lost in space now! And it all comes form not
understanding the role of expectations in Popper's notions. Ad hoc
hypotheses are the least bold, by definiton, for they are special
hypotheses which account for particular cases, they say nothing
beyond the particular case, and have minimal scientific value.
> (2) The first position views parsimony as an inductive
> step in hypothesis formulation - which is fine.
> The position that
> parsimony is a test belies the fact that most consider
> published trees to be hypotheses in a long series of
> attempts to model the past - not as theories or
> truth, but mere hypotheses.
Huh? How does this sentence hold together? First off, I told you
several times that parsimony is a criterion, congruence is the test.
And I never met a cladist who needed to be convinced that cladograms
are hypotheses, they are corroborated hypotheses; everyone has always
> To expect otherwise is
> positivistic. Some of the greatest contributors to phylogenetics
> have become skeptical of the utility of the trees that get
> published - Swofford and Maddison come to mind.
Swofford is one of the......what??????? (you lookin' for a job
And he has become skeptical of the utility of the trees that get
My god,,,,just imagine what the poor boy will be faced with once
everyone starts using PAUP* !!
> how could any
> scientists' faith in parsimony possibly effect whether
> the data they have collected are informative or misleading?
But since it has nothing to do with faith,,,,,,,,
> This would not be problematic if parsimony did not provide
> a set of optimal trees for ANY data set with variable
> character states.
Which is saying only that any data set can have some amount of
congruence. So what? We are testing specific data sets for the
congruent pattern that they have.
> Tom's opinion of parsimony is
> independent of whether or not, in a particular case, parsimony
> tests and provides corroboration, or has merely summarized the
> set of best hypotheses of homology in a congruent manner, which
> even then mau remain misleading.
My opinion of parsimony is based on the fact that it does just that.
It efficiently implements a congruence test. The congruence test
yields the hierarchical ordering of homologies which can be explained
by descent, and also indicates those which cannot.
> The focus
> should be on the evidence present in the data, and not what the
> data look like after they have been structured on a tree.
I dont understand this. You think you can detect (e.g) convergence
without a tree?
If you can,,,,,and you can do it convincingly, then I wont propose it
as a homology. Such concerns are addressed in the character defintion
phase. Whats the problem?
More information about the Taxacom