On realism, more

Richard Zander bryo at COMMTECH.NET
Fri Dec 12 13:20:21 CST 1997

Tom DiBenedetto wrote:
> Richard Zander wrote:
> >A realist believes in a reality based on all the facts at
> >one's disposal, then changes the concept when more facts appear and make
> >it necessary.
> No that is not my understanding of realism. I consider myself a
> realist. I do not believe that reality is based on facts. A fact is
> simply an idea which is deemed to be so well corroborated that it is
> accepted without further investigation (unless new phenomena
> impress upon us a reason to reopen the issue).
> Rather, I believe facts are based on reality, or our perceptions, and
> tested hypotheses concerning the reality that is accessable to us.
> >> What exactly are you denying the reality of?

I've got lots of real things around me right now. I deny nothing real.
Also I am raffling off the Brooklyn Bridge, and would you like to buy a
ticket? Actually, we agree on everything except the limits of what
"approaches" means in the phrase "the results of a well supported
cladogram appraches the true tree." Practically, I can work together
with anybody when we agree of what material things we call real and on
what ideas we call real, i.e. "as if" such were good instruments. The
product of our work is then compared with the product of other people's
work and we see who generates theories that are better at prediction.
Now Einstein's work was accepted long before its predictions (Venus
occultation, atom bomb, etc.) were tested, it being simple, elegant and
theoretically better than anything else. Given that there are two
competing theories now, cladistic and statistical, each of which is
touted as simple, elegant and theoretically better than the other, I'd
like to see some relative prediction testing.

> >... We can agree on what works,
> >pragmatically, and on what can be used (instrumentalism) be it ideas or
> >material things, to effect changes in our lives. We jointly aim at a classification of >organisms that facilitates prediction.
> Prediction of what? A parsimonious ordering of homologies has, it
> seems to me, the greatest predictive power of any ordering system.

Biogeographical relationships, medicinal properties, genes, stuff like

> >To the extent, however, that philosophical
> >realists among cladists claim to have approached truth more closely than
> >I think they have approached truth (with a view to furthering a joint
> >effort), I complain.
> As a cladist, I find my hypotheses to be believable to the extent
> that they are tested, and corroborated. I cant imagine a
> quantification of the distance I am from the truth, but I feel it is
> justified to prefer hypotheses which survive critical testing better
> than others; i.e. the most parsimonious tree.

Since parsimony and statistical studies seem to both get one tree, and
both trees seem reasonable, I think that the figures the statistical
people come up with to show confidence limits probably also apply to
cladistic results. I mean, say you're using the same data set: How can a
parsimony result converge on the truth (and therefore be a result we can
build on) and while a statistical result sometimes have such low
posterior probability? Besides, as I have argued in various threads, a
maximum synapomorphy result is not necessarily a maximum parsimony
result. This false equality has been hammered into the literature to the
extent that it is axiomatic.

> >All sentences are theory-laden. I call for an INDEPENDENT TEST OF
> >predictability of the trees obtained is compared and evaluated in such a
> >way that we can feel the work is worthwhile.
> Could you flesh this out a bit? Sounds to me like you are simply
> asking for more characters (for that is what they are).

See above. I figure we are doing two things: searching for truth, and
generating a useful classification. The first is moot, the second
doable, at least relatively. What makes a classification useful? That's
the test(s).

> >I predict that fine structure of trees from
> >both maximum parsimony and maximum likelihood analysis will not be worth
> >much as far as prediction goes.
> I think that is clearly nonsense. They already have (at least the
> parsimony trees).

Some circularity there.

> >> is adjudged to have low probability in light of some theory,,why does
> >> this indicate that it is false? Maybe the theory underlying the
> >> probability calculation is false.
> >>Maybe so. In fact, I'm surprized that both cladists and statistical
> >phylogeneticists don't shuffle their feet and look ashamed more often.
> This doesnt follow. Cladists dont calculate probabilities,
> understanding as we do, how illegitimate a frequency probability for
> unique historical data is.

Depends what "how" means. Poor resolution fine structure, sure, for now

> >>You seem to beleive that you can calculate the probability of
> >> something being TRUE. How can an anti-realist hold such a position?
> >Good retort. Answer is I think I can calculate the PROBABILITY of
> >something being true. Note emphasis.
> The emphasis wont save you. It will still be the probability of
> something being true under the assumptions of some theory.

Nuts. I though it might. On the other hand, statistical phylogeneticists
can calculate probabilities given certain assumptions. That always
refers to truth, and I figure truth in the past is similar
(uniformitarianism) to truth in the present, which generally is agreed
upon truth. Truth is, e.g. ((man chimp) dog), i.e. man and chimp are
phylogenetically grouped by our assumptive expectations about
speciation. The probability is 1, given morphology. No, you can't
calculate probabilities in cladistic analysis of relationships that are
arguable. Given certain assumptions, you can with statistical
phylogenetics, but I argue that the answers are pretty vague there, too.

> > However, a big however, calculating
> >probabilities of things that happened ONCE in the PAST could be
> >laughable
> IS laughable...
> >without the potential of molecular data from say a couple of
> >independent, selectively neutral genes.
> And just what on earth do you imagine that a couple of neutral genes
> will
> tell you? The probability that a given tree would represent the
> average phylogeny if we could run through the history of life a
> thousand times?

Jeez, I don't know. I'm no apologist for statistical phylogenetics. I'm
just suggesting that statistical phylogenetic and cladistics both
(often) claim to produce single trees that represent a thing that
happened once in the past, and confidence limits need to be established,
and assumptions spelled out.

Richard H. Zander, Buffalo Museum of Science
1020 Humboldt Pkwy, Buffalo, NY 14211 USA bryo at commtech.net

More information about the Taxacom mailing list