threads on patterns

[Kirk Fitzhugh]

Ken,

I sympathize with your reticence regarding philosophical/semantic
discussions. I'm on another list-serve that deals with pragmatics,
knowledge formation, and inference, and I often find myself wondering about
the relevance of what is said!

Unfortunately, or fortunately, philosophy of science must be at the
forefront of cladistics. I say this because it is only by way of addressing
the philosophical basis of a method that a group of individuals have the
opportunity to seriously and critically examine the nuances of a method and
make rational decisions as to what methodological aspects to keep and what
to dismiss. Cladistics has for a long time continued to remain perched atop
many different summits of thought. The disagreements continue, and will
continue, because discussions tend to take place within the context of
one's interests rather than stepping outside of that context and examining
methods under the broader umbrella of conceptual issues that dictate the
process of all scientific thought. For instance, arguments related to
cladistic "parsimony," inferential support for hypotheses, maximum
likelihood, total evidence, etc., can only be constructively addressed when
directly related to the philosophical underpinnings of each. Neither
parsimony, nor inferential support, nor maximum likelihood, nor total
evidence are issues that are philosophically disagreeable. But, to properly
apply the concepts, we must fall back onto the philosophical foundations of
each. A good example where this has not happened is with total evidence.
For all the talk for or against the requirement, I've not seen a cladistic
paper yet that correctly conveys the specifics of the requirement. It then
seems odd that some have even gone so far as to claim that we can
circumvent the requirement by way of certain "tests," when in fact there
are no philosophical treatments that would allow for such circumvention. To
use one of your statements as another example, "I have nothing
against...parsimony in moderation." This is sort of like saying, "I have
nothing against inference in moderation." I don't mean to sound factitious,
but the very loose criticisms that revolve around the topic of "parsimony"
have been so out of line with what the "criterion" of parsimony actually
refers to as to make the criticisms useless.

On the topic of "pattern," I really like your water stain example because
it points to how in cladistics we often misuse language. In your example,
you say, "Most of us might see a water stain on a wall and wonder what
might have caused it, but a few might discern a pattern in the stain of the
Virgin Mary crying and regard it as a miracle." If we agree there's a water
stain on the wall, then we have already provided an explanation for what we
observe, which is a discoloration on the wall. The pattern observed is the
discoloration relative to the rest of the wall, not the water stain.
Invoking the Virgin is just another explanation of the pattern. In the same
manner, cladistic analyses don't come up with patterns, they provide
explanations of patterns. The explanations might be real or not. At the
point of cranking out trees, what is real and non-real is entirely
irrelevant since the inference from data to trees is non-deductive, so
automatically even the "best" tree cannot even be considered remotely true.
In the same way, there are no "right" or "wrong" trees - there are only the
universe of all possible hypotheses, among which some are better confirmed
(non-deductively) than others. In the (correct) sense of confirmation, a
tree with length 100 is better confirmed than a tree with length 200 given
the same data (i.e., premises). In a sense, all the trees are worthless
because they are only vague descriptors of some of the causal events that
might have occurred to give us the patterns (= shared similarities) we observe.

So, there are no vulnerabilities to cladistics in this regard. But, this
does get us back to the philosophical basis of using the parsimony
criterion in cladistics. I think I mentioned in an earlier message this
week that the use of parsimony is part of the inferential act of maximizing
the explanation of shared similarities by way of common ancestry (the
inverse of minimizing homoplasy - which is nothing more than the
explanation of shared similarities by means other than common ancestry).
When one uses parsimony, to a maximum or lesser extent, it really does not
matter, all this means is that they have attempted to apply some causal
theory as much as possible to the explanation of shared similarities.
Indeed, it is incorrect to say there are "users of parsimony" and "users of
maximum likelihood." Both groups apply the criterion of parsimony, but by
way of using different causal theories (which is why a minimum-length tree
is a maximum likelihood estimate per some causal theory). The debate is not
whether to use "parsimony" or "maximum likelihood," but why use common
ancestry as opposed to a rate-based approach.

"What makes me nervous are unwarranted attempts to 'maximize' parsimony
(and oversimplify)." I hope I've been able to show that what is actually
unwarranted is the avoidance of using maximum parsimony! When one accepts a
tree of greater length, they are implicitly admitting that they don't
accept the causal theory they choose to use. There are no rules that
preclude one from doing this. This is not a point where logic has any
bearing. The choices one makes are purely within degrees of rationality. To
prefer a tree longer than the minimum length is to act in a
less-than-rational way.

Best,

Kirk