[Annelida] re: eunicid phylogeny

J. Kirk Fitzhugh kfitzhug at nhm.org
Thu Mar 23 16:08:45 EST 2006

The American writer Mark Twain once said, "Whenever you find you are on the 
side of the majority, it is time to pause and reflect."

I am glad to see Torsten's thoughtful comments and his willingness to 
pursue this discussion.  My comments will address the scientific merits of 
his reply, not his or Ken Halanych's accusations against my character.  To 
state that a method is scientifically unacceptable is neither rude nor 
insulting, given that I was speaking of methods, not individuals.  If I 
wanted impugn the credibility of anyone, then I would have structured my 
message quite differently.  What I pointed out is that certain methods are 
either incorrectly applied in phylogenetics, or lead to results that are 
meaningless or cannot be rationally interpreted.

On the matter of partitioned analysis, Torsten cites Lecointre and 
Deleporte (2005).  This is a good paper in that it discusses the 
requirement of total evidence (RTE) to some extent, as well as 
abduction.  But, Lecointre and Deleporte neither adequately discuss 
abductive inference or the RTE.  As such, one cannot rely solely on the 
systematics literature for a defense of partitioning.  Lecointre and 
Deleporte's understanding of the RTE is not entirely correct, especially 
with regard to the nature of positive or negative evidential relevance, 
which is a critical factor in correctly considering the merits of 
partitioning.  I specifically address their claim that partitioned analyses 
are necessary in my forthcoming paper in Biology & Philosophy:

'The present discussion calls into question the position taken by Lecointre 
and Deleporte (2005: 106, fig. 5; see also Grande 1994), that "Performing a 
simultaneous analysis right from the beginning is of little methodological 
value."  Rather, they contend that "points of disagreement between trees 
are either artifactual or due to a process of discord..." (Lecointre and 
Deleporte 2005: 107), indicating that some character data must be 
excluded.  The matter of deciding what sets of character evidence are 
abductively relevant to one another requires a three-step process according 
to Lecointre and Deleporte: (1) infer phylogenetic hypotheses for 
partitioned sets of data; (2) determine the extent to which character 
incongruence exists among those hypotheses; (3) if, according to the 
incongruence length difference test of Farris et al. (1995), hypotheses 
exhibit significant character incongruence, then "misleading" data are to 
be removed prior to the final objective, which is the inference of 
hypotheses from the total "relevant" data.  Contrary to what Lecointre and 
Deleporte (2005) state, the "misleading" data that are removed must be 
interpreted as "irrelevant" sensu Carnap (1950), rather than relevant.  The 
more significant problem with their approach is that separate explanatory 
hypotheses do not have the capacity to determine what sets of observations 
are irrelevant to the explanation of other observations.  Recall as well 
that relevance can be positive or negative, such that contradictory 
hypotheses inferred from the same causal theory... indicate that 
observations are negatively relevant.  As was noted for the inferences 
[presented earlier in the paper], and will be discussed in a later section 
(see The Myth of "Cladogram" Comparisons), applying the same causal theory 
to partitioned data sets for the same organisms results in hypotheses that 
cannot be rationally compared, as each of these hypotheses stands as an 
explanatory account that only pertains to the effects from which the 
hypotheses were inferred.  The fact that these hypotheses can contradict 
one another cannot be interpreted as "artifact" or "discord."  The best 
that can be claimed for such contradictions is, as mentioned above, that 
all data that are to be explained using the same theory are relevant 
(positively or negatively) to one another.  The matter of relevance must be 
given due consideration prior to inferring an explanatory hypothesis, 
rather than treating sets of hypotheses as the arbiters of rational 
reasoning.  Indeed, the matter of relevance is to be considered as part of 
the formulation of the causal questions and selection of applicable theory, 
and not within the inferential procedures that are employed after these 
tasks have been accomplished.'

In order to perform partitioned analyses, one must assume that the data are 
irrelevant to each other.  But, if data are irrelevant, then one could not 
causally account for both sets of data using the same causal theory, for 
this would violate the condition of relevance.  As the matter of relevance 
or irrelevance is something that must be decided prior to abductive 
inference, it is not possible to compare separate hypotheses from 
partitioned data and decide after the fact that those data are relevant, 
positively or negatively, or irrelevant.  Unlike what we see in 
systematics, the RTE is not a controversial issue as it is one of the 
central rules for rational, non-deductive reasoning.  If one wishes to 
defend partitioned analyses then they will have to show that the RTE does 
not hold.  Lecointre and Deleporte (2005), and other systematists, have 
never been able to successfully make this claim.  I agree with Torsten that 
one has to contend with matters such as paralogy or horizontal gene 
transfer.  The problem, however, is that explanatory hypotheses do not 
provide the basis for deciding what effects are in need of being explained 
by those hypotheses.  The criterion for such a decision is relevance, and 
this is considered prior to an abductive inference.

I will briefly dispel Torsten's misconception of the term 'fact,' 
especially his claim that 'an opinion is not a fact.'  Here is a definition 
provided by Mahner & Bunge (1997: 34, Foundations of Biophilosophy): '...a 
fact is either the being of a thing in a given state, or an event occurring 
in a thing.'  So, if I state, 'Partitioned analysis is erroneous,' this 
statement conveys a fact, that part of my state of being is to hold a 
certain opinion about partitioned analysis.  In other words, it is a fact 
that I hold that opinion.  Or more specifically, it is a fact that certain 
neurons in my brain are excited in a specific order such that I utter the 
words, 'Partitioned analysis is irrational.'  Just like stated opinions, 
observation statements are reflections of mental constructs that convey 
what one believes to be facts.

Torsten then claims that my 'foundation for phylogenetics [is] purely in 
the logical realm.'  No, incorrect.  All fields of science operate under 
the more general conditions of imposed by logic, epistemology, and 
philosophy of science, to name just a few.  What I have attempted to do is 
to investigate the philosophical foundations of phylogenetic systematics 
within the conditions stipulated by these broader realms.  We need to fully 
evaluate the nature of abductive inference as it applies to 
phylogenetics.  We have to evaluate the veracity of our methods against 
what we claim to be our goal in phylogenetics, as well as the goal in all 
sciences.  Torsten's comparison of my evaluation of phylogenetics as 'like 
building a house on water' only indicates to me that he does not comprehend 
the importance of philosophy of science to the critical evaluation of 
phylogenetic methods.

Torsten then addresses the topic of testing based only on the abstract of 
my Zootaxa paper.  He seems to confuse the testing of hypotheses by 
evidence as opposed to the inference of hypotheses by certain effects. The 
bottom line is that character data cannot be used to test any phylogenetic 
hypothesis since those data cannot be deduced as potential tests.  One 
cannot test a hypothesis inferred to explain the very effects used to test 
that hypothesis.  As phylogenetic hypotheses are causal claims, what must 
be tested are the statements regarding the specific events of character 
origin and speciation.  Character data cannot test those two classes of 
events.  Rather, the necessary test evidence must be of a form directly 
related to the events themselves that provide evidence that those specific 
events did occur.  Torsten fails to make the necessary distinction between 
effects used to infer a hypothesis and evidence used to test that 
hypothesis. This is not a matter of 'arguing purely in the realm of 
logic.'  The procedures for testing are clearly outlined in the philosophy 
of science literature, which again I cite in my paper.

And finally, Torsten repeats the mantra, likelihood is superior to 
parsimony, etc., etc.  For some reason he once again did not reply directly 
to the specifics of what I said earlier about those concepts.  In fact, I 
do not even state that I advocate parsimony over likelihood!  What I 
claimed is that neither has been correctly applied to the characterization 
of abductive inference in phylogenetics.  If list members want to pursue 
the relations between likelihood and parsimony as they apply to abduction, 
that bootstrapping is applicable to statistical, not explanatory 
hypotheses, and other amusing notions, then please take a look at my 
paper.  And, I encourage more discussion!


J. Kirk Fitzhugh, Ph.D.
Curator of Polychaetes
Research & Collections Branch
Los Angeles County Museum of Natural History
900 Exposition Blvd
Los Angeles CA 90007
Phone:   213-763-3233
FAX:     213-746-2999
e-mail:  kfitzhug at nhm.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.bio.net/bionet/mm/annelida/attachments/20060323/399ad3b0/attachment-0001.html

More information about the Annelida mailing list