[Annelida] re: eunicid phylogeny

Kirk Fitzhugh kfitzhugh at nhm.org
Mon Mar 27 22:51:34 EST 2006


All,

Regarding Torsten’s claim that all of logic and epistemology is metaphysical, if we take this reasoning to its ‘logical’ conclusion, then all of science is metaphysical, and all of reasoning is metaphysical, which leaves his argument that I am relying on the metaphysical entirely vacuous.  On the contrary, science operates within the combined guidelines of logic, epistemology, and metaphysics.  I provided earlier a definition of metaphysics that shows that my arguments are not ‘purely metaphysical.’  The matter of phylogenetic inference deals more with the interplay between logic and epistemology - the interplay between causal theories and observed effects.  Let’s move on.

Torsten states, ‘The methods we used and developed were neither scientifically unacceptable nor the results meaningless or not rational explainable. In contrast, they are well proven [sic] by empirical data.’  Let’s use an example to address this statement.  A person partitions their data into two sets, 1 and 2.  From 1, they infer hypothesis (A(B(C(DE)))); from 2 they infer (A(D(C(BE)))).  Each of these explanatory hypotheses were inferred to answer specific why questions.  Yet, the two hypotheses are contradictory, they make claims regarding past causal events that differ from one another.  Is it then rational to accept either (A(B(C(DE)))) or (A(D(C(BE))))?  What do we mean by ‘rational?’  A statement, hypothesis, etc., is rational when it has been provided on the basis of all relevant evidence available to the individual making the inference.  If one knowingly denies evidence from playing a part in their inferences, then we should not regard their action or conclusion as wholly rational.  What do we mean by relevant evidence?  Evidence is relevant to any non-deductive inference if it has the ability to positively or negatively support a hypothesis (sensu Carnap).  By partitioning data, we deny some relevant evidence to play a part in answering our why questions.  It would be less than rational to accept either (A(B(C(DE)))) or (A(D(C(BE)))) since neither hypothesis is based on all relevant evidence.  Given the nature of the data in 1 and 2, both sets are relevant to each other.  Similarly, it would be less than rational to simply treat these hypothesis by way of a consensus tree.  And just as erroneous, it would be meaningless to compare these hypotheses.  Two explanatory hypotheses that causally account for different sets of effects have absolutely no basis for comparison, and such comparisons do not provide any basis for recursively deciding irrelevance.  The problem is that the two hypotheses make contradictory causal claims - the conflict arises because the two hypotheses provide different causal accounts for the same group of organisms.  Given that the same causal theory is used for each data set, both sets of data have relevance to each other, in which case they all require explanation in the same inference.

So, when partitioned data are given separate explanatory accounts, it is meaningless to compare those results, it is less than rational to accept either hypothesis, and as a result, the hypotheses are scientifically unacceptable.  Can the example just presented be ‘well proven by empirical data,’ such that this violation of the requirement of total evidence can be saved?  No.  Separate explanatory hypotheses only have relevance to the data for which they provide explanations.  (A(B(C(DE)))) and (A(D(C(BE)))) are entirely irrelevant to each other - they are consequences of mistakes in reasoning by violating the requirement of total evidence.  Is what I have just presented just a lot of ‘metaphysics?’  Of course not.

Torsten states, ‘[Kirk] recognized the problem of paralogy or horizontal gene transfer, but did not propose anyway to solve the problem if he does not want to partition the data set. These are true problems with real data and this shows the problem when the foundation is not connected to the physical world.’  As the above example shows, partitioning data does not offer one any insight into what data are irrelevant.  All that one accomplishes with partitioned analyses is to produce separate explanatory hypotheses.  The easiest way to determine that data are relevant in phylogenetic inference is by way of the fact that one is attempting to explain shared similarities by way of descent with modification.  If the same causal theory is being applied, then clearly one is assuming those data are relevant to one another.

Torsten states, ‘To partition problems/data is what we do to get specific answers (it is called reductionism). Later the parts are put back together to get further insights. And it has to be pointed out that these parts are not irrelevant of each other.’  Let’s go back to the examples used earlier, where data set 1 produces (A(B(C(DE)))), and data set 2 produces (A(D(C(BE)))).  When we combine these data, we get yet another hypothesis, (A(C(D(BE)))).  In what way do we acquire ‘insight’ into the explanation of our observations by these three sets of inferences?  How can one compare explanation (A(B(C(DE)))) against (A(C(D(BE)))), or (A(D(C(BE)))) against (A(C(D(BE)))), or (A(B(C(DE)))) against (A(D(C(BE))))?  No rational comparisons can be made that would be empirically meaningful since these different explanatory hypotheses are not comparable.  They are the products of entirely different sets of premises.  This is not a matter of reductionism.  This is just irrational.

Torsten states, ‘Interestingly, [Kirk] does not directly address the problems of circularity and contradiction except for pointing out that effects and evidence are not not the same. Something I did not state. Furthermore, he does not put forward any specifics how to perform such tests if "Character data cannot test those two classes of events [character origin and speciation]".  I did not specifically address the matter of circularity in my previous reply since such circularity does not exist, and I did in fact state how to perform such tests.  Here is what I said: ‘The bottom line is that character data cannot be used to test any phylogenetic hypothesis since those data cannot be deduced as potential tests....  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.’  There is nothing circular here.  From an explanatory hypothesis one must deduce potential test consequences that would be the case if the hypothesis were true.  Such test evidence cannot be in the form of more shared characters since such additional characters are not deductive consequences of the hypothesis.  The causal claims in the hypothesis do not apply to other characters.  As I pointed out, what one must seek are effects directly related to the past events that caused character origin and speciation.  These latter effects are deductive consequences of the hypothesis.  These are effects that are independent of the properties of organisms in the form of shared similarities.  So, I did in fact address the problems Torsten raised, and showed that the problems do not exist.

Torsten states, ‘To test my hypotheses I need some kind of physical data and this will be new or old character data in anyway. A short remark, if the abstract (the summary of the main conclusions of a paper) already shows circularity and contradiction, what can we expect from the whole paper?’  Once again, character data cannot test any phylogenetic hypothesis since new character data cannot be deduced from a phylogenetic hypothesis.  As a phylogenetic hypothesis presents statements regarding character origin and speciation, it is only events associated with those causal conditions that can be deduced as potential tests, not characters.  So, the charge of circularity and contradiction is false.  What I have presented in the Zootaxa and Biosystema papers is the standard approach to testing explanatory hypotheses, which corrects the long-standing myth that character data can test such hypotheses.

Torsten states, ‘In his Zootaxa article Kirk states"The consequence is that parsimony has logical priority over likelihood in abduction,...". Thus, he advocates parsimony over likelihood contrary to his statement in the last e-mail.’  That parsimony has logical priority over likelihood in abductive inference does not translate into advocating parsimony over likelihood!  If you go back and look at what I wrote, you will notice that both parsimony and likelihood hold relations within any abductive inference.  So, it is impossible to advocate one over the other.  The logical priority of parsimony is indicated by the fact that it is a relation between one’s why questions and the hypothesis or hypotheses answering those questions.  Likelihood operates within the inference itself by way of the relations between premises and hypotheses, in the form of support.  So in fact, a most parsimonious hypothesis will also be the one with greatest likelihood since we want hypotheses that are most effective at answering our questions regarding what we observe.  This is simply a manifestation of abductive inference.  If we deny the truth of our observations, by denying parsimony a role in our inferences, then we not only compromise inferring hypotheses with greatest likelihood (most support) but also the very basis for scientific inquiry.  Parsimony and likelihood cannot be separated in the manner that has been claimed by advocates of 'likelihood' or 'parsimony.'  The schism that has been imposed is a false schism.  Unfortunately, the relations of parsimony and likelihood to abduction have not been discussed extensively enough.

Torsten states, ‘A myth would be Kirk's proper testing if it is not connected to any kind of physical data. A mythical explanation invokes a new layer of explanation for an observed phenomena without any connection to physical data.’  Huh????  How can testing ever be accomplished if not connected to data?  I never claimed the kind of myth Torsten associates with me.  I made very clear the nature of the empirical evidence that is required to correctly test explanatory hypotheses.

My thanks to Torsten for his comments.  I hope this exchange will prove useful to members of the list.

Kirk

-----------------------------------------------------
J. Kirk Fitzhugh, Ph.D.
Curator of Polychaetes
Invertebrate Zoology Section
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
http://www.nhm.org/research/annelida/staff.html
http://www.nhm.org/research/annelida/index.html
----------------------------------------------------

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.bio.net/bionet/mm/annelida/attachments/20060327/716e1c3e/attachment.html


More information about the Annelida mailing list