Character Polarity (from Molecular taxonomy: on way out?)

John Grehan jgrehan at SCIENCEBUFF.ORG
Wed Jul 20 13:35:31 CDT 2005

> Behalf Of Derek Sikes
> This Nixon & Carpenter paper is certainly cited as the 'key' paper
> justifying 1) not polarizing before analysis, 2) searching on
> unrooted trees, 3) rooting after the shortest tree is found which
> subsequently provides directionality, aka polarity, to the evolution
> of (many of) the characters.

Their paper is less a justification than an invocation of a particular
point of view - a view which does come into conflict with other
practitioner that they cite. For example, they have a section on
'polarity myths' in which they dismiss any view to the contrary. They
point out that it is in conflict with the procedure described by Farris
[wasn't he a pheneticist?]and they also say that the need for the sister
group is also a myth, but just cite a bunch of people asserting that
belief. They take what seems to be a snide shot at Brooks and McLennan
who "write introductory materials for young minds" [I picked up on this
as I am well aware that there is considerable animosity expressed by
some cladists towards others]. Nixon and Carpenter reduced outgroups and
polarity to the problem of rooting. I would disagree and assert that the
problem is being able to justify polarity of each character chosen -
there should be a defensible argument for each character by which the
different states show evidence of being connected. I doubt this will
really work for DNA sequences since there is no transformation of
characters, just a back and forth of any one sequence between two

I would like
> to see a paper that takes, say, a dozen or more published, empirical
> datasets that have been analyzed in the 'old style' of 'a priori'
> polarization and that reanalyzes them in the new, unrooted, Nixon &
> Carpenter method, with rooting providing polarization after the fact.

I would suggest a slightly different comparison - an 'old style' a
priori polarization and a 'new style' unpolarized character set. At
issue here, for me at least, is whether the initial character set should
be limited to derived character states for the ingroup.

> 1) if the hand-polarization was done in strict accordance to the
> rules used for outgroup polarization and combined with thorough
> application of an optimality criterion like parsimony, the results
> should be the same

I would expect the same result too. But whether polarized and
non-polarized data sets would necessarily give the same result would be
interesting (the non-polarized data set would have a whole lot of
characters missing in the polarized set).
> I hope such a paper exists. Nixon & Carpenter don't really
> demonstrate the difference using empirical datasets, but I'm not
> aware of such a paper. 

My reading of it a while back is that they just claimed the algorithm
would identify the proper derived states so that it did not matter what
went into the set.
> It seems that Grehan is not really that worried about his dataset
> having characters that are plesiomorphic for his ingroup as long as
> they are coded such that they CANNOT be mistaken for apomorphies -

Actually I would not include the plesiomorphic characters for the
ingroup analysis at all. I would only include derived states and resolve
any conflicting patterns of relationship.

> he's worried about a method that allows these characters states to be
> mistaken for apomorphies of his ingroup,

No, no, no. What I am concerned with approaches that include
plesiomorphic characters in the data set without identified as such
before the analysis.

> and repeating, and repeating that, in his opinion, the best way (only
> way?) is to eliminate these characters before analysis to prevent
> this mistake from happening.

I'm a broken record (or should is be CD? My ancient (pleiosmorphic?)
state is showing.

> What he fails to realize is that, especially for large complex
> datasets, and this was driven home by Wiley '81 and Paterson '82 that
> looking at characters in isolation can only get one so far- doing so
> provides only *hypotheses* of homology and if one chooses, polarity.

Of course I have never said that characters should be looked at
(analyzed) in isolation! I have said head character selected for the
data set should be individually justified as having a derived state for
the ingroup so they can be analyzed.

> These hypotheses are tested via congruence with other characters -
> the more additional characters, the better the the test. 

Yes of course. That's the whole basis of the orangutan argument (not
that it makes any difference as everyone has faith that DNA sequences
cannot be wrong while morphology is something the devil cooked up).

If one
> eliminates all possible homoplasy or characters that map such that
> one's favorite characters' polarities are reversed, one weakens the
> tests and biases the results towards what one imagined they should
> be. This is, of course, poor scientific practice but strongly and
> repeatedly advocated by Grehan.

I think Derek is running into the confusion that inevitably develops in
email discussions. If one is looking at the generic relationships of the
great apes and humans there will be four taxa and any possible
combinations of relationship suggested by different sets of derived
characters. The choice is not made by eliminating any of these sets, but
seeing which one provides the most congruent result according to
whatever clustering-tree drawing method one prefers.

> I expect John will say unrooted trees are not "cladistic", and, as
> usual, he will be wrong. He will also probably repeat something about
> how important it is to limit one's dataset to characters whose states
> are only derived for the ingroup (not understanding that one's
> ingroup could be someone else's outgroup!).

Oh I do understand that. And perhaps unrooted trees are cladistic (I
admit to not having analyzed the technique), but I do not see that
approach as solving the basic problem of failing to limit characters to
derived states. I am not worried about polarization mistakes - only that
algorithms cannot decide other than as an artifact of their treating
character coding in a certain way.

John Grehan

More information about the Taxacom mailing list