A "new" phylogeny & methodology

["Jerry D. Harris" <dinogami@hotmail.com>]

> Is it just me, or do other people find this endless hunt for ever more
> "characters" just plain CRAZY? The computer will, of course, grind out 
> trees
> as long as you give it "characters" to chew on, but do you really BELIEVE 
> in
> the resulting phylogenies, or are they just so much nonsense, to be 
> discarded
> when the next batch of computer-generated trees, crunching even more
> "characters," emerges? If so, why not hold out until you've found ALL
> possible "characters," grind out the one big tree and be done with it?

  As I see it (perhaps incorrectly), one could imagine a theoretical matrix 
of thousands, perhaps tens of thousands, of characters, resulting from an 
incredibly detailed analysis of thousands of specimens.  Would this tell us 
anything useful?  My guess would be: no, not really -- it would be loaded 
with "noise."  Using such an analysis, one could, for example, cite every 
percent grade in the length of a limb bone as a separate state -- something 
like "Ratio of femur/metatarsus length 0 = 50%, 1 = 49%, 2 = 48%...", etc., 
but surely this kind of thing is well within the range of individual 
variation and is statistical noise.  And that is the heart of the problem of 
"too many characters" -- once you have populations of specimens large enough 
and if you peer at them long enough, you're going to start seeing 
differences that are due solely to individual variation (or, on a slightly 
larger scale, clinal variation), but a cladistic-type analysis can't tell 
you that; it will still produce a branching graphic placing all the 
specimens into some kind of order when, in reality, no such order exists.  
(Although I can see why it would be a useful tool, this is the biggest 
danger inherent in doing phylogenetic analyses with individual specimens 
instead of species or other taxa.)

  At the other end of the spectrum, if one is too general with one's 
characters (or has too few), then the analysis can't make any sense out of 
real divergence patterns at all.  I have some problems with analyses that 
say things like "Prezygapophyses on caudal vertebrae 0 = short; 1 = long."  
This doesn't tell anyone anything -- what is short??? what is long???  These 
terms need very explicit explanations (short, for example, may mean "<50% 
length of centrum" or "does not extend beyond cranial end of centrum."  If 
they're ambiguous, then they aren't scientific because they can't be 
reanalyzed and retested by other scientists -- they can't read the mind of 
the original author to determine what, in his/her mind, "long" and "short" 
are, and as we are all aware, one person's "short" is another person's 
"long!" An analysis with characters thus defined, and with too few 
characters, is far more likely to result in erroneous and/or polytomic trees 
instead of something more sensible.

  Lastly, one must be quite careful in separating character states -- not 
functional decouplings (which is a whole 'nother ball o' wax...), but in 
having what are better stated as single, multi-state characters distributed 
throughout a matrix as separate, bi-modal characters.  For example, some 
analyses of the past have cited the astragalar ascending process:tibia 
length ratio frequently -- one character would be if the process is 0 = < 
1/6 the length of the tibi; 1 = > 1/6, and later we would see the same 
character state again, but with 0 = < 1/4 the length of the tibia; 1 = > 
1/4.  This is much more simply stated as a single character with states: 0 = 
< 1/6; 1 = 1/6-1/4; 2 = > 1/4.  (Of course, putting in too many state 
possibilities gets back to the scenario outlined above!)  This reduces the 
overall number of characters in an analysis, but retains all the 
information, and makes it easier to visualilze the evolutionary 
possibilities under consideration.  I believe that separating these out into 
separate characters in the past was done so that they could be discussed in 
the appropriate portion of a paper along with the node or stem for which it 
is diagnostic, but that same information can be read fairly easily off the 
data matrix and really isn't necessary.

  The problem here, of course, is that there are no hard-and-fast rules 
about what constitutes "too few" and what constitutes "too many" characters 
-- how do we know if we've got too few to perceive reality, and how do we 
know if we've got so many that reality is drowned out by the noise?  More 
isn't necessarily better here (although the number of characters is 
certainly going to rise as more taxa are added -- an analysis of all members 
of the Sauropsida probably ought to have more characters than an analysis of 
the Alligatorinae!) As far as I can tell, we're still running on a sort of 
"common sense," intuition-based system for assessing whether or not the 
number of characters in an analysis is too few, enough, or too many.  And, 
of course, no two people are really going to agree on where the lines should 
be drawn (which is why they haven't been drawn)...

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Jerry D. Harris
Dept of Earth & Environmental Science
University of Pennsylvania
240 S 33rd St
Philadelphia PA  19104-6316
Phone: (215) 898-5630
Fax: (215) 898-0964
E-mail: jdharris@sas.upenn.edu
and     dinogami@hotmail.com
http://www.sas.upenn.edu/~jdharris

_________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com.

Share information about yourself, create your own public profile at 
http://profiles.msn.com.