[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]

Re: Reduced Consensus (Was: Afrotheria revisited)



[I'm CC'ing this reply to the DML in the hope that it'll be useful to
someone else out there.  Hope you don't mind my making public the
quoted excepts of your message to me.]

Allen Hazen writes:
 > Dear Mike Taylor--
 >      Thanks for your reply!

My pleasure.  Having just learned all that, it was useful for me to go
through the exercise of explaining it (though, evidently, inadequately
-- see below).

 >      "Reduced consensus" sounds about like what I figured had
 > happened.  I looked briefly at the additional information (thank
 > goodness for a modern computer: my old one I wasn't able to access
 > that sort of thing even from journals our library gets!), but have
 > to look at it more carefully to see if it tells me anything.
 > Arsinoitherium doesn't seem to have a particularly high proportion
 > of missing data in their matrix, so it's not quite like yourr
 > hypothetical "Eodino".

Well, missing data is _one_ cause of "rogue taxa", and one that I am
particularly aware of since I work mostly with very fragmentary
sauropod specimens.  But there are other possible causes, including of
course homoplasy.

 > On the other hand, I can think of reasons why it might be a
 > problem.  It was the largest (body mass) of the taxa they
 > considered, and has famously unusual teeth (long row of bilophodont
 > postcanines, premolars pretty well molarized, as I recall, though I
 > haven't looked recently and am not a trained observer of dental
 > characteristics).  So I guessed it would have lots of special
 > autapomorphies, making it into a LONG BRANCH.  And we all know THEY
 > can be a worry!

Autapomorphies shouldn't be a problem, since they just make the OTU
distinct within the clade that its other characters assign it to.  Its
a funny mix of synapomorphies that will do for you.

 >      I'm still not 100% convinced it isn't cheating: I'm genuinely not 
 > sure.

If it's any use to you, I am :-) Seriously, it's not cheating.  If you
excluded it from the analysis it would be cheating.  Think of it like
a football league.  If you took Everton out of the league before it
started, that would be cheating since it would influence the results:
Everton always raise their game more when they play Liverpool than
when they play anyone else, so removing them _a priori_ would be in
Liverpool's favour.  But if you let them play all their matches and
then simply do not list them in the end-of-season table, then that _a
posteriori_ deletion does not change the relative positions of the
other clubs -- so Manchester United might come out a point ahead of
Liverpool due to having beaten Everton while Liverpool lost to them,
and the _a posteri_ "reduced consensus league table" will show that.

Of course, doing this in a league table is useless because the table
is a linear thing.  But doing it with a consensus tree is useful
because it allows you to see the consensus that your analysis --
including Everton, er, I mean Arsinoitherium, established between the
other taxa.

 > I think there can be legitimate and principled reasons for 
 > excluding taxa, but throwing away data always makes me nervous.

The data is still there :-)

 > Yes the data matrix gives a well-resolved tree for 22 taxa, but is
 > that because of "gerrymandering"?  Would it be possible to get a
 > DIFFERENT well-resolved tree by excluding one of the other taxa
 > (this is something that time and computer power could check!)?

Ah -- that's the key question, and the answer is a definitive NO.
(You don't need a computer to check!)  When using reduced consensus,
whichever taxon or taxa you remove, the trees you get are all
consistent with one another -- by definition, since they all taken
from the same initial set of MPTs.

 >      Micky Mortimer mentioned on the DML forum recently a published
 > criticism of a paper that had achieved a well-resolved tree by
 > excluding taxa from consideration: he mentioned it because he had
 > made a similar criticism some months ago on the forum.

Well, he should have published it :-)

 > I'm hoping he will comment: does he think Tabuce et al. are doing
 > something good, or bad like the people he mentioned.

Come on then, Mickey, assure Allen that I'm not cheating :-)

 >      I'm rather hoping that Tabuce et al. are in the clear, since 
 > their conclusions appeal to me.

They're clean.

 >      Leading to another mild discomfort I feel about Tabuce &c.  They 
 > say they have found the apheliscines to be a sister-taxon of the 
 > Macroscelidea, but their trees actually show Apheliscus as a sister 
 > of Macroscelidea+Paenungulata.

I can't comment on that, not having seen the paper (and not caring
about it since it's about furries.)  Sounds like a typo.

 >      Anyway... Thanks again for your comments.  As a
 > non-paleontologist (I'm a philosopher, specializing in logic) I'm
 > never confident that I'm saying anything that REAL paleos will even
 > think DESERVES comment!

I felt exactly the same a couple of years go; fortunately I have the
natural arrogance that is the core natural talent of every
palaeontologist, so I got over it :-)

 _/|_    ___________________________________________________________________
/o ) \/  Mike Taylor    <mike@indexdata.com>    http://www.miketaylor.org.uk
)_v__/\  "Arsene Wenger is trying to influence the panel ... and that's
         typical of someone driven by self-interest" -- Alex Ferguson.
         ("Hello kettle, this is Fergie.  You're black.")