[Taxacom] cladistics (was: clique analysis in textbooks)

John Grehan jgrehan at sciencebuff.org
Fri Aug 19 08:04:13 CDT 2011

For anyone who wants to wade through this further.

-----Original Message-----
From: taxacom-bounces at mailman.nhm.ku.edu
[mailto:taxacom-bounces at mailman.nhm.ku.edu] On Behalf Of Pierre

> I am now certain that we disagree about the definition of "amateur"

Words often have multiple definitions. No worries.

> not brilliant, basic: it's in textbooks and relevant literature (I
really, I am not just pretending that some possible textbooks... 

Brilliant because you get the point, and yes it is basic. That's the

> what is brilliant in my view is the recommendation by Peter Hovenkamp 
> that people interested in parsimony analysis should read the relevant 
>  literature first (you called it "crazy", 

Don't think so.

> but I presume it means "brilliant" in grehanian special jargon - or 
> "kangaroo" may be, who knows? esoterism is unfathomable...)

Explained above.

> 'Compatible' here is defined as those character states that are
compatible with the criterion of being uniquely shared within the

"compatible" is not here defined, it's defined in textbooks and relevant

literature, it's basic "cladistic" jargon for "compatibility analysis", 
also known as "clique analysis"

The original sentence was not mine - hence the ???? after it that I

> your approach is not really "clique analysis" by any standards,
> you are not analysing your complete data set (all taxa, all
> to find the clades supported by the largest clique of mutually 
> compatible and homoplasy-free characters (CI = 1 for the largest
> on this topology); rather, you are pre-defining ingroup versus
> and then you select the clique of characters fitting this elementary 
> outgroup/ingroup topology,

If limiting characters to those that are uniquely shared within the
ingroup for the ingroup analysis is not really 'clique analysis' is ok
with me. I don't have any strong view about what anyone wants to call

> so you have selected the series ("clique") of mutually compatible 
> characters _possibly_ uniquely derived in your selected ingroup - but 
> they are not necessarily "uniquely derived" in the details (see
> because they can still appear as homoplastic inside your ingroup, and 
> this you will know only after your will have performed your ingroup 
> analysis using surviving characters, when you observe a suboptimal CI
> 1 

They may not be uniquely derived with respect to individual taxa within
the ingroup, but they remain unqiue to the ingroup, whether or not they
may be postulated to be either independently derived within the ingroup,
or represent primitive retentions (i.e. a uniquely derived condition for
the entire ingroup, but now only retained in some member taxa
irrespective of their phylogenetic affinities within the ingroup_.

> (Sergio Vargas also explained at length how enlarging your ingroup 
> automatically changes your retained data set; hence possibly the 
> topology connecting the taxa of your previous ingroup inside this now 
> enlarged ingroup)

The orangutan study began with an assumed state for the outgroup. That
may be an Achilles heel and if someone shows that to be the case then so
be it. Its an interesting proposition. I might look into it sometime by
doing a larger analysis.

> Please explain the 'just so' rejection of thick enamel.

>I mean that it's just your opinion (I bet we agree on the meaning of 
> "opinion", you very frequently use this term)
> your opinion is that thick enamel is good despite being homoplastic in

> outgroups, hence it does not fit your general "potentially 
> homoplasy-free" selection criterion, but you retain this character 
> anyway - just so (maybe it's what Richard Zander called the part of 
> "intuition" in your "method")

Clarification understood. I agree that with the enamel thickness we have
made a subjective judgment (and we are explicit about this) that the
occurrence in two monkey genera represents an independent origin. This
is where a broader comparison of monkeys, apes and humans as an ingroup
with prosimians as an outgroup would presumably show, unless it ends up
that these two monkey genera (one OW one NW) are linked with humans and

> you acknowledged in a recent post that you could reject this character

> for this reason, so it seems that by moments you are conscious of this

> internal incoherence in your analysis (besides the other incoherence
> to your character selection criterion depending on the range of your 
> ingroup, as noted by Sergio, and above, and below)

If I were strict I would stick to character states that are uniquely
shared, but in the orangutan analysis we also argued for the inclusion
of features that are rare in the outgroup. If this is seen to be too
contradictory then one could reanalyze our data without such features.

>> 2) perform ingroup analysis by applying unweighted parsimony [not
>> compatibility analysis] so that now homoplastic characters can play
>> their role in defining subclades inside the ingroup, while this is
>> denied for outgroups [according to the goose / gander principle].
> What homoplastic characters?

> those characters that may finally appear to be not mutually compatible

> and homoplasy-free inside your ingroup, they may show some homoplasy
> your ingroup, so that you have no possible topological resolution with

> an optimal CI = 1 (despite your pre-selection of characters
> to the ingroup"); 

If they are homoplasious within the ingroup then they cannot define any
subclades within the ingroup.

> are you OK for "homoplasy"? (convergences, parallelisms, reversals...)
> is your model of character evolution rejecting the possibility of 
> reversals? (this could explain your character selection procedure, but

> not your internal ingroup analysis procedure: I call this incoherence,

> just like Sergio)

Well I don't agree.

> congratulations for using the term "parsimony"
> now, in jargon, "clustering" is reserved for overall similarity 
> analysis, i.e some phenetic clustering  (read Sergio, and textbooks)

Clustering is clustering is clustering. Group some things together and
you are clustering - however it is done.

> now you wrote: "Yes I am asserting that 'my' form of cladistics is 
> necessarily correct - or at least more correct or better than some 
> others." (down below) this is contradictory with "I am not tied to any

> particular analysis" (above)

No it is not.

> ...deliberate play of moving targets and shell games?... sheer 
> confusion?... you should know better than me

Don't worry.

> if you are not tied to a particular analysis, this means that there is

> no single "straightforward" procedure (see my supposedly "brilliant" 
> comment above), but strangely enough you are using no one of the usual

> known procedures (by the way, what do you mean by "straightforward", 
> algorithmically speaking?... not to question what "caveats"

Straight forward or not, no worries.

> I suggest that you use one of the classic procedures (because you
> care finally, if I understand well), the more universally used for 
> morphology being by far unweighted parsimony - i.e. treating your 
> ougroups like you treated your ingroup: no more incoherence

Then it would not be an outgroup.

> just put all characters in the data matrix, find the optimal
> parsimonious topology, root between outgroups and putative ingroup (if

> possible, so you are testing the coherence of your ingroup by the way:

> read PAUP user manual as an elementary textbook for this)

PAUP is not the definition of cladistics.

> In principle I would be happy to use three-item analysis, although
right now I do not have access to a program to do that.

> amazing... ney, fascinating...
> may I respectfully suggest that you read the relevant literature this 
> time and try to make up your own mind before you do that?
> e.g. in "Cladistics": Deleporte, De Laet (better, shot it dead),
> and Kluge, and quite recently Farris again (highly recommendable in my

> view, against pure classificatory formalism and fancy history of
> this could save you from professional suicide...
>(a variant of 3ia appeared as an involuntary reinvention of phenetics, 
> great fun - I mean true phenetics = clustering whatever character data

> set on the basis of overall similarity among objects, not grehanian 
> esoteric jargon concerning character selection)

Every clustering method can have its problems. I was interested, 'in
principle' in the idea of analyzing data where each character involved a
three taxon statement of relationship.

> at the very limit: retain only one species as outgroup, put all other 
> species in your ingroup,  and you save all the characters you
> rejected: do you get the point?

I get the point that such characters may now be operable in evaluationg
the relationships among a broader group of taxa, but one could continue
this indefinitely to have to analyze all living million plust taxa all
at once.

>they would bother at all with exposing your utterly incoherent method

Actually its not that incoherent, rather its something that you disagree

> listen: given the same data (characters and taxa), suppose your
> is now enlarged so that only one taxon is left as outgroup, then all 
> characters are necessarily "restricted to the ingroup" (to adopt your 
> esoteric jargon) - and you can also consider all the intermediary 
> possibilities: the larger the ingroup, the smaller the outgroup, and
> more characters are saved from rejection by your own criterion - do
> get it?

Yes, but this criterion one should analysize all of life at once. We
decided in our analysis of human and great ape relationships to accept
that the large bodied hominoids represented a monophyletic ingroup, and
that gibbons were the sistergroup, and that OW monkeys where the next
sistergroup, and that NW monkeys were the next sistergroup, and
prosimians the next sistergroup. We limited our outgroup to gibbons and
monkeys and proceeded on that basis. If our analysis is eventually
refuted, then so be it. 

> so, enlarging your ingroup would make you mechanically accept more 
> characters, 

Which is why we chose to make sure we had a large outgroup.

> we can also agree that you read no textbooks really (I insist that you

> should)

I guess there are some who see textbooks as the ultimate authority.

John Grehan

More information about the Taxacom mailing list