Re: "Ratite" polyphyly and paleognathous dromornithids

[David Marjanovic <david.marjanovic@gmx.at>]

> > How did they get those  guide trees? (I'll have access to the paper
>> in a bit over a week.)
>
> They considered all 15 possible rooted trees for 4 ingroup taxa
> (ostrich, rheas, tinamous, emu -- rheas and tinamous were
> constrained to be monophyletic and the interrelationships of outgroup
> taxa were fixed) and calculated their branch lengths using maximum
> likelihood.

So, they tried 15 guide trees, got 15 alignments, and then chose the one 
that gave the shortest tree as a result?

> However, the guide trees were  only used to assess alignment bias;
> their other analyses used manual alignments.

Ah.

> > I ask because ModelTest  used to derive its guide tree by
>> neighbor-joining (and the Jukes/Cantor model), while yours truly &
>> Laurin (2007, Syst. Biol.) noticed that changing it to using the
>> most parsimonious guide tree yields different results... and, at
>> least in the example we used, the most parsimonious guide tree is
>> much less horribly implausible than the one found by
>> neighbor-joining!
>
> Interesting. Would be surprised if that were a general rule, though,
> as neighbor-joining seems to be consistent over a wider region of
> parameter space than parsimony.

~:-| Even with Jukes-Cantor? And in which regions of parameter space do 
real examples most often occur?

It's difficult to imagine that a phenetic method would converge on the 
right phylogeny more often than a phylogenetic method.

> There is some methodological  weirdness in the analysis. For example,
> in order to test the mtDNA-based hypothesis of moas and tinamous
> being sister groups (Phillips et al. 2010), they ran a constrained
> analysis where "_Pterocnemia_, _Casuarius casuarius_, _Aepyornis_ and
> _Apteryx australis_ were forced to lie outside of a clade containing
> moa and tinamous" (Wothy & Scofield 2012:103). However, a few pages
> further, they wrote: "The analysis failed to unite tinamous and moa
> as a clade" (Worthy & Scofield 2012:107). Instead, they got two trees
> with moas grouping with rheas, ostriches, casuariids, elephantbirds,
> and kiwis to the exclusion of tinamous. Not only did the constraint
> not work as intended; it's not entirely clear what it actually did.

~:-| It really isn't clear. If they're talking about the same analysis 
both times, it's simply not possible that they enforced a constraint and 
it didn't work. Perhaps they forgot to write "enforce=yes" into their 
NEXUS file?