Chinlestegophis jenkinsi gen. et spec. nov.
Jason D. Pardo, Bryan J. Small, and Adam K. Huttenlocker (2017)
Stem caecilian from the Triassic of Colorado sheds light on the origins of Lissamphibia.
Proceedings of the National Academy of Sciences (advance online publication)
(First, the trivial reminder that the paper is not registered in ZooBank, so the name will only be validly published when the paper version comes out. But that shouldn't take long.)
I'm not so sure it really is a stem-caecilian. But the anatomical details aren't the funny part. Chinlestegophis certainly came out as a stem-caecilian in the Bayesian analysis the authors reported in fig. 2, repeated in fig. S7A. They also ran a parsimony analysis, and the results of that make up fig. S7B.
That's a majority-rule consensus. It looks a lot like the Bayesian tree, but I noticed that none of the nodes between the caecilians and the karaurids + frogs + salamanders occurs in 100% of the trees. I happened to be talking to the first author already, so I asked him what was going on in the other trees – after all, there's no further mention of this in the whole paper including the supplementary information. He wasn't quite sure and thought they may not have saved the individual trees, just the consensus. So I repeated the analysis overnight, or rather in the first 2 hours, 19 minutes and 13 seconds of the night.
I found the same number of trees with the same length, and some were compatible with the majority-rule consensus as expected. But others (including the very first tree I got, as it happens) had a monophyletic, exclusive Lissamphibia next to Gerobatrachus in the dissorophoids, a very mainstream hypothesis, while Chinlestegophis stayed with Rileymillerus in the stereospondyls. Yet others found Chinlestegophis as a stem-caecilian within otherwise the same arrangement. And yet others found a monophyletic, exclusive Lissamphibia next to (Chinlestegophis + Rileymillerus) in the stereospondyls. In all of these alternatives, the karaurids are salamanders as everyone had been thinking so far.
Keep in mind: all these different topologies are equally parsimonious. Not only aren't the percentages on a majority-rule consensus tree any kind of support measure; but the majrule consensus does not offer an overview over the most parsimonious trees either! It is no better than looking at one individual tree plus the strict consensus. (In fact it's a little worse, because PAUP* rounds these percentages: I've seen it happen that a node was marked on the majrule tree as "100%", but was absent from the strict consensus – it turned out it occurred in 99.6% of the most parsimonious trees. But this didn't happen in this case.)
Recommendations for authors
1) Save all trees you get.
2) Don't use the majority-rule consensus at all. If you get a forest and the strict consensus isn't resolved enough to give you a good picture of what you've found, look at a sample of the individual trees (20–30, evenly spaced through the forest, are enough in almost all cases in my experience) and then find a way to describe and/or illustrate your findings (I've been using two different ones lately depending on how much space is available).
Recommendations for reviewers
1) Don't let people get away with only publishing a majrule tree.
2) You aren't given enough time to actually check a data matrix for mistakes, sure. But you do have time (see above) to simply repeat the analysis in order to check what comes out of the matrix as it is. Do that.
PS: in the other new paper, the one on Lethiscus and Coloraderpeton, you'll find the strange phrase "at least stemward of". That should be "at least as stemward [better: rootward] as" – in some of the trees from that matrix, Lethiscus and Coloraderpeton are whatcheeriids, closer to Whatcheeria than to Pederpes. Again, fig. 3 won't quite tell you that, because it's a majority-rule consensus tree.