Testing Hill 2005

N.  Brocklehurst wrote,
“I think your repeated assertion that palaeontologists don’t test the relationships you suggest is a bit…well its just not true. Admittedly some havn’t been tested e.g. Tetraceratops (never tested outside synapsids), but a great many have. As just one example Hill (2005) uses a (mostly) genus-level taxon list which covers 80 taxa from almost all the major groups in Amniota with charater list almost 3 times the size of yours to show that Caseids do not go with Milleretids and Bolosaurids, Rhynchosaurs do not go with Rhynchocephalians, Ophiacodontids are not the sister to Therapsids, Synapsids do not go with Archosaurs, Captorhinids do not go with Lepidosaurs, Mesosaurids do not go with marine reptiles…I could go on.”

This is a topic worthy of a post. Coincidentally and several years ago I had studied Hill (2005) and submitted a manuscript describing its faults. It was rejected.

What Hill (2005) was looking for and how he did it
Hill (2005) sought to determine the phylogenetic position of turtles within the Amniota by increasing taxonomic sampling and including integumentary characters, like scutes on glyptodons and sauropods. Strange, mixing such taxa, only because they had scutes. But it was published, so hats off to Hill.

As in all supermatrices, no effort was made by Hill to cull the data or study the taxa. The data was presented ‘as is.’ Hill (2005) reported, “The morphological data set assembled here represents the largest yet compiled for Amniota.” He concluded with, “Turtles are here resolved as the sister taxon to a monophyletic Lepidosauria (squamates + Sphenodon), a novel phylogenetic position that nevertheless is consistent with recent molecular and morphological studies that have hypothesized diapsid affinities for this clade.”

I tested Hill (2005) back in 2006 (long before ReptileEvolution.com) in a three-step process.

1. Taking Hill (2005) as is. 
Hill 2005 created a supermatrix by combining published data and new data based on osteology and histology of the integument. Some strange pairings resulted (Fig. 1). Bulky Diadectomorpha nested as sisters to lithe marine Mesosauridae and as sister taxa to the Synapsida. None of these look very much alike. Round-faced Acleistorhinus nested with flat-faced Lanthanosuchus. Short-faced Trilophosaurus nested with long-faced Choristodera (Champsosaurus). The so-called ‘rib’ gliders nested with marine Sauropterygia. Turtles nest with Sphenodon and both are the sister taxa to Archosauria (dinos + crocs + parasuchia + aetosaurs). Parasuchians nest within aetosaurs. Aetosaurs nest within crocodylomorphs, derived from Protosuchus. Other than these misfits, the rest ain’t too bad. And we can’t blame Hill for this because, true to the method, he was just pulling together published trees (but without casting a critical eye on the data).

To one of Neil Brocklehurst points, note the sphenacodonts and basal therapsids are suprageneric in Hill (2005), so there was the opportunity for some cherry-picking of traits and key taxa. Stenocybus. a key taxon, was not included.

Hill 2005. See text for details.

Figure 1. Hill 2005. See text for details. Suprageneric taxa are marked by black squares.

2. Hill 2005 revision #1
A thorough examination of Hill’s data matrix revealed that hundreds of blank matrix boxes could be scored. Hundreds of others could be more accurately rescored, sometimes with additional character states to more accurately reflect characters. Since this was a supertree compilation, such a critical eye was not part of Hill’s process or method. I don’t like to let things slide.

Taxa that were difficult to access and contributed to excess polytomies using Hill’s scoring, such as Coahomasuchus, the titanosaur sauropods, Glyptosaurinae, Akanthosuchus, Goniopholis, Simosuchus and Mahajangasuchus were deleted. None of these taxa are basal to their respective clades.

Characters that were difficult to determine (foramina, braincase, notochordal opening) were left as Hill’s predecessors had scored them.

The resulting cladogram (Fig. 2) shows more appropriate tree topology with most clades in a more reasonable (more parsimonious) order (sister taxa look more alike overall and in detail), but pareiasaurs and turtles still nest here between lizards and Crurotarsi, which appears untenable.

 Hill (2005) revised.

Figure 2. Hill (2005) revised by the addition of more character scores. Note the topology changes.

3. Hill 2005 revision #2
The addition of a just few taxa (in red) to Hill (2005) revision #1 (Fig. 3) recovers a tree topology very much like the large reptile tree, including the major dichotomy at the base of the Reptilia (a hypothesis totally unknown to Hill in 2005). This underscores the importance of a wide gamut in a taxon list when exploring untested relationships. Here turtles nest with pareiasaurs, Procolophon and other lepidosauromorphs. Casea nests with Millerettidae, far from the Synapsida. Kuehneosaurus nests with similarly-shaped arboreal Lepidosauriformes. Synapsids and mesosaurs nest with sauropterygians and archosauriforms.

Adding taxa to the revision of Hill (2005).

Figure 3. Adding taxa to the revision of Hill (2005). Still not perfect, but a lot better.

It’s worthy to note that
the large reptile tree does not include any glyptodonts, derived crocodylomorphs or very many ornamented lizards. Instead the large reptile tree used more basal taxa to establish a wider gamut of relationships, leaving the above-mentioned highly derived taxa for other more focused studies.

It is also worthy to note that
even with so few taxa, and largely using Hill’s characters, the reptile tree dichotomy was recovered.

To Neil’s point about the Hill (2005) 3x larger character list
Once again: It’s the taxon list (not the number of characters) that needs to expand to figure out the amniote tree topology. As an example, see what just a few extra taxa can do to a tree? (Fig. 3). Any number of characters over 150 tends to flatten out the results from 95% consistency to 98% to 98.5% to 99.1%, never quite reaching, but very closely approaching 100%. On the other hand, every additional taxon provides an additional opportunity for any already included taxon to find a more parimonious partner somewhere on the tree. The larger the list, the better.

No. I tested that sucker. This is evidence.

As always, I encourage readers to see specimens, make observations and come to your own conclusions. Test. Test. And test again.

Evidence and support in the form of nexus, pdf and jpeg files will be sent to all who request additional data.

Hill RV 2005. Integration of Morphological Data Sets for Phylogenetic Analysis of Amniota: The Importance of Integumentary Characters and Increased Taxonomic Sampling. Systematic Biology 54(4):530–547.


10 thoughts on “Testing Hill 2005

  1. I’m not going to criticise your analysis of Hill. I am going to say that you missed the point of what I was trying to say. I wasn’t trying to say that the Hill analysis was the proof positive against your work, or that it should be taken as THE work on amniotes, and that it is flawless.
    What I was doing was objecting to your repeated assertion that the relationships that are widely accepted are only accepted because they are traditional, and that your suggestions have not been tested. I was trying to point out that there have been analyses that have tested the combinations of taxa that you have found and have not found them. Yes, you can pick holes and point out the flaws. This is good and science requires it (its this same reason why people try to pick holes in your arguments). But an analysis being flawed does not alter the fact that the analysis exists. There are more analyses that have tested widely, e.g. a variety produced by Gautheir, Reisz, Muller and others. They might have holes in them too. But you can not say that no one has tested many of the relationships that you have proposed.

  2. I understood your point, Neil. Keys to my statements include: 1) Taxon exclusion (at the generic level) and 2) the misuse of suprageneric taxa (cherry picking traits). If you can find a prior study of large gamut that does not contain suprageneric taxa, and includes key taxa, like Gephyrostegus (if we’re talking about the base of the Reptilia) or Daemonosaurus (and include Heterodontosaurus, not Ornithischia), then I’ll acknowledge that. Until then, I don’t think those studies are out there… otherwise someone would have drawn my attention to them. Everyone has been talking about Professor Merck working on this same problem for about a decade, I think. Still unpublished. That’s the closest others have offered. Hill (2005) comes in second, but its chock full of suprageneric taxa and unfilled scores.

    To be fair, I have pinpointed in each blog the key taxa missing from prior studies. I have also noted that prior studies seem to focus on various bumps and pits, rather than more readily viewed and comparable traits, which I use of necessity.

    Finally, the challenge is still out there to find taxa that are not correctly nested in the large reptile tree or large pterosaur tree and to provide a more parsimonious matching, on the generic or specimen level. Switching two or three nodes doesn’t really upset the topology. Someone will have to show why caseids belong with synapsids (on generic level) when Concordia, Orobates, millerettids and bolosaurs are included. Or why pterosaurs belong to archosaurs when various tritosaurs or fenestrasaurs are included.

    If true, it should be easy to demonstrate. If not, it’s time to change the textbooks.

  3. Certainly an interesting experiment, similar to what I do for theropod analyses at http://home.comcast.net/~eoraptor/EvaluatingPhylogeneticAnalyses.htm . I had actually not heard of Hill’s study before, though I must give credit to him for having a free pdf and an extensive database of images and codings on Morphobank. Good aspects include eliminating correlated characters and positively coded inapplicable states, coding the cells unaccounted for in the published matrices he combined (new character-taxon combinations) and using first hand examination for many taxa. Bad aspects include not ordering any characters, trusting published codings when available, using some suprageneric taxa, and an odd choice of some taxa (e.g. no basal saurischians or basal avemetatarsalians, no non-archosaurian archosauriforms).

    However, your evaluation could be problematic. First, while some analyses do genuinely have an issue with not coding scorable characters (so Hill’s might be one), you also think you see more detail than most paleontologists. So maybe some of the hundreds of boxes you scored actually can’t be coded, and are introducing bad data. For the same reason, some of your hundreds of recodings could be erroneous, though no doubt the matrices Hill took some data from were also miscoded in ways. Perhaps more importantly, your introduction of new states near certainly harmed the analysis, for while Hill understands reductive coding and lack of correlation, you don’t think it’s important.

    So just as before, you didn’t test your statement. Is it more taxa that’s important to getting your reptile tree? Or is it your construction of characters and choice of codings? I would guess that adding those seven taxa to Hill’s matrix before your alterations would not result in anything like your figure 3. And that’s the test we would need to evaluate your statement.

    Why don’t you use this matrix as a new starting point for your analysis? It has 60% more characters including the minutia I complain you leave out, and already has 30% of the taxa you use coded how you like.

  4. Your presumption toward the negative, rather than the neutral is not a good way to start.
    The fact that my sister taxa look like they’re related to each other also still hasn’t landed. Why is it so hard for you to accept that my nestings might be right? Why can’t you pInpoint problem areas and offer more parsimonious solutions? (That’s a rhetorical question, of course. I don’t expect to have an answer, as this same discussion drags on and on, like I’m talking to a Creationist). You have your assignment: Nest Daemonosaurus with theropods while including Heterodontosaurus and Massospondylus. That’s a half-dozen to a dozen taxa at most to deal with. Then we’ll compare answers.

    • I’m being neutral as far as I can see. I noted good and bad aspects of Hill, some in which your analysis is better, some are shared with your analysis, and some in which your analysis is worse. I said Hill may be right in not coding some entries, or you may be right in coding them. I said some of your recodings may be wrong, and some of Hill’s codings may be wrong. I have never stated all of your nestings are wrong, and have in fact stated I agree with some (e.g. pterosaurs outside Archosauriformes). So that’s you arguing against straw-me.

      As for sister taxa being similar, there are a few issues. One, you still think of cladograms the wrong way like when you say “The so-called ‘rib’ gliders nested with marine Sauropterygia” in Hill’s tree. There, rib gliders nest sister to a clade containing sauropterygians and Reptilia/Sauria. There’s no reason they should look like sauropterygians. The taxon at the base of the glider+(saurop+reptile) clade would be a lizardy thing (think Eosuchian grade if we were in the fifties), which split into one population eventually having gliders, and one which stayed lizardy. The latter lizardy population at the base of the (saurop+reptile) clade split into two population, one that became aquatic, and one which stayed lizardy to evolve into things like Sphenodon and basal archosauriforms. So why should a descendant in the population that eventually glided be very similar to a descendant in the population that eventually became aquatic? It’s like you saying the platypus nested with kangaroos in a standard mammal phylogeny, when actually they both came from opposumy things that also gave rise to placentals. Another issue is your analysis contains mostly easily visible characters, so will tend to have broadly similar-looking taxa clade together. And finally, as shown by Afrotheria vs. Ungulata and other examples, placing broadly similar taxa together is often wrong, being the result of convergence instead of a similar common ancestor.

      You ask “Why can’t you pInpoint problem areas and offer more parsimonious solutions?”, but I’ve done so extensively. For example…

      I noted of 25 characters supporting Ophiacodontidae, you include 6. Thus you can’t say Varanosaurus is not an ophiacodont, because you didn’t use three fourths of the evidence that say it is. While your analysis does include “several times more taxa” than Benson’s, most of these are not particularly close to ophiacodonts, and it includes many that are close to them but which your analysis lacks (e.g. Stereophallodon, Pyozia).

      I noted more than fifteen of your characters that are correlated, so weight certain features more than others.

      Here I noted you only included 3 (17%) of saurischian characters and 1 (8%) of dinosaurian characters, so it’s no surprise you didn’t recover either clade. That could do a lot to explain how you get Daemonosaurus in Ornithischia.

      Here I looked into your nine dental characters and found five needed adjustment, in addition to some miscodings.

      After posts like these, you haven’t fixed your character states or ordering and you haven’t added the additional characters. Yet you constantly claim no one has challenged your analysis with specific critiques. Funny you should mention the creationist comparison, as it seems apt from my side too. Always saying there are no transitional fossils, no species have been observed to form, etc.; and ignoring all the examples given. I know adding characters and fixing characters is tedious work, but no one said fixing your analysis would be easy. Or how you have no explanation for how nuclear genes, mitochondrial genes and other morphological analyses can result in the same tree which is different from your own, just like creationists have no explanation for how all the different types of radioactive decay, tree rings, ice cores, etc. can arrive at the same dates which are older than they think the Earth is. You know the different genes must be wrong because your tree is right, you just don’t know why they’re wrong yet.

      • You missed the point, Mickey. What I am asking you to do is adjust my tree topology to suit your hypothesis. Then we can compare your changes to the large reptile tree. I’ll put up your figures and compare them with mine. Then we’ll look at synapomorphies. The list of characters is infinite for any taxon, as you can get down to individual chemistry and DNA if you like. So, merely listing a few to a dozen characters is not a solution. Rearrange the tree topology like you want to, then we’ll talk.

        With regard to DNA topologies, as above, offer me your solution based on DNA we’ll put those taxa together, your way and my way and see how they compare visually.

        With regard to arboreal taxa nesting next to aquatic taxa, you’re stretching credibility. We have more parsimonious intervening taxa. There is no reason to ignore them.

    • I won’t take up your challange on Theropods (my knowledge of Dino morphology isn’t great) but I may look into the parareptile/Caseid bit when I next go to America.

      For now I will point out an objection I have to another argument you keep using: that your taxa ‘look better’ together than the ‘traditional’ relationships (the ‘unusual bedfellows’ argument). You used it here: “Bulky Diadectomorpha nested as sisters to lithe marine Mesosauridae and as sister taxa to the Synapsida. None of these look very much alike. Round-faced Acleistorhinus nested with flat-faced Lanthanosuchus. Short-faced Trilophosaurus nested with long-faced Choristodera (Champsosaurus). The so-called ‘rib’ gliders nested with marine Sauropterygia.” etc etc

      There are two problems with this argument. Firstly, your large reptile tree, whilst haveing an enormous varity of clades, does not include the basal members of those clades. For example, your Caseid representatives include only the large herbivorous forms Casea, Ennatosaurus and Cotylorhynchus. However, not all Caseids were large herbivores. The basalmost published one (Oromycter) was a medium-small omnivore. I have also heard in a talk about a possible even-more-basal Caseid (in review) that was a small insectivore.

      The exclusion of these taxa hugely affects the character polarities. Yes, Casea and Cotylorhynchus don’t look much like their putative sisters in the synapsids. But this is not what Caseids looked like early in their evolution. By missing out basal taxa, you hugely increase the probablilty of convergences being taken as synapomorphies. Other examples include the use of Ophiacodon but not Archaeothyris, Sphenacodon and Dimetrodon but not Cutleria and Cryptovenator, and Edaphosaurus but not Ianthasaurus. Saying two derived members of putitive sister clades don’t look like eachother is pointless.

      In a perfect world, we should obviously have relationships where we can see a gradual transition from one species to another. In the real world, do we expect to see this? NO we most certainly do not. Why? Because the fossil record is incomplete. And in the areas you’re examining (the base of Amniotes, during the Carboniferous and Permian) the fossil record is frankly pretty dire. Obviously wer’re not going to see every taxon look like its sister; the early record is so patchy.

      Remember: if we had no fossil record at all, Birds and Crocodiles would be unusual bedfellowes.

  5. To your points, Neil: for Caseasauria I do include the basal Oedaeleops and Eothyris, and beyond them, Concordia. I will find Oromycter and add it, if it is substantial. Archaeothyris and Cutleria are included in the pelycosaur/therapsid tree, distinct from the larger reptile tree. I blogged about Cutleria here earlier. Cryptovenator is known from only a partial mandible, so I did not include it – yet.

    re: the incomplete taxa, there is little to no controversy in the relationship of Dimetrodon and Edaphosaurus. Many of the taxa I include there is no controversy.

    Saying two taxa look alike is not pointless. It is THE point. Morphology drives phylogenetic analysis. Convergence overall does not apply to skeletal details as in the example of dolphins and ichthyosaurs. It’s the skeletal details that come together in analysis.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s