Basal Archosauromorpha paper – Ezcurra 2016

Another paper repeating the ‘sins’ of the past,
based on an incomplete taxon inclusion list (that also includes taxa that should not be included). And a huge amount of otherwise excellent work! It proves once again that first hand access to specimens and an overly large character list will not bring full resolution to a small taxon list cobbled together by tradition, rather than testing.

I envy, am proud of and have to feel sorry for 
Martin Ezcurra (2016). He went around the world gathering data, obviously took a huge amount of time studying the specimens and writing this paper, but he’s stuck with that less than adequate traditional taxon list rather than the testing offered by the wide gamut taxon list (large reptile tree) in He’s using 96 taxa (vs. 674 at He’s using 600 characters. That should be more than enough, and it is more than enough (less than half that number will do), but more taxa is really what Ezcurra needs.

Just a few notes
Ezcurra wrote: “Jesairosaurus lehmani was described in detail by Jalil (1997). Despite its short neck, this species has been considered since its original description as a member of “Prolacertiformes.” Nevertheless, the phylogenetic position of this species has not been further tested in more recent quantitative analyses.” Yes it has, Here Jesairosaurus nests with drepanosaurs at the base of the Lepidosauriformes, not with Macrocnemus, as shown by Ezcurra (Fig. 1). Drepanosaurs were excluded by Ezcurra.

Figure 1. Ezcurra 2016 tree of basal archosauromorphs. He has basically repeated the mistakes of Nesbitt 2011 here.

Figure 1. Ezcurra 2016 tree of basal archosauromorphs. He has basically repeated the mistakes of Nesbitt 2011 here. Colors denote taxa that lie outside the gamut of the Archosauriformes + Protorosauria under study here.

Ezcurra is still including the thalattosaur, Vancleavea which nests with Doswellia in the Ezcurra tree. It just doesn’t belong in a study on archosauriforms.

He still holds to the tradition of a monophyletic Diapsida proven invalid here.

Ezcurra is still including pterosaurs in an archosauriform study
Proterochampsia (now including Vancleavea) is still recovered by Ezcurra as the proximal outgroup. Phytosauria and Lagerpeton are sister taxa. How is this possible? What characters do they share? They certainly don’t look alike. He notes Peters (2000) then writes, “The phylogenetic analysis conducted here [Ezcurra 2016] constitutes the best data matrix compiled so far to test the position of pterosaurs within Archosauromorpha because of the broad sample of Permo-Triassic species, including the undoubted pterosaur Dimorphodon macronyx.”  Martin, but you’re not looking, really looking at your results. Your proximal outgroup should look kind of like a pterosaur. Right?

Ezcurra notes that 33 extra steps
are needed to place Dimorphodon with Tanystropheidae and 19 synapomorphies support the Ornithodira. That might be true in his study. Hard to imagine how that is possible though. I will try to plow through his 600 characters to figure it out.  Convergence is rampant in the Reptilia. More synapomorphies support pterosaurs outside the Ornithodira when pertinent taxa are not excluded (see below).

Ezcurra writes, Future analyses focused on testing the higher-level phylogenetic relationships of pterosaurs should also incorporate a broader sample of early pterosaurs and some enigmatic diapsids that were found as more closely related to pterosaurs than to other archosauromorphs by Peters (2000) and are not included in the current taxonomic sample (i.e., Langobardisaurus pandolfi, Cosesaurus aviceps, Sharovipteryx mirabilis and Longisquama insignis). However, it seems extremely unlikely that the addition of these enigmatic diapsids, which are unambiguously considered to not be members of Archosauriformes (e.g., Peters, 2000Senter, 2004), will affect the higher-level phylogenetic position of pterosaurs.”

In Science, the word ‘seems” and “extremely unlikely” need to be tested, especially when Langobardisaurus, for instance, shares so many traits with Tanystropheus and Macrocnemus. And especially when they have been tested sixteen years ago (Peters 2000). The word “enigmatic” is inappropriate here, unless Ezcurra just preferred to avoid them and stay with the traditional nod and move on.

Many good color photos of specimens here.
Precise descriptions. Like Nesbitt (2011) he’s just not playing with a full deck — of taxa.

Ezcurra’s tree
had 1.8 million+ possible MPTs. The large reptile tree was fully resolved with a single tree and high Bootstrap values. His analysis 3 recovered 40 MPTs by dropping largely incomplete taxa. That’s often a good idea. No reconstructions were offered, except for some skulls. No gradual accumulations of derived traits for odd partners like pterosaurs, Vancleavea, Doswellia, etc. Many purported sisters do not look alike.

Still not sure how
these trees don’t nest Tropidosuchus and Lagerpeton together. They are virtually identical.

Figure 2. Ezcurra tree with Bremer supports AFTER pruning incomplete taxa.

Figure 2. Ezcurra tree with Bremer supports AFTER pruning incomplete taxa. Many oddly paired sisters still show up here.

Ezcurra comments on Choristodera
“The problematic phylogenetic position of choristoderans may be a result of an unsampled early evolutionary history. The phylogenetic position of choristoderans is also ambiguously resolved in this analysis, but is constrained to the base of either Lepidosauromorpha or Archosauromorpha.” Actually the early history is sampled (here), just not included in this analysis.

Ezcurra has to be feeling pretty confident.
He writes, Much of the general topology of the phylogenetic trees recovered in this analysis agrees with that found by several previous workers (e.g., Sereno, 1991Dilkes, 1998Gottmann-Quesada & Sander, 2009Ezcurra, Lecuona & Martinelli, 2010Nesbitt, 2011Ezcurra, Scheyer & Butler, 2014).”

I’d feel more confident
if all sister taxa looked alike and a gradual accumulation of traits could be traced for every taxon. Ezcurra needs more taxa to weed out the problems here. This study carries with it the sins of past studies.

I was unable to open the Ezcurra data files on either Mesquite or MacClade.

Ezcurra MD 2016.The phylogenetic relationships of basal archosauromorphs, with an emphasis on the systematics of proterosuchian archosauriformsPeerJ 4:e1778
Nesbitt SJ 2011. The early evolution of archosaurs: relationships and the origin of major clades. Bulletin of the American Museum of Natural History 352:1–292 DOI 10.1206/352.1.
Peters D 2000. A reexamination of four prolacertiforms with implications for pterosaur phylogenesis. Rivista Italiana di Paleontologia e Stratigrafia 106:293-336


29 thoughts on “Basal Archosauromorpha paper – Ezcurra 2016

    • If no intervening taxa separate two nodes, they are indeed sisters. Sorry for the late reply. Just got to this to approve the comment. That’s the way WordPress operates with no repliers. The large reptile tree tests many other taxa and there Silesaurids nest with other poposaurs. Lagerpeton with Tropidosuchus (same odd feet you can check this out) and pterosaurs nest with the taxa Ezcurra thought to mention, but did not include.

      • David uses a definition of “sister taxon” which nobody else does. It’s been pointed out numerous times before.

        In any case, there are 26 differences between Lagerpeton and Tropidosuchus in his matrix. I do agree Ezcurra leaving out megalancosaurs, Langobardisaurus, Cosesaurus, Sharovipteryx and Longisquama is a bad thing. The 1.8 million vs. 40 MPT example should really prove to you that having a low number of MPTs is nothing to brag about. It doesn’t mean the analysis is better, only that either taxa that are too incomplete to place exactly aren’t included or that characters that resolve certain parts of the tree weren’t included.

  1. Lagerpeton isn’t the sister taxon of Phytosaurs here? Ezcurra clearly got Lagerpeton as the sister taxa of Dinosauriformes.

  2. Ezcurra didn’t recover Lagerpeton and Phytosaurs as sister taxa. That’s not how sister taxa work. Lagerpeton is clearly the sister taxon of Dinosauriformes and Phytosaurs to more derived Pseudosuchians, according to Ezcurra.

    • Sorry for the late reply. Just found this in “pending”. If two taxa in a cladogram are not separated by another taxa they are sister taxa, even if the cladogram splits to provide two or more sister taxa.

  3. Adding intervening taxa removes the sisterhood status. All you have to do is add taxa between phytosaurs and pterosaurs to remove their sisterhood. No one using Nesbitt’s tree ever seems to do that.

    Sister taxon definition: A sister group or sister taxon is a systematic term from cladistics denoting the closest relatives of a given unit in a phylogenetic tree.

    re: Low number of MPTs. Mickey these cladograms are attempting to model actual evolutionary events, of which there was only one.

    Lack of resolution, in my experience, comes from incomplete taxa nesting with other incomplete (in a different way) taxa, incorrect scoring, or three taxa that are so alike they score alike. There are 3x more characters here than are necessary. Nesbitt 2011 is used as a basis, so I know there are some scoring issues here. These need to be weeded out, not copied out of trust.

    • The intervening taxon is Dinosauromorpha. You could say dinosauromorphs are on the pterosaur stem. What you’re saying is that on the traditional tree of living amniotes (ignoring turtles), lepidosaurs are the sister group of monotremes. But no scientist thinks that’s true because they all use a different definition of the term than you. See the trees-


      EXACT same pattern. But monotremes’ sister group is not Lepidosauria, it’s Theria (Marsupalia+Placentalia). Ditto pterosaurs and phytosaurs. What you’re doing is not taking the “closest relatives of a given unit”, it’s taking the smallest clade of the second closest group. So why do you keep using a different definition than everyone else?

      As for resolution, note that besides incorrect scoring, the reasons you and I listed don’t make an analysis’ results worse. There’s also actual character conflict. If you include more evidence for one hypothesis than another, while I include all of the evidence for your hypothesis plus an equal amount for mine, I’ll have less resolution. But I’ll have the better analysis because it’s using more data so is more honest about reflecting what we know.

      Finally, what you say about the number of necessary characters, that’s not based on any data besides the number you arbitrarily started with when you began the reptile analysis. It’s going to vary widely based on the characters and taxa used. Once you run out of taxa to add, you should really try adding new characters. You might be surprised by the results.

      • “You might be surprised…” : ) Innuendo is not science, Mickey. That’s flirting with Science. Get real. The 150 number of characters comes from Wiens 2003, as I recall. It’s the same statistical solution as the game of 20 questions. You’ll usually get the answer by 20. You don’t need 40 or 60. Less resolution, in my experience, and I’ve had a week of it lately, comes from incorrect scoring. AND from trying to nest a cranium with a post-cranium. Your monotreme examples don’t make the point, as most suprageneric examples seem to. Let’s get back to A-B-C, like the example when you google the term, “sister taxa.” They correctly note that sister taxon to A is B. But they never note that B has two sister taxa, A and C. That’s what’s going on here. If you don’t like pterosaurs nesting NEXT to phytosaurs, just add some intervening taxa, as I have done. And try to avoid suprageneric taxa. It’s as bad as anything that has come along in paleontology.

      • Wiens (2003) shows the opposite of what you claim. We’re using his figure 7, because incomplete taxa in your analysis are not ideally placed and missing data are randomly distributed since you use mostly big obvious characters where missing data depends on the elements preserved. Within that figure, we’re using graphs e and f where 3/4s of the taxa are incomplete because you wisely include every taxon you can code. Within e and f, we’re using option 2 for the top row choices- “data sets 1 and 2 combined (with missing data).” This is because you have missing data but don’t prune taxa as in option 3. Your analysis has 228 characters, so is basically between the 100 and 500 character options of the graph. We’ll use the average of those. The other variables are between best and worst case scenarios. Graphs in e assume incomplete taxa are evenly placed, graphs in f assume they’re placed to maximize long branch attraction. Similarly, the top set in each sub-figure assumes short branch lengths while the bottom set assumes long branch lengths. Finally, your missing data percent is ~27% in your 2013 matrix, which has no reason to vary drastically from your current one. So based on this…

        Short branch length + evenly placed incomplete taxa- ~81% accurate
        Short branch length + poorly placed incomplete taxa- ~81% accurate
        Long branch length + evenly placed incomplete taxa- ~38% accurate
        Long branch length + poorly placed incomplete taxa- ~38% accurate

        So ideally, assuming no miscodings or poorly formed characters, 20-60% of your tree is likely wrong. Note if you want the tree to be basically correct (Wien’s 95-100% division) with ~25% missing data in these circumstances, you’ll need between 500 and 1000 characters in his table. I REALLY hope you take this to heart, because it took me a while to complete.

        “But they never note that B has two sister taxa, A and C. That’s what’s going on here.”

        Because B doesn’t have two sister taxa. It has one- A. That’s what I’m trying to get you to see here. They never note it because it isn’t true. That’s not even what you believe, since by that reasoning it would be Pseudosuchia that is sister to pterosaurs in Ezcurra’s tree. Whatever B is, it’s not a sister taxon according to everyone else in the world who uses the term. So why do you use the term in a different way than everyone else?

  4. In what way is the large number of characters in Ezcurra’s analysis a bad thing? You’ve pointed out that it has an “overly large character list,” but why do you think that? More data is usually a good thing.

    Of course, there is only one tree of life. However, due to the incompleteness of the fossil record, incorrect or ambiguous phylogenies are possible, especially without enough good character data. A large number of MPTs doesn’t mean evolution happened in multiple ways simultaneously, it just means there isn’t enough data yet–and with an analysis including taxa such as Exilisuchus, known from only an ilium, a statement of ignorance about the precise relationships of a taxon (i.e. multiple equally parsimonious trees) is inevitable.

  5. At 150 characters the curve starts to flatten out considerably. Above that number your benefits are increasingly incremental, if that makes sense. In this case, as you correctly note, there needs to be some overlap to avoid multiple MPTs. AND the matrix you set as your base should be excellent and free from errors. Best to start with a good tree with complete specimens then go ahead and have fun with partial skeletons.

    • Wiens 2003 was using simulations containing 16 taxa, with between 100 and 2000 characters. Never did he show that using less characters than taxa can produce an accurate result.
      Moreover, in every single one of the Wiens graphs, it shows that the larger the number of characters the greater the accuracy. In every single one, 2000 characters is shown to produce greater accuracy than 1000, which produces greater accuracy than 500, which produces greater accuracy than 100.

      Moreover, the quality of the characters is important. Yours are full of redundancy and poor character formations (as you have been told numerous times), and these have huge effects on the results. Don’t take my word for it, it has recently been demonstrated quantitatively:

      Simões, Tiago R., Michael W. Caldwell, Alessandro Palci, and Randall L. Nydam. “Giant taxon‐character matrices: quality of character constructions remains critical regardless of size.” Cladistics (2016).

      These are such a simple concepts I am forced to assume one of two things. 1) you are deliberately misrepresenting the result of the Wiens paper because you can’t be bothered to add more characters; 2) you simply don’t understand papers about phylogenetics. Based on our past conversations I’d believe either
      one, but since I prefer to assume ignorance to malice, I’ll simply say you should actually become knowledgeable about phylogenetics and read these papers properly before trying to make your own phylogeny.

  6. And what if A was never discovered? Or included? Everything in the phylogenetic hypothesis is based on the taxon exclusion set, which is driven by tradition, presently, not testing, as Ezcurra confesses. So, sister taxa are those that are closest to one another on the tree.

    Mickey, you’re trying to elevate theory over practice. Sometimes theory (what should happen) falls short of what actually does happen. Theory needs to be adjusted to fit reality, not the other way around. In that case, 228 characters are more than sufficient to lump and separate all 674 taxa. And I can probably add several hundred more.

    It would help, perhaps, to tell me which specimen-based taxa should not nest together in the present trees and we’ll have a real talk about real relationships.

  7. One more thought on the Wiens paper. Remind me, did that paper consider only binary scores? Perhaps the numbers change when 3-9 options are available for some to many scores. Again, theory is usually based on such predetermined constraints. The large reptile tree does not have that particular constraint.

    It dawned on me that your concept of sisters may be those taxa that descend from a common ancestor, not the ancestor itself. That’s fine. Thus A and B are sisters, but basal C is not. Along those same lines, then AB and C are sisters to an un-shown D ancestor, etc. etc. But if we want to avoid suprageneric taxa, then A _and_ B are sisters to C.

  8. “Thus A and B are sisters, but basal C is not. Along those same lines, then AB and C are sisters to an un-shown D ancestor, etc. etc.”

    YES! In the sense that the A+B+C clade is sister to D. Sister groups can be and often are suprageneric. Similarly, they can easily be made of multiple OTUs in an analysis. That’s not just my concept of sister groups, it’s the concept of everyone except for you. And yes, it’s purely based on the OTUs in a cladogram, not the very messy situation in reality. Remember that in reality the genus is no more real than the order or family and there’s not a common ancestor but rather a network of interbreeding individuals that at some point mostly diverge into populations that stop interbreeding and at some later point are mostly unable to breed with the population at that divergence. So what’s the “real” sister to a certain species? Is it the ancestor population which can usually not interbreed with the current population, the last population that couldn’t usually interbreed and diverged? All of which are basically unknowable for ancient fossils. No, it’s just based on lines in a cladogram, which is itself only a certain model for how phylogeny works but is not supposed to be reality.

    “Mickey, you’re trying to elevate theory over practice.”

    You’re the one saying a certain number of characters “is more than enough” and that the statistic is “the same statistical solution as the game of 20 questions.” Seems you liked theory until you learned it defied your practice. You say “228 characters are more than sufficient to lump and separate all 674 taxa”, but we’re not talking about how many characters you need to get a resolved tree, but to get the correct tree. Yeah you get a tree, but probably a significantly inaccurate tree.

  9. “Probably” ? Listen to yourself. It’s a weasel word. Either the tree is inaccurate or it is not. Take a stand. “is more than enough” = practice, not theory. Your words twist your own thoughts. See this in yourself, change, and set yourself free to accept valid hypotheses you seem resistant to. “to get the correct tree” what is the correct tree? Is it the tree that Ezcurra promotes? Take a stand, Mickey. Support it or trash it. Every argument you’ve made is one of semantics, not results. Nitpicking, not substance. And BTW, sisters can include the original stock that continues essentially unchanged from which an isolated group gets separated.

    • “Probably” because Wiens’ analysis leads to a conclusion your tree is between 19 and 62 percent wrong, BUT there are always outliers, plus Wiens’ paper near certainly isn’t the only source on the topic (its main focus was on how incomplete taxa and characters that can be scored for few taxa affect analyses). “Probably” is a science word, not a weasel word.

      “The correct tree” is the tree that results from an analysis with no missing data. So it’s not even the real Tree of Life, but just what your characters and taxa would show if we had perfect specimens of everything. Basically Wiens’ analysis is saying that your missing data percentage and character number combined lead to trees that are on average ~20-60% different than they would be if all data were known. In other words, to make up for ~25% missing data you need a lot more characters (~500-1000) so that enough known scorings are present to get to the real signal in your data.

      And yes, the arguments I’m making here are about process not conclusions. If the process is gravely flawed, the conclusions have no chance. It does no good to argue taxon X actually has state Y for character list Z so should be in position A if your matrix is ~32.5% miscoded, lacks ordered states, has composite and correlated characters, doesn’t include suggested characters that contradict your tree, etc.. All things I detailed in depth about your analysis in 2013.

    • Well, you haven’t added or altered any characters to my knowledge, so the facts your matrix “lacks ordered states, has composite and correlated characters, doesn’t include suggested characters that contradict your tree” are all still 2016 problems. That’s one reason your analysis has such a low accuracy ceiling- you won’t actually fix fundamental issues. You can throw all the taxa at it you want and correct all the codings, but Wiens’ results (which are also what every other scientist has been telling you for years) show that without more characters any result will likely be significantly inaccurate. As a scientist do you honestly think the right response is to think that your personal results are the exception to the statistical data?

  10. Please take a stand, Mickey, deal in reality and tell me which taxa are not correctly nested and where they should go instead. I suggest you cut and paste the taxa, put them to scale in a new layout to make comparisons easier. Write down the synapomorphies if you want to. If you hurry and pick the right taxa, I guarantee your success. There are at least two taxa that are currently wrongly nested. I will update them this weekend.

    I do think I have recovered previously unknown relationships because I have increased the number of opportunities to nest for all included taxa. Some have found more parsimonious nodes and closer sisters than in traditional trees.

    You think I have fundamental issues. I have proven that 228 multiple option characters are sufficient to lump and split 674 taxa and smaller subsets of the latter. That’s a fact that you are having a hard time accepting. I think you have fundamentalist issues, an inability to accept new theories of relationships.

    • I don’t have time to go through your analysis in depth a second time when the first time had no effect on your philosophies. This is a perfect example. I said above-

      “You say “228 characters are more than sufficient to lump and separate all 674 taxa”, but we’re not talking about how many characters you need to get a resolved tree, but to get the correct tree. Yeah you get a tree, but probably a significantly inaccurate tree.”

      And yet you just said-

      “I have proven that 228 multiple option characters are sufficient to lump and split 674 taxa and smaller subsets of the latter. That’s a fact that you are having a hard time accepting.”

      I just explained above that finding a resolved tree (which you obviously do) is not the same as finding a tree that’s correctly resolved, even using ‘correct’ as Wiens does to simply mean “what your characters and taxa would show if we had perfect specimens of everything.” That’s what I have trouble accepting, as does every other scientist who has ever told you that you use too few characters for such a huge swath of taxa. As does Wiens’ statistical data. Your only reply is personal incredulity. Yet Wiens tested 200 matrices for each of his 768 analysis variants, which any scientist would value over their experience with their own tree. There’s not even a way you could know your own tree is the “correct” tree as Wiens uses the term, since no one has perfect specimens of all the taxa you use. That’s why we need models.

      My point is that when I point out issues with your data like this, you don’t understand their importance or incorporate the solutions into your analysis. I don’t know why. But hopefully you at least will stop claiming that 600 characters is more than the amount needed to find correct (sensu Wiens) trees. And recognize that C is not a sister taxon.

  11. Mickey, your attempts at blackwashing are not succeeding because you claim the tree is wrong in total but do not or cannot show where the current mistakes are and why. Until you do so, you are considered one of the lunatic fringe along with Mayasaura, whoever that is.

    • I DID show the current mistakes- “your matrix lacks ordered states, has composite and correlated characters, doesn’t include suggested characters that contradict your tree.” Why do you only consider first-order mistakes of coding accuracy important compared to higher-order mistakes of character construction? Really- why?

      As for pejoratives, the majority agrees with my critiques, so I’m by definition not fringe. The proper pejorative here would be ‘establishment’ or ‘consensus.’ “Lunatic” is a word you’d use as a reason to block my post if I used it towards you.

  12. Which taxa are mistakenly nested? You keep avoiding this and harping on factors that do not impact the the nesting of individual taxa. That what puts you in the fringe. And stop bugging me until you do find mistakes in the resulting cladograms.

    • But how do you know the factors I harp on “do not impact the the nesting of individual taxa”? Have you tried adjusting those factors in your analysis to test that, or know of studies that suggest e.g. correlated characters aren’t that bad? Wiens says the opposite, for instance. All the other people who do cladistics have harped on the same things for your analysis, so again it’s not fringe.

      You still haven’t answered my questions, namely-
      – Why do you only consider first-order mistakes of coding accuracy important compared to higher-order mistakes of character construction?
      – Do you think the right response to Wiens’ study showing analyses like yours are probably significantly wrong is to claim that your personal results are the exception to the statistical data?

      I keep bugging you because until you make these changes, your analysis is probably never going to be good. Every character can be scored perfectly and your results will still be probably wrong (sensu Wiens).

  13. Waitaminute — Dave, did you just refer to Mickey as being guilty of “blackwashing” AND of being part of the ‘lunatic fringe’? Oh, wow, you did. (stops laughing) No, no: you have it all wrong, just like you do for the term ‘sisters’ and for how you might get robust results from a phylogenetic analysis. But, then, I do admit I only come here for the comments… Funny how there are all those nefarious “blackwashers” out there, isn’t it.

      • Sorry for the late reply. Maybe a more basic question for both of us: how do you determine right from wrong? Consensus? Parsimony? Opinion? Dead reckoning? Or should we just present two sides of the argument and let others fight it/figure it out? It is clear I am not getting through to you. If the results are wrong and you cannot point out where they are wrong, we are at an impasse.

        Darren and Squid whatever, same answer applies. Enjoy the moment.

Leave a Reply

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

You are commenting using your 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 )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.