Earlier we discussed in parts 1, 2, 3, 4, 5 and 6 the criticisms leveled at ReptileEvolution.com by Darren Naish, author of the Tetrapod Zoology blog titled “Why the world has to ignore ReptileEvolution.com. Here is part 7.
Darren grants credit for my attempt to use species-level taxa and for my criticisms of previous analyses for not including enough of these. He disagrees with my novel nestings because “(1) he finds the ‘conventional’ positions to be far more convincingly supported by character data, and (2) because Dave’s alternative positions are based on erroneous or incomplete analysis of the data available.” Unfortunately, once again, no details are given.
Darren suggests my character list needs to be longer than 228. By comparison, he is working on a project in which 800 characters are employed. I don’t have the references, but I remember the thesis on a paper on cladistic analysis. The more characters the better, of course, but after 150, the logarithmic curve starts to flatten and you don’t get greatly different results thereafter. You’re already in the high 90th percentile of reliability between 175 and 200.
Darren was deeply perplexed by my statement that I didn’t intend to add more characters to my matrix. Of course that would involved reexamining every one of my over 300 taxa for each new trait and I’m going to leave that for the next guy. I have a job, a life, a blog… I can only give paleontology so much and no more. I hope you all understand. Adding taxa is relatively easy by comparison and with a single tree result, so far so good.
Darren asked, “What the hell happened to Test. Test. and Test again?” Well, Darren, I have tested again and again over the past two decades. I am adding taxa and none have shaken the tree. I do make corrections. When I say, “Test,” that’s for those guys and gals who refuse to test lizards and archosaurs and pterosaurs and thalattosaurs together. Get it?
Darren considers my character list to be cherry-picked. Well the list was put in place long ago when their were just 150 to 200 taxa. Since the list has grown to twice that size, I don’t think I a priori picked any cherries. I don’t include the number of characters for ANY particular clade needed for a more focused study. My aim was to include characters that could be applied to as many reptiles as possible. That’s why there is no “plastron” associated with my turtles and no “wing” associated with my pterosaurs.
Darren notes that I really don’t know how to read a cladogram. He says it’s not such a good thing to recover only a few trees in a cladogram (complete resolution). First time I’ve heard that. There are no “findings” without resolution. I also misunderstand the concept of sister taxa, according to Darren. Lucky for me, PAUP understands. Darren then explained that pterosaurs and phytosaurs nested together because they diverged from a common ancestor. That common ancestor was derived from a sister to Erythrosuchus, according to the tree. I don’t know about you, but I’d like to see much less of a mental leap between related taxa, as I found closer sisters among the little fenestrasaurs. And it scares me when professionals stand by such hypotheses.
Concluding, Darren, provides personal praise, but notes that my work will never be accepted (so be it), that (quoting Bennett 2005), I have made no attempt to convince academic palaeontologists of my findings (btw, Chris is the one who said he refuses to read my emails that attempt to convince him of my findings), and I have taken my ideas directly to the general public and presented them as true and correct, (doesn’t everybody??).
All I can say is what I’ve said before: 1) Don’t be afraid to test lizards and fenestrasaurs with pterosaur and archosaurs. 2) Don’t be afraid to trace fossils, even if they seem daunting and difficult. 3) Don’t be afraid to let your results give you your hypothesis, even if it differs from the status quo. 4) Don’t be afraid to make changes when necessary and 5) When others shun you, hang in there if you’ve done good work.
Tomorrow, we’ll ‘cleanse the palate’ with a “nude” Archaeopteryx.
See you then.
The Pterosaur Heresies 
“I don’t have the references, but I remember the thesis on a paper on cladistic analysis. The more characters the better, of course, but after 150, the logarithmic curve starts to flatten and you don’t get greatly different results thereafter. You’re already in the high 90th percentile of reliability between 175 and 200.”
That’s simply not true, as the number of characters necessary to test a hypothesis is at the very least strongly correlated with the number of taxa included. If you have four taxa, then 150 characters should be great. If you have 300 taxa, it’s terribly insufficient.
But more importantly, I think you misunderstand what the point of using PAUP is. Let’s say I want to figure out paravian relationships, which are controversial right now. Is Archaeopteryx a deinonychosaur, is Anchiornis a troodontid, is Rahonavis a dromaeosaurid, etc.. So I find 250 characters that vary between paravians and code every known Mesozoic paravian, which has never been done before in a published analysis. PAUP spits out a tree, and let’s say I’m lucky and actually get only one most parsimonious tree, and furthermore that most of the nodes are very well supported. Say Archaeopteryx comes out as an avialan and it would take ten more steps to move it to Deinonychosauria. Should I feel confident that my tree is accurate? No, I should not.
Why not? Because there’s no guarantee I included all or even most of the characters others have used to find or suggest alternate topologies. If I didn’t include Paul’s (2002) or Xu et al.’s (2011) supposed deinonychosaur characters that Archaeopteryx has, or only included a few, I haven’t really given their idea a chance. Characters are more than just a mass of raw data that can be exchanged for an equally sized mass with little change in the resulting tree. In morphological analyses especially, each character is itself a piece of evidence supporting a certain relationship. The point of PAUP is to give us a way to weigh all of the suggested evidence together objectively to find the tree supported by most of it. If you don’t include most of the suggested evidence, you’re not doing anything useful.
And this is why any statement of “only X characters are needed to analyze y taxa” is wrong. It’s not the total number which is important, but which of the suggested ones you’ve included.
Good luck, Mickey.
So, will you reply to Mickey, or won’t you?
I ask because he’s right.
I’ll test this hypothesis this weekend. Look for a report soon. Hopefully others will do likewise.
Chippendale and Wiens 1994 may be helpful here. I think that’s the reference I was looking for.
Thanks for the reference. The problem is that their figure 1 which shows the relationship you describe is only for one dataset. That particular dataset may only need 300 characters to get the right topology, but that will be different for different datasets. Note too, that simulation lacked missing data, which our fossil analyses have in spades.
Perhaps most important, note it only finds the right tree with 300 characters or more if the right characters are included. These are the “slow characters” that exhibit little homoplasy. When only “fast characters” are included, it never gets above 75% accuracy even with 1000 characters. Since your analysis has a CI of ~0.1, almost every character is lost and gained 10 times, making almost all of them fast characters. So even if your dataset followed this same curve and was accurately and well coded, your tree would only have ~55% accuracy.
And now that I acquired Bull et al. (1993) which that simulation is based on, I can add further details. The dataset is molecular and consists of only 4(!) hypothetical species in an unrooted tree. So there are only three possible arrangements. Whereas in your rooted tree of 300 taxa, there are well… I don’t know the name for numbers that high. There are dodecatillions of possible unrooted trees for 30 taxa, and the number increases exponentially, so yeah. Thus 300 characters will find the right tree out of three possibilities 100% of the time with no missing data. Not similar at all to functionally countless possibilities with missing data. As further proof, consider that molecular analyses have used hundreds if not thousands of characters for years but don’t result in identical trees.
“Darren was deeply perplexed by my statement that I didn’t intend to add more characters to my matrix. Of course that would involved reexamining every one of my over 300 taxa for each new trait and I’m going to leave that for the next guy. I have a job, a life, a blog… I can only give paleontology so much and no more. I hope you all understand. Adding taxa is relatively easy by comparison…”
EXACTLY! It’s long, hard work. That’s why no one’s done it yet, and why questions like what pterosaurs or turtles are related to are still unanswered. I was chosen to run an analysis for a new taxon of troodontid, and if I did it your way, I would have been finished last year. But since I know 250 characters isn’t enough to test the proposed hypothses, I’m still working every day and currently have 740 characters. No one’s blaming you for not taking the years required for such a venture, we’re blaming you for thinking you have the answers without doing the needed work. Now you’ve said before yours is the Wright Flier of amniote phylogenies, but then you also claim to have “answers to those persistent mysteries” and write as if your phylogeny is correct. If you wrote more cautiously about your ideas, you would not encounter such resistence. And on that note…
“I have taken my ideas directly to the general public and presented them as true and correct, (doesn’t everybody??).”
No, we really don’t. Well, the better scientists among us don’t. I have a lot of novel ideas, but look at the language I use. “Thus Poposauroidea is the most likely candidate”, “Pending comparison to a greater variety of lepidosaurs (including fossil taxa), Patricosaurus is provisionally considered more likely to be a choristodere, perhaps closer to neochoristoderes than Cteniogenys”, “In conclusion, Embasaurus is most similar to the basal tyrannosauroid Xiongguanlong, which is also close stratigraphically and geographically. It is less similar to the generally earlier basal tetanurines, so is tentatively referred here to Tyrannosauroidea”. I don’t think it’s true that Embasaurus is a tyrannosauroid, I think it’s most likely with the data I took into account.
“When I say, “Test,” that’s for those guys and gals who refuse to test lizards and archosaurs and pterosaurs and thalattosaurs together. Get it?”
That’s hypocritical. Testing by adding taxa is one way, sure. But testing by adding characters is just as important. So you’re saying others should test by adding taxa but that you shouldn’t test by adding characters. The reality is, you and others should both test by adding taxa and characters.
“I don’t include the number of characters for ANY particular clade needed for a more focused study. My aim was to include characters that could be applied to as many reptiles as possible.”
That’s a problem, and also why no one has done a good genus-level amniote analysis yet. If you were to code 10 divergent pterosaurs for your reptile analysis, you probably wouldn’t get the same topology that you do for your pterosaur analysis. Pterosaurs might be so divergent that they would at least all group together, but that’s not always the case. When I first added several enantiornithine birds to my coelurosaur analysis before adding characters that were meant to group and organize enantiornithines, they didn’t go where they were supposed to, despite being coded for 500 other characters. One was more basal than enantiornithines should be, one was sister to more derived birds. And they had non-enantiornithines like Confuciusornis separating them. This is probably what’s happening in your tree quite often, like with Ophiacodon and Varanosaurus, or with Squamata. I bet if you added Gauthier et al.’s (2012) characters, your squamate tree would mostly match theirs.
Btw, to use a character that can’t be applied to all taxa, just code them with a hyphen in PAUP (-) to indicate they are inapplicable.
*standing ovation*
And then make sure PAUP* is set to interpreting gaps as missing data, not as a 5th base/21st amino acid. Fortunately, treating them as missing data is the default, but check it.
In PAUP*, “-” is the default symbol for gaps in a molecular sequence. It is not actually supposed to occur in a morphological matrix.
Indeed. All I’ve proposed is a model. Does the model work better than other models (on the large scale)? So far it is the only model on the large scale built on species, so it stands alone. Here and there leaves might switch twigs, but the branches are what I’m promoting and hoping others will test in their own way. Discussion will hopefully result, not censure.
You’re very inconsistent on this. Sometimes your rhetoric is like the above, sometimes it’s more like “Answers to those persistent mysteries” and “There’s something very wrong with our pterosaurs”. You oscillate back and forth.
No, it doesn’t stand, as Mickey just explained.
Sorry you feel that way. There will always be fault found in any study. The results, sister taxa with gradually accumulating derived characters, is unmatched. You attack everything but the taxa. Please attack the taxa.
I don’t feel. I’ve explained the reasons for why I think the way they do. If I haven’t explained them with sufficient clarity, ask!
The Relativity of Wrong
It is matched by every single MPT ever, because that’s how parsimony works. In different trees, different characters may accumulate like this, but many always do; this is inevitable.
What? No, we attack the problems.
To your mind’s eye, are there any taxa in the large reptile tree that should nest elsewhere? If so which and where? Or do all sister taxa resemble one another more than any competing candidates within the tree?