Monday, August 30, 2010

Recoding Holtz 1994 Part 2- 36% miscoded and alternate placements

Holtz's analysis was impressive at the time for its size, in both characters and taxa used. The analysis has a refreshingly low Consistancy Index of 0.51, indicating it was not designed with a particular result in mind. Only 7% (9/126) of the characters are composites and only 3% (4/126) are correlated with others, which is quite an improvement over Gauthier's (1986) analysis. Another improvement is the use of lower level taxa for OTUs, with only Megalosaurus being a potential composite due to the inclusion of Duriavenator's states. The use of polymorphies is an improvement over most prior analyses, but the all-zero outgroup is again an issue as it should be coded as apomorphic or polymorphic for at least 24% (30/126) of the characters. While Holtz did not test any alternate topologies, he did illustrate a consensus cladogram showing which avetheropod nodes collapsed with one more step, and noted in three more steps basal tetanurine phylogeny collapsed. The two main issues with the study were both mentioned by Clark et al. (1994). First, many characters are vague (e.g. depression in periotic region; narrow nasals; periotic region highly pneumatized; chevrons attenuated distally) and a much larger number are more precisely stated but still unquantified. As none were described in more detail, and the cited sources like Bakker et al. (1988) were just as vague, it's difficult to code these unambiguously. Yet the greatest problem are the miscodings. Some are simply due to specimens which were unknown at the time, like most Majungasaurus material, Currie and Carpenter's (2000) Acrocanthosaurus skeleton, Sues' (1997) Chirostenotes skeleton and the MOR Troodon material. Many are due to depending on the pre-1994 literature, especially for coelophysids, Dilophosaurus, Megalosaurus, basal ornithomimosaurs, and non-Oviraptor oviraptorids. Yet it seems as if in most cases, taxa were coded based on what they should have instead of what they were known to have. This is most obvious in the many codings based on elements which are not even preserved in the taxa in question. For instance, Ornitholestes is coded for at least nine characters involving the pubic boot, proximal femur, tibia and tarsus though these are not preserved in the only specimen, which Holtz examined personally. Similarly, Torvosaurus is coded for at least ten characters involving the occiput, axis, distal caudals, carpals and femur though none of these are preserved (the axis and femur are preserved in the "Brontoraptor" specimens some authors assign to Torvosaurus, but these were discovered after Holtz's paper was submitted). Another manifestation of this flaw is when the character states are obvious, but don't match the clade the character was meant to diagnose. Examples are the circular orbits and anteromedially projected femoral heads of coelophysoids and the outgroup, or the large laterotemporal fenestrae of oviraptorids. Since Gauthier used circlar orbits as a coelurosaur character, non-coelurosaurs were seemingly just assumed not to have it.  Similarly, the anteromedially projected femoral head and large laterotemporal fenestra were used to diagnose neoceratosaurs by Bonaparte, so any taxon outside that clade was apparently assumed not to have it.  In all, over a third of the entres (36%- 905/2520) were miscoded.

Regarding Elaphrosaurus, Holtz's original matrix moderately supports it as a ceratosaur, and moderately rejects it as a coelophysoid (7 more steps) or basal ceratosaur (6 more steps).  It strongly rejects placing Elaphrosaurus in Ornithomimosauria (13 more steps).  In the recoded matrix, it takes 4 more steps to put it in Ceratosauria or 5 more to put it in Coelophysoidea.  Making it an abelisauroid is 7 steps longer though.  This all indicates its position in Holtz's cladogram is due to miscodings.  Elaphrosaurus does have some coelurosaur-like characters, such as broad anterior articular surfaces on its anterior cervical centra, no obturator foramen in its pubis, and a triangular obturator process on its ischium.

In Holtz's original matrix, Acrocanthosaurus becomes a basal coelurosaur with a single extra step, while in the recoded matrix is shifts from being a basal coelurosaur to being a carnosaur in three steps.  So the interrelationships of Allosaurus, Acrocanthosaurus and Coelurosauria are weakly supported in both versions.  More interestingly, Acrocanthosaurus is a tyrannosauroid with only a single extra step in the recoded matrix (compared to a moderate 5 steps in the original dataset).  For those wondering, making it AND tyrannosaurids carnosaurs adds 6 steps in both versions.

Holtz's Oviraptoridae + Arctometatarsalia clade (which he named in his thesis) takes 9 extra steps to recover after recoding, making it moderately rejected.  Arctometatarsalia itself takes 15 more steps after recoding, meaning it is strongly rejected.  It breaks down in the original matrix with only 3 added steps, indicating it was poorly supported from the start.

The ornithomimosaur-troodontid clade Bullatosauria breaks down with a single step in the original matrix, if one groups ornithomimosaurs with tyrannosaurids instead.  In the recoded matrix, it only takes 5 extra steps to get Bullatosauria back, so it's still not strongly rejected.  However, in Holtz's original matrix it took a whopping 20 extra steps to enforce Deinonychosauria.  In the recoded version, it only takes one extra step.  This indicates that far from being caused by an absence of transitional taxa like Sinovenator, Bullatosauria existed due to miscoding the derived taxa that were known at the time.


  1. (I have nothing to say in response to this, but I want to register to be notified by email when further comments are posted, so that I won't miss it if Tom responds. As far as I know, the only way to do that in Blogger is by first submitting a comment. Very dumb.)

  2. (Oh -- turns out I was wrong and this blog has NO way to register to be sent email alerts for new comments. Darn.)