[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]

RFTRA -Reply

So much to cover.... Obviously, I think this post was full of tremendous 
problems and I guess I need to cover them

>>> "Jonathan R. Wagner" <znc14@TTACS.TTU.EDU> 04/16/99 02:41pm  >  I 
>>> understand that this may be an unpopular attitude amongst dinosaur
>workers, but I don't like RFTRA. It is an interesting way of showing you >the 
>difference between reconstructions. However, it does not explicitly >address 
>allometry, the "RFTRA coefficients" generated are by no >means objective 
>measures of difference (note the clustering of points in >Chapman and 
>Brett-Surman's and study,  and, as Chapman points out >(in _Dinosaur 
>Systematics_), it is highly sensitive to the reconstruction >used. This adds a 
>level of uncertainty to the uncertainty already present >when >dealing with 
>fossil material.

JW seems to suffer from the disease where he thinks all morphometric approaches 
must attempt to settle all questions simultaneously. This is, at best, a naive 
assumption and, at worst, an indication of great problems in the basal 
scientific understanding of the person doing the science. 
Allometric/conventional multivariate morphometric approaches are useful and I 
still use them tremendously and have done, I suspect, orders of magnitude more 
of them than JW ever has. they also have their own problems as well, in that 
they provide information that is disembodied from the original geometric of the 
organisms being studied and as such can be difficult to relate back to the 
original material.

Frankly, there are no objective ways to do this with any technique, all are 
dependent on what things are measured and how you choose to measure them. To 
think otherwise is extremely simplistic. Take two researchers who choose to 
analyze allometry in long bones and they will, inevitably have significant 
differences in the measurements they take and, as a result, the interpretation. 
Yes, all geometric shape analysis methods have a strong component of dependence 
on the actual points analyzed, but that is also a strength, in that the 
researcher chooses what he wants to zero in on. To think that allometric 
studies don't include this component as well is simply wrong. Period.

> Here is an excerpt from a review I wrote on the Chapman and
>Brett-Surman paper for a class:
>Chapman, R. E., and M. K. Brett-Surman. 1990 Morphometric >observations on 
>hadrosaurid ornithopods. pp. 163-177 in K. Carpenter >and P. J. Currie (eds.), 
>Dinosaur Systematics: Approaches and >Perspectives. Cambridge University 
>Press, Cambridge, New York, 
>        "This study provides an interesting case-study in the use of
>morphometric (phenetic) date to evaluate evolutionary difference and >support 
>phylogenetic inferences. The study provides characterizations >of 
>morphological variation in advanced ornithopods, and indicates some
>interesting correlations between shape analysis and phylogenetic >analysis, 
>but fails to present either set of results in an effective, cogent,
>applicable manner. Part of the reason for this inefficacy is an inability on
>the part of the authors to draw an explicit philosophical and functional
>separation between phenetic and phylogenetic evidence.

I'm sorry you don't care for the writing in the paper but I suspect you should 
have read it more carefully. At the time it was written Horner was starting to 
bang the drum for a polyphyletic Hadrosauridae and we endeavored to answer a 
single question which is, if you use geometric shape analysis methods to 
examine this question, do they support a single or double (or more) origin of 
the hadrosaurs. That simple. That we also wanted to note the levels of 
variation we observed in the groups was an added bonus and what I would expect 
anyone to do.

>        "This is further complicated by a lack of clarity in outlining the
>goals of the study. The exploratory aspect of the study is coupled with >an 
>implicit confirmatory aspect. The same methodology is used for both
>functions, and both purposes are served by a highly interpretive >approach to  
>the handling of the results. No criteria for confirmation are >established, 
>and exploration is largely limited to conscious or >unconscious attempts to 
>support a priori observations. For example, it is >noted that the landmarks  
>showing the greatest change in >lambeosaurines are those associated with the 
>snout and crest, while >other landmark positions remained conservative in 
>their location. >However, landmarks were specifically concentrated in areas
>which were expected to show the most change, and fewer landmarks >were placed 
>in other parts of the skull (see below).

Try reading the paper again, we don't try and establish a phylogeney with the 
methods at all and state explicitly, and quote me in  the other paper on this 
in the volume, that we are not doing that and should not do that. So we are 
only looking at one purpose. We put landmarks where we could get them and there 
are plenty in the more conservative area. There are 4 in the snout region, 3 on 
top and back of the skull and a bigger bunch in the more conservative area. You 
are flat wrong here.

>        "Any study which attempts to use morphometric (and therefore
>phenetic) similarity to evaluate phylogeny treads on unsteady ground. In
>this case, the authors do not establish a reasonable criterion by which
>phenetic assessments are to be used to evaluate a phylogenetic >hypothesis. 
>The authors state that since it is more parsimonious to >explain morphological 
>similarity in the context of homology (i.e. it is >inherited from a common 
>ancestor), if morphological similarity among >phylogenetically disparate taxa 
>cannot be "recognized" as >convergence, then the phylogenetic hypothesis must 
>be re-evaluated. >How convergence is to be "recognized", and what weight this 
>is to be >given with respect to phylogenetic analysis is not addressed. For 
>>example, they evaluate the cranial similarities between Ouranosaurus >and 
>Anatotitan as convergent, presumably within the context of another 
>>phylogenetic study. They do not explain how this convergence is
>"recognized", except to point out that both taxa posses an elongate >snout 
>with an unexpanded narial opening.

Again, totally incorrect. The main approach we used to do this was to find 
those taxa among the lambeosaurines which were closest morphologically to them 
among the iguanodonts and lambeosaurines. All these "nearest neighbors" were 
among the hadrosaurines. Therefore, it is more parsimonious to assume that this 
is due for genetic reasons rather than convergence. Period. We do not construct 
a tree with this data, we simply see the implications of the morphology for 
answering this simple question.

>        "The explanation of the overall greater similarity among
>non-lambeosaurines within the context of hadrosaurid monophyly is >restricted 
>to a vague assertion that this indicates a high degree of >morphological 
>evolution among the lambeosaurines. Given their earlier >statements the 
>authors should be required to explain [how they] >"recognize" this apparent 
>convergence among the non-lambeosaurines. >Obviously, this is a case of 
>symplesiomorphy, with the less derived >members of the group exhibiting a 
>greater phenetic affiliation for each >other, and hardly warrants 
>"recognition". Direct reference to this >concept would both keep the reader 
>aware of the implications of the >study and assure the reader that the authors 
>are aware of those >implications. However, throughout the discussion of their 
>results, the >authors make no explicit attempt to relate the findings of this 
>study to >phylogenetic concepts, nor do they recognize the implications of 
>their >results in such a context, beyond very general statements such as the 
>>one above.

I'm glad JW seems to think he is the world's arbiter of what are interesting 
questions but, when such data are available, it would be nice to see that is 
the case rather than assuming it. I suspect JW needs to stop assuming his 
results before he gets them - or doesn't bother to. That we don't have the 
exact discussion he wishes to see is, frankly, not all that interesting to me. 
Whether the lambeosaurines were less derived was one of the questions that was 
answered and part of the question. If they were not part of the Hadrosauridae, 
then they would not be more derived, necessarily.

>        "As a means of summarizing data the UPGMA dendrograms may >be useful, 
> however conclusions drawn from them are phenetic, and >are therefore not 
> necessarily useful in elucidating phylogeny. The >authors do not explicitly 
> interpret the meaning of the dendrograms, and >references to them are usually 
> in the context of a confirmation of >previous observations, or noting where 
> the dendrograms produce >unexpected clusters. In the analysis of pubes, it is 
> noted that the >clustering of Camptosaurus and specimens of Iguanodon reflect 
> >"family-level similarities." This reflects the use of the dendrograms to 
> >confirm previous non-phylogenetic classificatory hypotheses rather >than 
> elucidate phylogenetic relationships. Among the lambeosaurine >crania, the 
> nearest neighbor among non-lambeosaurines is always >Saurolophus, yet this 
> taxon is in a completely separate cluster on the >UPGMA dendrogram. 
> Additionally, juvenile "Procheneosaurus" was the >nearest neighbor for all 
> non-lambeosa!
urines. Little interpretation of >these phenomena is given, despite their 
importance for phylogeny and >ontogeny respectively. How these various phenetic 
techniques relate to >phylogenetic patterns is left >unclear.

Cluster analyses are used as graphical tools and are useful to those who 
understand fully their usefulness and limitations, which JW obviously does not. 
They show a first estimation of similarity in shape in this context. That 
Saurolophus is in a different cluster than the lambeosaurines is no surprise 
given the nature of the morphologies. That does not in any way contradict the 
nearest neighbor data presented. I suggest you work harder to make cluster 
analyses a part of your toolbox for research rather than just throw the 
technique away for dubious reasons. If you understand the technique, it can be 
useful for deducing some information. Similarities come from many levels and we 
did what we could to try and explain what we saw. At the time this was written, 
there were no reasonable and ver useful phylogenetic studies available for 
ornithopods that would be all that useful, so we worked with what was available 
at the time. I would do lots different now, but so what?

>        "Further, the authors use paraphyletic groups (iguanodonts and
>possibly hadrosaurines) in their analyses. Paraphyletic groups are less
>useful than monophyletic groups in elucidating phylogeny. The use of
>paraphyletic groups demands a series of conceptual considerations >which the 
>authors do not appear to have made. Instead, the authors >consistently treat 
>these groups as if they were real evolutionary >entities, potentially 
>undermining the utility of their results. 

We used genera as individual units, so this is not particularly scintillating 
in its bite. We also used the prevailing opinion on the relationships at the 
time of the writing of the paper. That these relationships are now different, 
so what? The last sentence above is actually pretty historically ignorant. We 
treated them as they were and, frankly, the inferences we made still hold up 
using the level we did. That we used the term iguanodon at times does not 
require apologies. 99+% of all dinosaur people did then. WE treated the 
individual taxa as different units. 

>        "The authors do make broad attempts at relating the overall results
>with respect to a particular phylogeny. They suggest that conventional
>phylogenetic hypotheses are supported by the fact that lambeosaurines >show 
>greatest overall similarity with hadrosaurines among >non-lambeosaurines, 
>followed by iguanodonts and the Camptosaurus. >While this statement makes 
>intuitive sense, it lacks a theoretical basis. It >is not difficult to 
>construct an hypothetical example where such a >correspondence would not be 

Irrelevant. Yes it could have happened and it would have been interesting to 
try and extract that pattern from the data. But it didn't. Bad criticism, 

> While such a case would require "recognition" of some other >phenomenon (as 
> the authors state above), such a recognition will come >directly from the 
> analysis of derived similarity which produced the >original phylogenetic 
> hypothesis. The phylogenetic analysis, being >based on derived similarity, 
> would carry more weight than a phenetic >one. Unless the phenetic analysis 
> develops new characters not >considered in the previous phylogenetic 
> analysis, then that analysis >sufficiently "recognized" the convergence 
> before the phenetic analysis >began. Correspondence between morphometric and
>phylogenetic studies is a welcome occurrence, but it is by no means a
>requirement. It is unclear what theoretical basis the authors have for
>testing phylogenetic inferences using phenetic methods. 

Again, we never advocate not doing the standard phylogenetic analysis, just 
that morphological data, here morphometric analysis of the geometry of the 
heads, support certain reconstructions better than others. Period. Learn to 
read and not put words in authors' mouths.

>        "While RFTRA seems to be a powerful tool for describing shape
>changes, it may not be the best one for the purposes of this study. By
>isolating size and concentrating on shape variations, size-related >variation 
>and extreme allometric effects may have been eliminated >which might adversely 
>affect any measure of distance as applied to a >phylogenetic study. RFTRA may 
>thus confound evolutionary or >developmental interpretations of shape 
>differences, and may >exaggerate differences among taxa.

RFTRA is not the only method worth using. Right. Never said it was and I use a 
huge range of methods and always have. You cannot use everything in one paper 
and this was an exploration of the use of this one approach to this problem and 
the world is open to additional analyses, which we assumed were going to 
happen. Still needs to. We've been planning more allometric analyses for years 
of the postcranial regions but the problems with getting skeletal data that can 
be related to other units and composite skeletons, have made it a lot of work 
which I have yet to get to and had heard, continuously for more than a decade 
now, that others were doing, especially students of Jack's. They seem never to 
do it, however, and I regret never starting it. This is a lesson for us all. No 
matter who you hear is doing a project, never not do one you have planned 
yourself in anticipation of it. Most often, they never happen.

> For example, RFTRA output suggests that the difference between the prepubic 
> >processes of Edmontosaurus and Parasaurolophus (representative >hadrosaurine 
> and lambeosaurine taxa, respectively) are manifested in >increased depth in 
> the latter taxon. However, in the text it is explained >that the prepubic 
> process of Edmontosaurus is absolutely longer than >that of Parasaurolophus. 
> A visual evaluation of the outline drawings >provided will show that if the 
> two pubes were scaled to the same >acetabular dimensions, shape differences 
> in most
>parts of the bones would resolve themselves neatly. The primary >difference 
>between the pubes of the two taxa would then be >expressed as an increase in 
>the length of the prepubic process. The >RFTRA analysis might be useful in 
>determining the likely taxonomic >affinity isolated bones of the same size as 
>those used in the study, but >it is unclear whether this will extend to bones 
>of different sizes. A >similar situation is apparent in their analysis of the
>ischium, where the primary difference again appears to involve a >lengthening 
>(and possibly thickening, see below)of the shaft coupled >with distal 
>expansion. As before, the RFTRA analysis redistributes >these effects over all 
>landmarks, resulting in comparison which >appears less biologically useful. 
>While RFTRA is touted as being >capable of analyzing localized differences 
>among landmarks, it seems >unable to isolate localized variation. In the 
>process of finding a best fit, >better fits in one part of the landmark 
>configuration must be sacrificed >for the overall fit. Distance coefficients 
>from such analyses may thus >exaggerate overall differences if static 
>allometries among groups vary >significantly from isometry, as is likely the 
>case in most phylogenetic >contexts.

We are rather upfront about the lower success level of the postcranial analysis 
but wanted to try it. Rather than complain about the limitations of one 
technique - which still presented interesting results - the better approach is 
to shotgun the whole situation and apply various techniques, all with their own 
limitations and strengths. Again, allometric techniques do not keep the 
original geometry and, consequently, are also limited in providing data. they 
are far from the panacea that JW seems to think they are.

Oh, and I suggets you reread the descriptions of the algorithm. RFTRA 
specifically allows localized change and does not portion the variance out as 
you say. Least-squares methods do, although they are very useful as well. I 
suggest you get to know a techniques before you malign it. be bold enough to 
try it, actually, and find its usefulness.

>        "There are also inadequacies in the landmark configurations used >in 
> the study. The authors make repeated statements concerning the >utility of 
> their cranial study for phylogenetic purposes, as opposed to >the study of 
> the pelvic bones, due to the use of homologous landmarks. >However, casual 
> scrutiny reveals that 11 of 20 cranial landmarks >(numbers 1-4, 6, 10-13, 16, 
> 20) are in actuality pseudolandmarks, and >two (7, 8) are defined as a 
> pseudolandmarks although they may >represent true homologous points as they 
> are applied within the study. >Further, these landmarks are arrayed about the 
> lateral profile of the >cranium to characterize expected differences in 
> morphology among >hadrosaurids rather than quantify overall shape. The
>resulting geometric coverage of the form is good. It is unclear, >however, 
>whether the distance measurements arrived at truly reflect >overall similarity 
>among the hadrosaurids included in the study.

The landmarks chosen are as good as any landmarks used in such analyses. 
Experimental tests by Louis Roth have shown extreme points can be very useful 
and even more conservative than the triple junction type ones, which are more 
like tectonic plates than points. However, they provide useful landmarks. There 
are better ones in ostracodes that are extensions of the nervous system, but 
these are classic vertebrate landmarks. the standard measures used in 
allometric studies tend to use much less good ones - classic caliper 
measurements. This is an awful paragraph that seems to impress an instructor 
more than anything. The landmarks used are probably etter and more stable than 
that used to develop characters used in most phylogenetic analyses. Wrong again.

>       "The authors employ non-hadrosaurid taxa apparently as in parallel
>to the outgroup method of polarizing characters in phylogenetic studies.
>However the landmark configuration is designed to quantify variation >among 
>hadrosaurids, not among ornithopods in general, and may not be >optimal for 
>characterizing difference among all ornithopods. That >hadrosaurines appear 
>more similar to iguanodonts than lambeosaurines >may be due to the use of 
>distance measurements derived from such a >landmark configuration. Further, 
>landmarks are clustered around areas >of interest, and the combined effect of 
>their changes among >lambeosaurines and hadrosaurines may serve to overweight
>the differences between these two groups relative to the differences >between 
>hadrosaurines and other non-lambeosaurines.

May may May - get off your butt and prove it. Do the follow up paper to show 
that is the case. Otherwise you are just blowing smoke. The points concentrated 
in the non-changing parts of the head. You are wrong. Yes there were others, 
but "iguanodonts" also play in those areas as well. Wrong!

>      "In their interpretations of the results of the cranial analyses,
>the authors must repeatedly caution that similarities among
>non-lambeosaurines are partially due to the derived nature of the skull >of 
>lambeosaurines. This is manifested as very high RFTRA distances >between 
>lambeosaurines and non-lambeosaurines, due to the >tremendous differences in 
>positions of landmarks associated with the >premaxillary-nasal crest. In a 
>modern approach to discovering >phylogeny (i.e. phylogenetic systematics), 
>autapomorphic character >states are not useful in elucidating phylogeny. In a 
>morphometric study >aimed at elucidating the phylogenetic relationships among
>these groups, and specifically addressing the supposed diphyletic >nature of 
>the Hadrosauridae, it might be appropriate to select a landmark
>configuration which ignores the very derived nature of premaxillae and
>nasals of lambeosaurines. This might reduce the very high RFTRA >distances 
>between lambeosaurine and non-lambeosaurine, and >increase the low differences 
>among the lambeosaurines, allowing a >more balanced evaluation of overall 
>similarity. However, this would >affect the utility of the point configuration 
>in assessing overall >differences among the taxa studied. The effectiveness of 
>a particular >point configuration is dependent on the purpose of the study. 
>Since this >study had the twofold purpose of characterizing variation and 
>>evaluating phylogenetic hypotheses, the conclusions suffer from using >a 
>landmark configuration used to address both goals.

Yes, a phylogenetic analysis may yield different patterns. No news there. We 
did not do a phylogenetic analysis here, not the point. Read and understand. No 
paper solves all problems - this was one more unit of information useful for 
solving the long-term question of the relationships of these groups.

>        "The authors bemoan the lack of homologous points for >comparison of 
> the pelvic bones. They manage to identify several useful >pseudolandmarks on 
> the pubis and ilium. Here, as opposed to the >cranial study, their use of 
> pseudolandmarks is more understandable. >The landmark configuration on the 
> ischium fails to capture at least one >important difference among taxa, 
> namely the thicker shaft. The lack of >landmarks along the shaft of the 
> ischium may also account for the >peculiar configuration of the outline 
> drawings which resulted from the >analysis, and, as with the pubis, may have 
> unduly affected RFTRA >distance coefficients (see above). It seems unlikely
>that, with such a bewildering array of pseudolandmarks in the other >bones 
>studied, that some pseudolandmarks could not have been >produced to more 
>adequately model the ischium.

might might might. It's the job of the next study to answer these questions. 
Might is not particularly interesting here and I've already touched on this 

>        "Although the analyses in this study demonstrate that RFTRA is
>capable of isolating size to allow a pure shape comparison, this >comparison 
>is only useful insofar as the point configurations reflect the >actual shape 
>of the element.

You interpret results knowing your configurations and what you can do with 
them. period. Always the case, even in allometric analyses.

> It is unclear whether isolating size produces biologically
>meaningful information. Such an analysis might be useful for broadly
>comparing shapes of bones and groups of bones for classification, but >may not 
>be useful in evaluating a phylogenetic hypothesis. Further, >given a 
>phylogenetic hypothesis, it is unclear whether RFTRA will be >useful in 
>characterizing derived difference in morphology. RFTRA, and >the use of RFTRA 
>distance in dendrograms, is not an "assumption-free" >analysis. RFTRA distance 
>is only useful insofar as the landmark >configuration reflects the 
>requirements for solving the problem at hand. >Various size-dependent and 
>landmark-dependent phenomena might >obscure whatever relevant information is 
>contained in the forms. >RFTRA distance can be useful in providing some 
>measure of >evolutionary change among organisms. Utilizing this distance as a
>test of a phylogenetic hypothesis requires a theoretical foundation >which is 
>lacking in this study. Without both a theoretical basis and a 
> practical basis for using RFTRA to test a phylogenetic hypothesis, it >might 
> be best to restrict its role in phylogenetic analysis to developing 
> >characters, and characterizing changes among taxa.

Shape independent from size is of great interest. As is shape independent from 
size related back to size to see if size really has been removed, etc. When you 
find an assumption free algorithm for doing any analysis, let me know.

>        "The results of any such analysis must still be evaluated in a
>broader context, as they may not accurately reflect variation among
>specimens. It is clear from the above discussion that RFTRA should not >be the 
>only method used for the purpose of analyzing variation among >taxa. 

Sure, never would I say anything otherwise. This has been one of my main 
thrusts of my career. Can't do everything in one paper and we stated we were 
exploring this one method.

> Some form of orthogonal decomposition technique, for example >principle 
> component analysis, coupled with a landmark method to orient >distances and 
> allow reconstruction of forms for comparison, may prove >more capable of 
> isolating localized variation and analyzing similarity and >difference within 
> a more biologically informative size-shape >decomposition.

Actually, RFTRA and related techniques are the best way of looking at isolated 
change. Period. much better than PCA methods, which I use all the time and 
love, by the way. You don't know what you are talking about here.

> Using any morphometric technique to evaluate phylogenetic >relationships 
> requires a more solid foundation than that which is >provided here, and no 
> further suggestions can be made with regards to >an appropriate technique 
> until such a theoretical basis is established."

Actually, the foundation is finally catching up with us as better phylogenetic 
work is finally available. Exploratory analyses have their utility and this one 
did as well. hind sight jabs simply don't do this or other works justice. Lull 
& Wright applied D'Arcy Thompson grids to ceratopsians and generate neat info. 
I could state what is wrong with it with modern eyes but it is still an elegant 
study and one of my favorites. i'm not saying this paper was a classic but it 
was one of the first studies of its kind, and it has led to some interesting 

>        I suspect I spent a little too much time on this assignment...

Actually, I suspect you spent too little time since you didn't bother to get a 
good understanding of techniques other than the ones you knew, which you seem 
to overemphasize and not understand all that well either. You also didn't 
consider the paper historically. Papers for classes can be unsupported and 
incomplete because of time limitations and knowledge limitations. that doesn't 
mean they should be broadcast, especially in their totality, to the rest of the 
world. Go out and do the follow up if you like, I'm gonna get to it myself some 
day and I suspect many things will change and many patterns actually will hold 
up. note the hadrosauridae still exists and Jack's pretty much been beaten down 
on it. It is silly to throw out data such as this. Learn the methods and 
extract the good that they produce. 

Sorry for the length but this "review" frankly was distortive and simplistic 
and, although I would do the original paper very differently now with my 
current knowledge and new techniques, and better available phylogenetic work, 
it still did some good things. I hope others will think before dumping term 
papers onto the list. There was nothing new in this one, and a lot wrong, in my 
humble opinion.

Ralph Chapman, NMNH

Ralph Chapman