I fear, Kieran, that you're right: papers are still published with
per-landmark analyses. There's even an R package that "provides the code
... to calculate per-landmark rate and variance".
Per-landmark analyses produce a good looking numerical artifacts. I am
not hopeful that this will stop very soon, but at some point it will. It
was the same with the analyses of partial warps 25 years ago: after Jim
conclusively showed they were wrong, papers went on being published for
a while but, now, luckily, it's very rare.
At least, with partial warps the mistake was more understandable: it was
the early days of Procrustes-GMM and the mathematics of those variables
is much more difficult for biologists. The error was common and made
even by some of the best morphometricians. With per-landmark analyses,
in contrast, the issue is so obvious that it's been known since before
the time of modern GMM.
The problem is that scientists are interested in the questions
apparently addressed by per-landmark analyses, and these analyses
produce results and good looking visualizations, but none of those make
sense: for meaningful results, one must first demonstrate how a LEAST
SQUARE superimposition, such as Procrustes (plus, maybe, sliding
semilandmarks by minimizing bending energy, "a metaphor borrowed for use
in morphometrics from the mechanics of THIN METAL PLATES"
https://sbmorphometrics.org/glossary/gloss1.html) can be used to
estimate variance of individual points in biological tissues. None has
done it, to my knowledge, and that's because it can't be done.
Claims that the approach is justified because it produces results
congruent with previous analyses are invalid for two simple reasons: 1)
at best, they demonstrate precision but precision is not accuracy (i.e.,
the correct answer); 2) they lack external validity (i.e., even if they
got the right answer for a specific dataset - which is hard to
demonstrate if at all possible - that cannot be generalized to other
datasets).
The editors of PNAS, Nature, Plos etc. cannot be competent on everything
and finding competent reviewers is hard, which is probably why several
of those wrong and misleading analyses are published in prestigious
journals.
In fact, I doubt we're really interested to know if a single point wins
the prize for the most variable or fast evolving. We may want to know if
a region of a bigger structure is more variable or 'evolvable', but we
can't use single points as a proxy to get the answer. Subsetting
landmarks after a common superimposition ('within a configuration'
approaches) has the same logical issue of per-landmark analyses, because
results (including the visualization) depend on the choice of the
superimposition. That's something it took me much longer to understand,
but probably others had already got it and it's what Jim meant (Jim,
correct me if I am wrong) in his comment some time ago in morphmet: “in
addition to trying to assess variation at a single landmark, there is
also a problem with … covariation among landmarks … These limitations
call into question some methods used in studies on the very popular
topics of integration and modularity. Geometric morphometrics is an
excellent tool but it cannot do the impossible. The fact that one can
compute interesting statistics does not mean that they are reasonable.”
Cheers
Andrea
On 17/05/2022 08:20, [email protected] wrote:
Dear all,
Dr. Andrea's thread "analysis of landmarks one at a time in Procrustes
shape data" has been recently expanded into a paper: Procrustes Shape
Cannot be Analyzed, Interpreted or Visualized one Landmark at a Time. Evol
Biol (2022). https://doi.org/10.1007/s11692-022-09565-1.
Recently, I find landmarks are still being analyzed one at a time in Nature
Genetics.
The paper (Attached):
Genetic variants underlying differences in facial morphology in East Asian
and European populations. Nat Genet. 2022;54(4):403-411.
Legend for Fig 4f reads "P values of each quasi-landmark". Although I have
limited understanding in genetics, it seems very likely that landmarks were
analyzed one at a time. *So my question 1 is if I have misinterpreted
anything such that landmarks were not analysed separately in this paper. *
[image: image.png]
Best,
Kieran
--
Dr. Andrea Cardini
Researcher, Dipartimento di Scienze Chimiche e Geologiche, Università di
Modena e Reggio Emilia, Via Campi, 103 - 41125 Modena - Italy
tel. 0039 059 4223140
Adjunct Associate Professor, Centre for Forensic Anthropology, The
University of Western Australia, 35 Stirling Highway, Crawley WA 6009,
Australia
E-mail address: [email protected], [email protected]
WEBPAGE: https://sites.google.com/view/alcardini2/
or https://tinyurl.com/andreacardini
--
You received this message because you are subscribed to the Google Groups
"Morphmet" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/morphmet2/f76120ff-3f79-770c-f316-80191cdf83af%40gmail.com.
