----- Forwarded message from andrea cardini <[email protected]> -----
Date: Fri, 25 Jan 2013 03:25:43 -0500
From: andrea cardini <[email protected]>
Reply-To: andrea cardini <[email protected]>
Subject: Re: Replicates for error checking when digitizing landmarks
To: [email protected]
Dear Sive,
in my view, option 3 is better. That's what we
did in Viscosi & Cardini
(http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0025630)
on all leaves we were measuring and what I did
also when working on 3D data from monkeys. The
protocol for monkeys is exploratory (not a test)
and should be described here: Cardini A., Elton
S., 2008 - Does the skull carry a phylogenetic
signal? Evolution and modularity and in the
guenons. Biological Journal of the Linnean
Society, 93: 813–834. As you will see, because we
were working on more than 1300 specimens, which
are actually a subsample of the ca. 4000 I
measured, we assessed error in a small subsample.
That was a bit sloppy but probably OK for a study
where we were concerned with fairly large differences (species/genera).
I would probably now try to use more individuals
(not all of them, however!) and select
representatives of the main groups I am going to
compare. Using Klingenberg's ANOVA approach, you
are able to partition variance and test different
levels of variation. Viscosi and I applied a
simplified version of Chris' approach and did not consider symmetry/asymmetry.
Besides Chris' approach, probably the other test
of error which I really like is in a paper by the
Viennese people. It's somewhat related to the
ANOVA idea, if I remember well, as it is also
relating the error to the effect being studied. I
think it's the one on dental arch asymmetry
(AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY
129:132–142 (2006)). I am not sure if they
assessed only digitizing error or digitizing plus
repositioning, but it can be clearly extended to
do that. They also, I believe, worked on a subsample.
However, as you seem to be very rigorous,
consider that even approach 3 is not testing the
2D to 3D approximation. If you wanted to do that,
you should have landmarks on images but also
their replicas measured with a 3D digitizer such
as the Microscribe. I suspect that you could even
build a design in your replicas (in a subsample)
so that you can assess all those different
sources of error: digitizing, repositiong, 2D to
3D. Overall, I'd tend to say that what really
matters is that the total error is negligible.
Partitioning sources of errors might help to
detect if there's a level which is particularly problematic.
Another ref. where you might find some ideas on
these issues is the one below. There, we were
interested in comparing measures from 3D
reconstructed skulls to measurements directly
taken on the real bones using calipers. We also
assessed inter-operator errors. As in the Viscosi
& Cardini paper, we tried to provide an example
dataset for people to be able to replicate the
protocol. Compared to V&C, we used a resampling
method (permutational ANOVA), which is likely to
be better than a parametric test (resampling
makes less stringent assumptions). The reprint is
not yet available in my webpage but I can send it
to you together with the example data.
Franklin D., Cardini A., Flavel A., Kuliukas A.,
Marks M.K., Hart R., Oxnard C., O’Higgins P.
Concordance of traditional osteometric and volume
rendered MSCT interlandmark cranial measurements.
International Journal of Legal Medicine, in
press: DOI: 10.1007/s00414-012-0772-9.
Good luck!
Cheers
Andrea
At 20:17 24/01/2013, you wrote:
>----- Forwarded message from Sive Finlay <[email protected]> -----
>
>Date: Wed, 23 Jan 2013 05:27:24 -0500
>From: Sive Finlay <[email protected]>
>Reply-To: Sive Finlay <[email protected]>
>Subject: Replicates for error checking when digitizing landmarks
>To: [email protected]
>
>Dear all,
>
>I am studying morphological convergent evolution
>in the skull and limb shapes of tenrecs compared
>to other small mammals. I have approximately 100
>species and aim to both measure and photograph
>10 representatives of each species. I have a
>query relating to the optimal way to check for
>error when digitizing landmarks on my photographs.
>
> From my reading of other studies, I've come up
> with 3 possible ways to approach error checking;
>
>1) Take one photo of each specimen, make copies
>of that one image, digitize the same landmarks
>on each copy and then measure the variability
>between these landmarks to estimate digitizing error.
>
>2) Take multiple photos of each specimen (is 3
>enough to aim for in this case?) and use these
>to digitize landmarks separately on each image
>
>3) Take multiple photos of each specimen but
>re-position the skulls/ limbs between every photo
>
>The aim of options 1 and 2 would be to account
>for error in my own ability to be consistent in
>digitizing the landmarks. In contrast, any
>landmark variation with the option 3 approach
>would be a combination of both digitizing skill
>error and discrepancies associated with how the
>specimens are positioned for the photographs
>(e.g. if they're not completely flat in the plane of focus).Â
>
>The issue with this 3rd option would be that
>it's far more time consuming since each specimen
>would have to be positioned and photographed on
>3 separate occasions while options 1 and 2 only
>require me to set up each picture once.
>
>Given that my aim is to collect 1,000 specimens
>(10 replicates of 100 species), I would be very
>grateful for any advice on which error checking
>approach is practical or necessary. In addition,
>any modified or alternative suggestions for
>error-checking approaches would be very much appreciated.
>
>Thank you!
>
>Sive
>--
>Sive Finlay
>IRC EMBARK Initiative Postgraduate ScholarÂ
>
>Macroecology and Macroevolution Research Group
>Zoology building
>School of Natural Sciences
>Trinity College Dublin
>
><mailto:[email protected]>[email protected]
>http://www.tcd.ie/Zoology/research/ncooper/people.php
>
>
>----- End forwarded message -----
>
>
----- End forwarded message -----
