Hi Göran,
I am not sure I understand your point (2). I do not know what
you mean by “true” landmarks in the introductory paragraph, or actual
landmark locations in all dimensions, in point (2).
I assume location and orientation would necessarily vary across repeated
measures, or the repeated measures would be of little value. For my
purposes, ideally the ratio of measurement error standard deviation to
biological standard deviation for each landmark could be calculated. I can
understand how measurement error s.d. for each landmark can be calculated.
I do not know how the biological standard deviation at each
landmark can be calculated. The variances of the registered landmarks, as
far as I know, are not estimates of the variances about the population mean,
before registration and translation. For example, if the underlying
variance at one landmark is large in the raw data, when that data is
registered the variance of that landmark typically shrinks, while the
variance at other landmarks increases. How can the value of a landmark,
measurement variation over biological variation, be assessed without a good
estimate of biological variation?
Thank you for your reply,
Ben
----- Original Message -----
From: "morphmet" <[EMAIL PROTECTED]>
To: "morphmet" <[email protected]>
Sent: Wednesday, July 27, 2005 2:56 PM
Subject: Re: Repeatability
Ben - Provided that you have a set of objects, for each of which you have
recorded the same set of landmarks repeatedly, there are at least two
simlpe ways in which you can compare the impact of random measurement error
across "true" landmarks [following a Procrustes superimposition if location
varies across repeated measures].
1. In absolute terms: simply compare SD across repeated measures within
objects for the actual landmark data (in all dimensions). The smaller the
SD's, the better the landmark in terms of being less affected by random ME.
2. In relative terms: calculate the repeatabilities of actual landmark
data/locations (in all dimensions). The higher the repeatability, the
larger is true variation in landmark location across objects relative to
measurement error for that landmark.
Does this help?
/Göran Arnqvist
At 02:06 PM 7/27/2005, you wrote:
> I don't yet see a solution to the repeatability problem. While the
>repeatability at each landmark can be assessed at each landmark, the ratio
>of the repeatability to the biological variation at each landmark cannot be
>calculated using any method I have seen. The work done in this area has
>focussed on repeatability/biological-variation of shape variables (Arnqvist
>& Martensson, 1998). The ratio is not calculated for any specific
>landmarks. The landmarks are the measurements, so how can we know which of
>our measurements are good and not-so-good?
> Ben
>----- Original Message -----
>From: "morphmet" <[EMAIL PROTECTED]>
>To: "morphmet" <[email protected]>
>Sent: Thursday, July 21, 2005 12:17 PM
>Subject: Re: Repeatability
>
>
> >
> > I'm a bit late in jumping on the repeatability bandwagon, but for what
> > it's worth, regarding semilandmarks, allow me to throw in my ten cents'
> > worth.
> >
> > Their initial positions along the outline are arbitrary, but I assume
> > one could probably derive new coordinates for them after having "slid"
> > them in TPSRelw, and then treat them as true landmarks for the purposes
> > of repeatability.
> >
> > If one felt compelled to do so.
> >
> > That would, however, not be statistically valid since their final
> > positions will be determined by the positions of those terminal
> > landmarks which anchor them.
> >
> > Repeatability of the semilandmarks will therefore be related to
> > repeatability of the terminal landmarks, and my personal feeling is to
> > not worry about the semilandmarks too much, but take very special care
> > in minimising digitising error in terminal landmarks.
> >
> > > Lets see, take a x number of individuals, make three repeated
> > > measurements, preferably by different people, and estimate the
> > > proportion of the variation that is explained by individual versus all
> > > the available variation. Either landmark by landmakr, of all at once
> > in
> > > a MANOVA.
> > >
> > > For semi landmarks, no idea, never worked with those.
> > >
> > > Good luck.
> > >
> > > Kim
> > >
> > > morphmet wrote:
> > >
> > >> I was simply wondering what would be the best way, the most
> > appropriate
> > >> statistical test, to evaluate the repeatability of your landmark
> > >> placement.
> > >>
> > >> Furthermore, would this method also work with semi-landmarks?
> > >>
> > >>
> > >> Thanks
> > >>
> > >> Martin
> > > --
> > > Replies will be sent to the list.
> > > For more information visit http://www.morphometrics.org
> > >
> > >
> >
> >
> >
> >
> >
> > ________________________________________________________________________
> > This message and attachments are subject to a disclaimer. Please refer
> > to http://www.it.up.ac.za/documentation/governance/disclaimer/ for full
> > details. / Hierdie boodskap en aanhangsels is aan 'n vrywaringsklousule
> > onderhewig. Volledige besonderhede is by
> > http://www.it.up.ac.za/documentation/governance/disclaimer/ beskikbaar.
> > --
> > Replies will be sent to the list.
> > For more information visit http://www.morphometrics.org
> >
>
>--
>Replies will be sent to the list.
>For more information visit http://www.morphometrics.org
********************************************************************
Associate Professor Göran Arnqvist
Animal Ecology
Department of Ecology and Evolution
Evolutionary Biology Centre
University of Uppsala
Norbyvägen 18d
SE - 752 36 Uppsala
Sweden
Email: [EMAIL PROTECTED]
Phone: +46-18-471 2645
Fax: +46-18-471 6484
New! Homepage with complete PDF downloads and more at:
http://www.ebc.uu.se/zooeko/GoranA/GA.html
********************************************************************
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
