On Nov 3, 2005, at 7:44 AM, morphmet wrote:

> Good morning morphometricians
>
> ...
> c) Again when comparing groups within a sample, is there some way of
> evaluating which are the landmarks that contribure more to the observed
> differences?-

Antigoni et al.,

I will respond to question (c), which is of most interest to me.  It is 
clearly a multivariate analysis-of-dispersion question which can be 
answered by analysis of the aligned data using an 
analysis-of-dispersion package.  This would include developing a design 
matrix that would  include how your groups within a sample are related.

You would then analyze the Y matrix of aligned data using the design 
matrix producing a Sums of Squares and Crossproducts from that analysis 
which would have SS and Crossproduct matrices due to each design 
feature, plus a Residual Sums of Squares and Crossproducts.  Then  
these would be submitted to the Test-for-Additional Information.   For 
instance, one could ask the question:

Does trait 1x,y  provide any additional information above and beyond 
trait 2-10x,y?

or

Do traits 1-5x,y  provide any additional information above and beyond 
traits 6-10x,y?

This powerful Analysis of Dispersion approach is described in Rao 
(1965).

Rao, C.R., Linear Statistical Inference and Its Applications. 1965, New 
York NY: John Wiley. 522pp.

I have implemented a package that uses this approach over the years in 
each subsequent language that I have been using: Fortran, Pascal, APL, 
J and now R.   I have abandoned most of the prior language 
implementations and now only have sure access to the R version.

I provide a set of scripts which work in the J environment that were 
completely functional under the J computing environment current in 2002 
but I have not maintained it .  This J version script is reasonably 
annotated with what it will do:

http://www.bio.umass.edu/biology/kunkel/pub/j_scripts/

I use the R version of the Analysis of Dispersion part of it regularly 
but can not vouch for the Test-of-Additional-Info module yet.

http://www.bio.umass.edu/biology/kunkel/pub/r_scripts/andy/

  As with much personal software, it may not be friendly enough for 
general use by individuals unfamiliar with the math but people are 
welcome to play with it and modify it to their own needs.

They are a simple implementation of the matrix algebra from Rao.

I have never seen them implemented in another commercial package (I 
really have not looked) and would ask our present company of 
morphometricians if they have another approach or available 
implementation that would do the same trick?  We should all not keep 
patting ourselves on the back for reinventing the wheel but it is 
a-good-thing that some of us keep checking that we can do the basic 
math -and- help others learn to do the relevant stuff also.

This is not something that Jim's TPS suite will do and I know that he 
usually recommends that people take the aligned data and analyze it 
with available special packages.  Should we urge some sort of matrix or 
tree of techniques be created that would allow us to direct new users 
to a protocol or key to be followed out from the basic alignment 
procedure ... what things can be done by the TPS or other suites and 
what needs require looking toward other packages?

How about a concept map of the tools that can be used to answer 
morphometrics questions?

Joe Kunkel
[EMAIL PROTECTED]


> --·.  .·  `·.  .><((((º>`·.  .·  `·.  .><((((º>`·.  .·  `·.  .><((((º> 
>  .··.· >=-           =º}}}}}><
Joseph G. Kunkel, Professor
Biology Department
University of Massachusetts Amherst
Amherst, MA 01003
http://www.bio.umass.edu/biology/kunkel/

-- 
Replies will be sent to the list.
For more information visit http://www.morphometrics.org

Reply via email to