----- Forwarded message from Dean Adams <[email protected]> -----

     Date: Fri, 17 May 2013 10:23:48 -0400
      From: Dean Adams <[email protected]>
      Reply-To: Dean Adams <[email protected]>
      Subject: Re: test for antisymmetry
      To: [email protected]

Adrien,

As far as I am aware, antisymmetry has only been applied to univariate 
data. Typically, antisymmetry is identified by taking the differences 
between the R and L measures (R-L) and plotting these for a set of 
specimens as a histogram.  When this histogram is bimodal about the 
origin, antisymmetry is inferred. This is often accompanied by 
additional statistical tests of the properties of this distribution. 

For GM shape data, since shape is multi-dimensional, the analogous 
procedure is not at all straightforward. One reason is that differences 
in shape between R and L objects are represented as distances, not 
difference scores (because there is more than one dimension to shape 
space). Thus, the distribution of R vs. L shapes will only have positive 
values, rendering the standard statistical approaches to antisymmetry 
not informative. 

A second reason the concept of antisymmetry is not easily generalized to 
multivariate data is that it requires some a priori axis upon which 
deviations can be defined as 'positive' or 'negative'.  For univariate 
traits, this is easy: by using (R-L), negative values mean L is bigger 
than R, and the converse for positive values.   But how does this extend 
to multivariate shape data? I'm not sure that it does.  Think of the 
following example. If antisymmetry is present in GM data, one 
possibility is that there are two 'clusters' of shapes in shape space, 
where for each object, its R & L shapes are found in different clusters, 
but where across objects some R shapes are found in cluster 1 while 
other R shapes are found in cluster 2. In other words, the 'deviations' 
between R & L are always present, but go in different directions when 
comparing among specimens. I believe this would represent antisymmetry 
in a multivariate context, but identifying this pattern is rather ad-hoc 
and descriptive (i.e., one must look for clusters, and then see whether 
they correspond to R & L shapes, etc.). Of course, since GM shape space 
is highly dimensional, there must be other ways that the concept of 
antisymmetry could be satisfied. 

That brings us back to the a priori axis. If one could identify some 
directional axis in shape space upon which R vs. L differences could be 
projected to identify  '+' and '-' values, it is possible that one could 
infer antisymmetry from this, in a manner analogous to that used for 
univariate data. However, I know of no mathematical or biological theory 
that could be leveraged to identify this axis, so that the univariate 
concept of antisymmetry could be extended to multivariate data (though 
admittedly I've not thought long and hard about this issue). 

If others are aware of any GM implementations of antisymmetry I would 
very much like to know of them. 

Hope this is helpful. 

Dean

-- 
Dr. Dean C. Adams
Professor
Department of Ecology, Evolution, and Organismal Biology
Department of Statistics
Iowa State University
Ames, Iowa
50011
www.public.iastate.edu/~dcadams/
phone: 515-294-3834

On 5/17/2013 1:27 AM, [email protected] wrote:
> ----- Forwarded message from [email protected] -----
>
>       Date: Thu, 16 May 2013 08:42:42 -0400
>        From: [email protected]
>        Reply-To: [email protected]
>        Subject: test for antisymmetry
>        To: [email protected]
>
> dear morphometricians,
>
> would you know how could I test my data for the presence of antisymmetry 
> using morphoJ and R?
> I have already places 32 landmarks on all my skulls and performed procrustes 
> fit followed by a procrustes anova. 
>
> I know that I'm supposed to test my data for skewness and kurtosis but I 
> can't figure out how to concretely do that... 
>
> I was thinking of exporting my asymmetric components from morphoJ and use the 
> "moments" package from R which contains skewness and kurtosis functions. 
> But I'm not sure this is a valid method. 
>
> Thank you very much for your advices
>
> Adrien, (a young morphometrician in progress)
>
> ************************************************************
> *///////////////////// Ph.D. Student //////////////////////*
> ************************************************************
> *////////////////// Université de Liège ///////////////////*
> *//Bâtiment B22; Laboratoire de génétique des populations /*
> *////////// boulevard du Rectorat 27; 4000 Liège //////////*
> *//////////////////////// Belgique ////////////////////////*
> *//////////////// Tél. ULg : +32 4 3662130 ////////////////*
> ************************************************************
> *///////////////////// McGill University //////////////////*
> */ Department of Biology; W3/19 Stewart Biology Building //*
> *// 1205 Ave Docteur Penfield; Montreal, H3A 1B1 Canada ///*
> *////////////// Office: +1 514 398 5965 ///////////////////*
> ************************************************************
>
> ----- End forwarded message -----
>
>

----- End forwarded message -----


Reply via email to