Re: MANOVA with 3D landmark data

[morphmet_moderator]

----- Forwarded message from Dean Adams -----

Date: Tue, 26 Jun 2012 08:54:21 -0400
From: Dean Adams
Reply-To: Dean Adams
Subject: Re: MANOVA with 3D landmark data
To: morphmet@morphometrics.org

Lia,

Done properly, Procrustes coordinates projected orthogonally to tangent space, PC scores from these, and the set of relative warp scores will all yield identical statistical results using MANOVA; once dimensions with zero variation have been eliminated, or if generalized inverses are utilized. They will simply be orthogonal rotations of one another in tangent space.  See e.g., Rohlf 1999 J. Class. for a demonstration of this.

Given this, 'summary(manova(lm(y~x)))' in R will be suitable for your MANOVA on the set of shape variables (again, with the dimensions with zero variance eliminated).

Dean

--
Dr. Dean C. Adams
Associate 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 6/25/2012 5:41 AM, morphmet_modera...@morphometrics.org wrote:

----- Forwarded message from lia betti -----

Date: Wed, 20 Jun 2012 09:14:57 -0400
From: lia betti
Reply-To: lia betti
Subject: MANOVA with 3D landmark data
To: "morphmet@morphometrics.org"

      

        

          
Dear all, 
          
 
          
I'm trying
            to test for sexual dimorphism in shape in different
            populations, and I
            would like to do a MANOVA to test for significant effects of
            sex, of
            population, and the interaction between the two variables. I
            am using 3D
            landmark data, and I realise that the Procrustes
coordinates
            I
            obtain after Procrustes superimposition are in a
            non-euclidean space, and
            therefore unsuitable for normal MANOVA.  I read that I can
            use relative
            warps as data to input in my MANOVA, and solve the problem
            this way, but I would
            prefer to use either the Procrustes variables, or the PC
            scores obtained after
            projecting the Procrustes variables on a euclidean space
            (because I then proceed
            to use Procrustes variables to estimate indices of sexual dimorphism). 
          
My
            questions are:
          
1) Can I
            apply a normal
            MANOVA on the PC scores, given that they are in Euclidean
            space? I haven't found
            any paper or book mentioning this method, and I don't
            understand why.
          
2) I think
            I could in
            principle use Goodall's F test as an alternative to MANOVA
            on the Procrustes
            variables, but MorphoJ specifies that the Procrustes
ANOVA 
            run by the
            program is suitable only for studies of asymmetry or
            inter-observer error. Again, I am not sure why. In R, the
            function for this test is again mentioned
            only in relation to studies of fluctuating asymmetry. Is
            there a
            software I can use to run this test to check for significant
            effects of sex and
            population?
          
 
          
Thanks!
          
 
          
Lia
          
email:
            liabe...@yahoo.it

        
      
      

      

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

      

    

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

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