----- Forwarded message from adrien.an...@doct.ulg.ac.be -----
Date: Wed, 17 Apr 2013 17:02:18 -0400
From: adrien.an...@doct.ulg.ac.be
Reply-To: adrien.an...@doct.ulg.ac.be
Subject: Fa analyses protocol
To: morphmet@morphometrics.org, "Virginie Millien, Dr."
<virginie.mill...@mcgill.ca>
Dear morphometters,
My name is Adrien ANDRE,
I have recently started working on my Ph.D. in co-direction between the
university of Liège (Belgium) and the McGill University (Canada).
My first goal is to quantify the stress level (using fluctuating asymmetry) in
different mice populations, from places where the species is present for many
years, toward the border of the species range in order to know if populations
from the range edge have a higher stress lever than the others.
I have trouble with my statistical analyses and I haven't found anyone in my
direct environment able to help me, so I'm turning to you!
I've read many things about FA statistics but there are few things that are
still dark in my mind, especially, how to compute the different analyses using
morphoJ.
I'll write here the statistical protocol I thought about, and It would be nice
if you could help me with the aspects that are still dark...
Here is the question I'd like to answer:
Do the FA levels vary between populations depending on how far they are
situated from the species range edge?
I have about 220 skulls that have been photographed on their ventral side.
I placed 31 landmarks on each pictures, using TPS software. (28 paired
landmarks).
Here are the things that I need to check before computing any FA scores.
1 : Perform procrustes Fit with all my pictures.
(I'm not sure wether I should align by principal axes, or use my two extreme
unpaired landmarks...)
2 : check for the presence of any outliers
3 : measurement error:
Because I'd rather not position landmarks twice on each picture, I'd like to
measure the error with a subsample of my pictures.
thus, positions of landmarks were captured twice on 25 random skulls in order
to estimate the error measure related to the landmark positioning.
One skull has been digitized 10 times and landmarks were then positioned in
order to measure the error related to the digitizing.
Here is what I did on MorphoJ :
-import the 60 pictures
-procrustes fit
-check for outliers
-in the procrustes anova tab:
I have created two classifiers :
the first one (name) is the id of the skulls. Thus, it is the same for the
replicants
the second one (repet) corresponds to the replicant number.
Here is how it looks like :
name-repet
skull1 -1
skull1-2
skull2-1
skull2-2
..
skull25-1
skull25-2
skull26-1
skull26-2
skullx-3
..
skull26-10
Here are the results :
<<
Classifiers used for the Procrustes ANOVA:
Individuals: name
Error 1: repet
Centroid size:
Effect SS MS df F P (param.)
Individual 1,456924 0,060705 24 1088,57 <.0001
Error 1 0,001840 0,000056 33 0,22 0,9817
Residual 0,000505 0,000252 2
Shape, Procrustes ANOVA:
Effect SS MS df F P (param.)
Individual 0,03426891 0,0000492369 696 14,61 <.0001
Side 0,00023053 0,0000079493 29 2,36 <.0001
Ind * Side 0,00234620 0,0000033710 696 5,96 <.0001
Error 1 0,00108309 0,0000005659 1914 -0,21 NaN
Residual -0,00031671 -0,0000027302 116
Shape, MANOVA tests of effects:
Symmetric component of shape variation:
Effect Pillai tr. P (param.)
Individual 21,90 <.0001
Note: the test for 'Individual' used the symmetric component of Error 1 as the
'error' effect.
Asymmetry component of shape variation:
Effect Pillai tr. P (param.)
Ind * Side 19,17 <.0001
>>
Because I was not sure if I'm allowed or not to mix the 25*2 replicants and
the 1*10 replicants, I performed another procrustes anova without the ten
replicants of the same skulls. Thus, in the next analysis results, there are
only the 25 skulls and their replicants. (50 indiv) included in the analysis.
here are the results :
<<
Classifiers used for the Procrustes ANOVA:
Individuals: name
Error 1: repet
Centroid size:
Effect SS MS df F P (param.)
Individual 1,266177 0,052757 24 1034,54 <.0001
Error 1 0,001275 0,000051 25
Shape, Procrustes ANOVA:
Effect SS MS df F P (param.)
Individual 0,02574950 0,0000369964 696 12,13 <.0001
Side 0,00021630 0,0000074587 29 2,45 <.0001
Ind * Side 0,00212228 0,0000030493 696 6,09 <.0001
Error 1 0,00072640 0,0000005010 1450
Shape, MANOVA tests of effects:
Symmetric component of shape variation:
Effect Pillai tr. P (param.)
Note: the test for 'Individual' used the symmetric component of Error 1 as the
'error' effect.
Asymmetry component of shape variation:
Effect Pillai tr. P (param.)
>>
Here are my questions relative to this analysis:
Is it right to mix the 25*2 replicants and the 1*10 replicants in the same
analysis? did I do it right?
In morphoJ, it's possible to use 2 classifiers for two different kind of errors
(imaging and digitizing). How should I proceed in my case?
In the case where this analysis would have been well conducted, what would be
the exact procedure in order to know if additional analyses of asymmetry
variation are justified?
According to Palmer paper " Fluctuating-asymmetry analysis: A step-by-step
example".
MS from Ind*side should be higher than MS from Error.
I haven't read anything like that for MorphoJ , so I'm still wondering if I can
use that simple method in the upper results.
4-directional asymmetry
I know that I have to test my results for presence of directional asymmetry.
I have read that this DA is quantified by the "side". In both analysis, Side
effect is smaller than Ind*side effect. Does this mean that directional
asymmetry can be neglected for the next FA analyses?
5-antisymmetry
I know that I'm also supposed to test for the presence of antisymmetry, but I
don't know If I can (and how) test this in morphoJ.
6-FA scores
Once those preliminary analyses will have be done, I'll have to get the FA
scores for each individuals.
Here is the way i think I should do:
-start a procruste fit for my 220 skulls.
-perform a procrustes anova.
-export the procrustes FA scores.
include this FA score in my table in order to relate this variable with others!
I'd like to thank you in advance for your precious help.
Adrien
adrien.an...@ulg.ac.be
************************************************************
*///////////////////// 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 -----
