On 6.2.2015 22:54, Eugenia Leone Gold wrote:
Hi everyone,
I am trying to run a non-parametric test for group mean differences using PC
scores and I can’t seem to find a function to do that. I am using PC scores
instead of Procrustes Coordinates so as to have fewer variables than specimens.
Does anyone know of a function that can do this? I came across the function
adonis in Vegan, but that seems to use distances so I don’t think it’s
applicable to this situation. Thanks for any help!
Best,
Eugenia
Hi Eugenia,
If you want (just) to make a statistical inference regarding difference
between two shape means, you can use permutation test. You can do it in
MorphoJ, geomorph, PAST...All of them are distance based.
If you want to perform hypothesis testing that describe patterns of
shape variation and covariation in regard to some factor(s), solution is
to perform appropriate linear model. However, regardless whether
dependent variables are original procrustes residuals or PC scores of
them, small sample size may be a problem in traditional (parametric)
approaches. There is recent paper by Collyer and Adams published in
Heredity
(http://www.nature.com/hdy/journal/vaop/ncurrent/abs/hdy201475a.html)
which demonstrated efficacy of non-parametric approach that is not
constrained by low sample size compared to number of variables. That
approach can evaluate effect size of factors in the model even if sample
size is small relative to number of variables. Procedure is described
and justified in the paper and you can perform it in the geomorph
package (procD.lm function).
Regarding non-parametric tests, as far as I know, most of them use
distance based methods. E.g. see here:
http://www.entsoc.org/PDF/MUVE/6_NewMethod_MANOVA1_2.pdf
Hope this help.
All the best.
--
Marko Djurakic,
Teaching assistant,
Faculty of Sciences
Department of Biology and Ecology
University of Novi Sad
Trg Dositeja Obradovića 2
21 000 Novi Sad
Serbia
PhD student at Faculty of Biology,
University of Belgrade
Studentski trg 16
11000 Belgrade
Serbia
e-mail: [email protected]
--
MORPHMET may be accessed via its webpage at http://www.morphometrics.org
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
