Dear all,
as an addendum to my previous message on tests for multivariate
normality, I've just noticed that a new R package for this (MVN by
Korkmaz et al 2014 -
http://journal.r-project.org/archive/2014-2/korkmaz-goksuluk-zararsiz.pdf ) is
out.
Best,
Carmelo
--
Carmelo Fruciano
Marie Curie Fellow - University of Konstanz - Konstanz, Germany
Honorary Fellow - University of Catania - Catania, Italy
e-mail [email protected]
http://www.fruciano.it/research/
Carmelo Fruciano <[email protected]> ha scritto:
Patrick Arnold <[email protected]> ha scritto:
Dear morphometrics,
I have two questions about applying MANOVA on shape data:
1. MANOVA assumes the data to be normal distributed. What is the
best way to test normal distribution of multivariate shape data?
And what is the best (free) software for this issue (is this
embedded in PAST)?
Dear Patrick,
I normally use other software. However, it looks like PAST has some
option for this at page 100 of the manual. On the other hand, you
might want (also given the small sample sizes you mention below) to
use resampling-based approaches...
2. The differences between my groups are very distinct in the CVA.
This is quite normal and well known (i.e. the fact that CVA shows
often very clear separation of the groups). A detailed discussion of
this can be found in
Mitteroecker, P. and F. Bookstein (2011). "Linear Discrimination,
Ordination, and the Visualization of Selection Gradients in Modern
Morphometrics." Evolutionary Biology 38(1): 100-114.
Most importantly, Mitteroecker & Bookstein suggest using
between-group principal component analysis as an alternative
ordination technique. This is gaining popularity over time and, in
my experience, it has been useful already in multiple cases
(Franchini, Fruciano et al. 2014 - Molecular Ecology; Fruciano et
al. 2014 - Biological Journal of the Linnean Society).
I want to test whether these differences are statistically
significant, too. As my sample size only slightly exceeds the
degrees of freedom, I want to apply the MANOVA onto the CVs (i.e.
decrease in the number of dependent variables). Is this possible or
is it redundant?
Although I have seen it done, I don't think this is a very good
idea, as you have already suspected. Again, maybe resampling-based
approaches might be a better choice (while, as Andrea Cardini was
suggesting in some previous post, also recognizing the limitations
of one's analysis).
I hope this helps,
Carmelo
--
Carmelo Fruciano
Marie Curie Fellow - University of Konstanz - Konstanz, Germany
Honorary Fellow - University of Catania - Catania, Italy
e-mail [email protected]
http://www.fruciano.it/research/
--
Carmelo Fruciano
Marie Curie Fellow - University of Konstanz - Konstanz, Germany
Honorary Fellow - University of Catania - Catania, Italy
e-mail [email protected]
http://www.fruciano.it/research/
--
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].
