----- Forwarded message from morphmet_modera...@morphometrics.org -----
Date: Tue, 01 Apr 2014 22:06:14 -0700
From: morphmet_modera...@morphometrics.org
Reply-To: morphmet_modera...@morphometrics.org
Subject: Re: Group differences
To: morphmet@morphometrics.org
----- Forwarded message from "Cabo-Perez, Luis" <lc...@mercyhurst.edu> -----
Date: Fri, 21 Mar 2014 20:13:37 -0400
From: "Cabo-Perez, Luis" <lc...@mercyhurst.edu>
Reply-To: "Cabo-Perez, Luis" <lc...@mercyhurst.edu>
Subject: Re: Group differences
To: "<morphmet@morphometrics.org>" <morphmet@morphometrics.org>
Patrick, if I am interpreting your question correctly (why do I find
differences in the canonical roots but not in the raw data and PCA's), a
possible explanation is that the differences that CVA is using to separate the
groups are mostly related to size. In DF and CVA you are asking the analysis to
find out the variables that maximize group differences, and in living organisms
size is often the main difference. The Procrustes coordinates and PC's are
expressing shape (including the part of shape that is explained by allometry),
so it is possible that individuals of the same size display a very similar
shape independently on the group to which they belong. Something that the PC's
and Procrustes would convey better than the canonical roots if the latter are
separating the groups mostly according to allometric size.
The number of canonical roots extracted would give you a clue: if the number is
low, the explanation above would be pretty likely (e.g., if a single root
explains a very large chunk of the variance). If that is the case the first
canonical roots should also be highly correlated with centroid size.
Hope that I am not too off-base and this helps,
LC
> Date: Mon, 10 Mar 2014 08:59:14 -0400
> From: Patrick Arnold <patrick.arn...@uni-jena.de>
> Reply-To: Patrick Arnold <patrick.arn...@uni-jena.de>
> Subject: Group differences
> To: morphmet@morphometrics.org
>
> Dear morphometrics,
>
> my data contains 5 a priori determined groups. I did CVA in MorphoJ
> but I know CVA not necessarily reveals significant groups. Thus, I
> want to verify my groups statistically. Manova with CVs in Past reveal
> significant groups, comparison between groups in MorphoJ and Past,
> too. But, if I only use Procrusted coordinates or PCs, Manova in Past
> says there are no significant different groups. What do theses results
> say about my groups. Is significance with CVs in Manova enough to say
> that the 5 groups are really different in shape??
>
> Patrick Arnold
>
> Institut für Spezielle Zoologie und Evolutionsbiologie
> mit Phyletischem Museum
> Friedrich-Schiller-Universität Jena
> Germany
----- End forwarded message -----
----- End forwarded message -----
