----- Forwarded message from a.vanhete...@roehampton.ac.uk -----
Date: Fri, 26 Apr 2013 04:05:17 -0400
From: a.vanhete...@roehampton.ac.uk
Reply-To: a.vanhete...@roehampton.ac.uk
Subject: RE: Linear models for cranial variability
To: morphmet@morphometrics.org
Dear Milos
Just my two cents, in case you do opt for regression. In any regression you
might want to do, I would use habitat types as independent variables and PC
scores (or any other measure of shape) as dependent variables. Regressions are
very asymmetrical and the analysis you describe assumes that habitat types vary
according to skull shape, whereas the reverse (skull shape adapting to various
habitats) is a much more realistic assumption, and will give different results.
Also, I would suggest using all your data, rather than just sample means (12
data points, really is not a lot). If the reason you used population means is
that the relationship between habitat and skull shape is different between the
various populations, may I suggest you consider a regression analysis pooled
per species. This analysis retains all the data (so all the information), but
allows for a certain degree of difference between populations (i.e., the slope
of the regression is the same for all, but the intercept is allowed to vary per
population).
Lastly, don't try to force the data to fit your hypothesis. You may not have
had much luck with regressions so far, because cranial shape may not actually
be dependent on habitat, but on something else. Or perhaps sexual dimorphism or
allometry is so strong that it obscures any shape-habitat signal. You may want
to experiment with correcting for allometry (can be done by regressing your
data onto some measure of size and continuing your analyses with the residuals)
or separating males and females (either in the same analysis by pooling or in
separate analyses).
I hope this helps somewhat.
Best wishes,
Anneke van Heteren
________________________________________
From: morphmet_modera...@morphometrics.org
[morphmet_modera...@morphometrics.org]
Sent: 25 April 2013 22:45
To: morphmet@morphometrics.org
Subject: RE: Linear models for cranial variability
----- Forwarded message from Milos Blagojevic <spearsata...@hotmail.com> -----
Date: Wed, 24 Apr 2013 08:34:17 -0400
From: Milos Blagojevic <spearsata...@hotmail.com>
Reply-To: Milos Blagojevic <spearsata...@hotmail.com>
Subject: RE: Linear models for cranial variability
To: "morphmet@morphometrics.org" <morphmet@morphometrics.org>
Thanks to Pere and Carlo,
I haven`t thought about PLS or correlation/covariance matrix approach to this
problem at all. Maybe you could provide some papers that utilize such
methodology for correlating cranial dimensions with environmental variables?
I have tried this approach and always there is a low value of the Rv
coefficient.
The only approach I had success with was Beta-regression (Dirichlet
generalization) when using proportions of habitat types (forest, meadow, plow)
as dependent and PC scores (mean per population) as independent variables.
Still I don`t know if this is ok and have failed at finding some reference
papers.
Best regards,
Milos
> From: morphmet_modera...@morphometrics.org
> To: morphmet@morphometrics.org
> Subject: Re: Linear models for cranial variability
> Date: Tue, 23 Apr 2013 20:53:14 -0700
>
>
> ----- Forwarded message from carlo.mel...@unina.it -----
>
> Date: Tue, 23 Apr 2013 04:01:24 -0400
> From: carlo.mel...@unina.it
> Reply-To: carlo.mel...@unina.it
> Subject: Re: Linear models for cranial variability
> To: morphmet@morphometrics.org
>
> Dear Milos,
>
> you can try using Partial Least Square that allows to look at
> correlation between one block of variables (cranial dimensions) and
> the second block of variables (environmnetal variables). Make sure you
> standardize the variables (e.g. for cranial dimension it would be good
> using log transformation of measurements and for environmental data
> try to standardize by subtracting mean so that data values are not too
> disparate or large).
>
> Alternatively, if you want to make predictions you can perform a
> multiple multivariate regression or a Generalised Least Square model.
> However, they have more assumption dealing with multivariate data
> normality while PLS has not.
> You can do PLS using the current version of the free software PAST
> that has a user friendly interface. For multiple multivariate
> regression and Generalised Least Square NTSYS or SPSS or specific
> scripts in R.
>
> All the best
>
> Carlo Meloro
>
> morphmet_modera...@morphometrics.org ha scritto:
>
> >
> >
> > ----- Forwarded message from Milos Blagojevic -----
> >
> > Date: Mon, 22 Apr 2013 15:12:47 -0400
> > From: Milos Blagojevic
> > Reply-To: Milos Blagojevic
> > Subject: Linear models for cranial variability
> > To: "morphmet@morphometrics.org"
> >
> > Dear Morphometricians,
> > Drifting a little bit from the field of GM I have a question about
> > the formulation of a linear (or possible any other) model that has
> > to account for cranial variability in relation to certain
> > ecological parameters.
> > My dataset consists of 50 linear measurements taken on roe deer
> > skulls from 12 populations. After PCA and optional discriminant
> > analysis I have individual scores that should enter possible linear
> > model as dependent variables. Ecological data consists of
> > proportions of forest to meadow to plowland areas (expressed either
> > as proportions that add up to 1 or as absolute areas in Ha) within
> > every population and population density (individual/area or
> > absolute numbers). Any ideas on what kind of a model could be
> > suitable for this dataset and for testing the hypothesis that
> > cranial dimensions are predicted by these independent variables
> > (habitat structure and abundance or population density)?
> > Best regards,Milos BlagojevicDepartment for Biology and
> > Ecology,Faculty of Science,Kragujevac,Serbia
> > Here is sample data (with absolute numbers but they can be expressed
> > as proportions as well)
> > PCx score population abundance forest plow meadow -0.6033788
> > ADA_BEC 1500 61154 12000 32313 0.3250981 ADA_BEC 1500
> > 61154 12000 32313 0.5577059 ADA_BEC 1500 61154
> > 12000 32313 -0.1596194 PM 23980 89499 579870 8178
> > -1.3089952 PM 23980 89499 579870 8178 -2.1693392 SP
> > 2500 38000 47098 432432 -0.9669080 SP 2500
> > 38000 47098 432432 -1.8857842 SP 2500 38000 47098
> > 432432 0.7242678 DKN 65908 181133 12400 1233
> > 1.6815373 DKN 65908 181133 12400 1233
> >
> > ----- End forwarded message -----
> >
> >
> >
> >
>
> ----- End forwarded message -----
>
>
----- End forwarded message -----
Consider the environment. Please don't print this e-mail unless you really need
to.
This email and any attachments are confidential and intended solely for the
addressee and may also be privileged or exempt from disclosure under applicable
law. If you are not the addressee, or have received this e-mail in error,
please notify the sender immediately, delete it from your system and do not
copy, disclose or otherwise act upon any part of this email or its attachments.
Internet communications are not guaranteed to be secure or virus-free.
University of Roehampton does not accept responsibility for any loss arising
from unauthorised access to, or interference with, any Internet communications
by any third party, or from the transmission of any viruses.
Any opinion or other information in this e-mail or its attachments that does
not relate to the business of University of Roehampton is personal to the
sender and is not given or endorsed by University of Roehampton.
University of Roehampton is the trading name of Roehampton University, a
company limited by guarantee incorporated in England under number 5161359.
Registered Office: Grove House, Roehampton Lane, London SW15 5PJ. An exempt
charity.
----- End forwarded message -----
