[R-sig-eco] A beginner's question to constrained ordinations with vegan

[Stephen Sefick]

Tim,

I would take a look at Numerical Ecology with R. This book may not 
address your particular question, but should be useful as a general 
reference for using R for quantitative ecology.

Some questions:
Are you interested in what is structuring the community along an 
environmental gradient? What is the rational for investigating the 
sub-gradients?

My own curiosity:
Is there a literature source with quantitative data demonstrating that 
particular ordinations "better" uncover true environmental/distance 
relationships?

A couple of comments (please correct my misunderstandings):
RDA and PCA followed by envfit will give different results because they 
are doing very different things. From my understanding, rda uses the 
predicted value matrix from a multivariate regression of Comm_Mat ~ 
Env_Mat and then preforms a PCA on the resulting matrix (mean value 
given the environmental predictors; constrained). A PCA on the 
(appropriately hellinger transformed?) Comm_Mat is unconstrained by the 
environmental variation and projects sites along the direction of 
maximum variance in Comm_Mat only. Therefore, these techniques will give 
very different results.

I hope that helps, and my explanation is not very far from the reality 
of the techniques.

kindest regards,

Stephen

On 10/09/2014 07:26 AM, Tim Richter-Heitmann wrote:
> Hi there,
>
> i have a typical ecological problem (modelling abiotic parameters to
> bacterial abundances - i have 9 of these explanatory variables (but also
> a variety of spatial and biotic parameters, who may serve as
> explanators), many bacterial species and hundreds of sites).
>
> My species gradients seem to be very long in the DCA, so i began my
> analysis with CCA modelling all 9 abiotic parameters to the species
> matrix, and using the triplot as a final result.
>
> However, i have two very distinct bacterial communities in the DCA with
> a huge gap on the x-axis between them (one community is defining 90% of
> all samples, and the smaller one is found in 10% of the samples), so i
> was fiddling around with performing rda's
> (which i believe is recommended for small species gradients) on the two
> subsets.
>
> Now, a colleague was actually recommending me to use unconstrained
> ordinations like PCA and use envfit to fit the explanatory variables later.
>
> ord.OTU <- rda(OTU)
> ef <- envfit(ord.OTU, Env, perm=999)
>
> instead of
>
> ord.OTU <- rda(OTU~., Env)
>
> However, i fail to grasp the ideas and differences behind and between
> the two approaches - in my case, an envfitted PCA looked different than
> the "equivalent" RDA. As far as i have been taught, constrained
> ordination techniques like RDA or CCA search for the best explaining
> variables in the direct gradients, so i would use those for problems
> like mine per default. So, what are the benefits in using the
> unconstrained techniques first?
>
> Since i am new to the field, i lack the experience to evaluate this. Any
> advice would make me a very happy student.
>
> Thank you very much, and my apologies if i have asked something that was
> asked many times before. In fact, i tried to find the answer online, but
> wasnt too successful.
>
>
>

-- 
Stephen Sefick
**************************************************
Auburn University
Biological Sciences
331 Funchess Hall
Auburn, Alabama
36849
**************************************************
sas0025 at auburn.edu
http://www.auburn.edu/~sas0025
**************************************************

Let's not spend our time and resources thinking about things that are so 
little or so large that all they really do for us is puff us up and make 
us feel like gods.  We are mammals, and have not exhausted the annoying 
little problems of being mammals.

                                 -K. Mullis

"A big computer, a complex algorithm and a long time does not equal 
science."

                               -Robert Gentleman