Dear Prof David Warton
Thanks a lot for your nice introspection on my data. I appreciate your
valuable comments. I am also trying to explore gamm or VGAM to match its
suitability with data. Its fine. However, I am thinking to reduce my data
structure by removing some of the species showing interspecific
correlation. Honestly speaking I do not have thought of it. Can you please
give more insights regarding this (interspecies correlation). I am also
interested in studying species-environment relationship (not by CCA or RDA).
Your kind comments are highly appreciated.
With Best Regards
Rajendra M Panda
School of Water Resources
Indian Institute of Technology Kharagpur, India
On Wed, Feb 18, 2015 at 4:36 AM, David Warton <david.warton at unsw.edu.au>
wrote:
Hi Rajendra and Greg,
A couple of quick thoughts:
Firstly, Rajendra the method that is applicable to your data really
depends on the research question - what is it that you are trying to
achieve. It is always hard to offer help on what analysis method is suited
to a question without knowing the original research objective. The gamm
function for example might be useful to you if you are primarily interested
in predictive modelling, and also if you think that you have a common
nonlinear response to environmental variables with some "noise" around this
pattern for different spp (which can be represented as random effects).
You could alternatively use this function to fit a separate smoother for
each spp but that would be a pretty complicated model and few would have
sufficient data to justify that level of model complexity. VGAM y Thomas
Yee offers and option in between these two.
Secondly, something you need to worry about with this type of data is
interspecies correlation - for various reasons (including species
interaction), it is widely thought and even better often observed that
species are correlated in abundance (or presence/absence, whatever) even
after accounting for environmental predictors. This makes the problem
multivariate. If you care about making joint inferences across species and
you don't account for correlation between species you can get things quite
wrong. The gamm function I think could handle residual correlation, but
not the way you specified it, and it would have a lot of trouble, unless
you have only a handful of species and quite decent abundance data on
each. On the other hand if you are just making predictions separately for
each spp then you don't need to worry too much about this.
All the best
David
David Warton
Professor and Australian Research Council Future Fellow
School of Mathematics and Statistics and the Evolution & Ecology Research
Centre
The University of New South Wales NSW 2052 AUSTRALIA
phone (61)(2) 9385-7031
fax (61)(2) 9385-7123
http://www.eco-stats.unsw.edu.au/ecostats15.html
[[alternative HTML version deleted]]