On Sun, 2011-11-13 at 11:31 -0600, Stephen Sefick wrote:
> Vegan (on CRAN) may be of help. Particularly look at the ordistep,
> ordiR2step etc.
But do note the warnings that as these models don't really have a log
likelihood and hence the don't have a deviance nor AIC. The AIC
implemented in vegan uses the method of:
Godínez-Domínguez, E. & Freire, J. (2003) Information-theoretic
approach for selection of spatial and temporal models of community
organization. _Marine Ecology Progress Series_ *253*, 17-24.
Read Jari's warnings in ?deviance.cca regarding its usage.
ordiR2step uses the forward selection method of:
Blanchet, F. G., Legendre, P. & Borcard, D. (2008) Forward
selection of explanatory variables. _Ecology_ 89, 2623-2632.
which employs and adjusted R^2 criterion.
Note that any form of forward selection applied to these multivariate
methods is just as likely to be subject to all the problems of stepwise
selection methods familiar to the application of linear regression. It
would be helpful if we could combine these ordination methods with the
concept of shrinkage (e.g. the lasso) so that selection could be
performed in a single step *and* the effects of selection be taken into
account. (There has been some progress in this regard in some
[non-ecological] parts of the literature.)
Or, better still, think before fitting the model and only include those
terms that you wish to test that correspond to the hypotheses you wish
to test.
HTH
G
> On Sun 13 Nov 2011 01:25:46 AM CST, David_Hewitt wrote:
> >
> > On Sat, 12 Nov 2011, Michel Rapinski<mrapi...@uottawa.ca> wrote:
> >>
> >> Hello,
> >>
> >> There is a function in R's basic library (stats), step(), which allows
> >> step by step selection of variables (forward, backward, both) on multiple
> >> linear regression models based on AIC scores.
> >>
> >> Unfortunately, and correct me if I am wrong, it only works for lm,
> >> aov and
> >> glm models.
> >
> >
> > The package AICcmodavg handles many other types of linear models. It's
> > on CRAN.
> >
> >>
> >> In the case of selecting variables for canonical analysis,
> >> more specifically redundancy analysis (RDA), are there functions that
> >> enables these same test on rda models? I figured that since RDA is
> >> basically a multivariate extension of the multiple linear regression, it
> >> should work, but no luck!
> >
> >
> > There are important differences between ordination and linear models.
> > Beyond
> > that, the issue of selecting important variables is far more complex than
> > just an automated routine to search through them for "significance"
> > (of any
> > kind). Patrick's recommendation to have a look at the book by Burnham and
> > Anderson is a good one -- start with pp 84-85 and section 4.4 (pp 167 ->).
> >
> >>
> >> I have succesfully managed to use the forward.sel() function in
> >> library(packfor), for selecting variables in my multivariate RDA models,
> >> but I also wish to do backward and alternating selection to help in the
> >> selection of my variables.
> >>
> >> Help will be greatly appreciated.
> >>
> >> Michel
> >>
> >> Michel Rapinski, MSc. Student
> >> Inst. of Plant Biology Research, Montreal Botanical Garden
> >> Université de Montréal
> >> Montréal, QC H1X 2B2
> >> Tel: 514.772-1710
> >> Fax: 514.872.9406
> >> michael.rapin...@umontreal.ca
> >> University of Ottawa
> >> mrapi...@uottawa.ca
> >>
> >>>
> >>> Hi Minda,
> >>>
> >>> AIC scores depend upon the statistical models used. I think R does the
> >>> best job of providing these scores, for example in the context of
> >>> multiple
> >>> linear regression and generalized linear models.
> >>>
> >>> The literature on R or on stats using R is growing rapidly. You will
> >>> find
> >>> readable treatments of AIC in Crawley's 2007 R book or in Zuur et
> >>> al'sv2009 Mixed Effects Models and extensions in Ecology with R.
> >>>
> >>> And do not forget to examine ( I am not sure read is a realistic option)
> >>> the valuable book by Burnham and Anderson 2002, Models Selection and
> >>> Multimodel Inference.
> >>>
> >>>
> >>> Patrick Foley
> >>> bees, fleas, flowers, disease
> >>> patfo...@csus.edu
> >>> ________________________________________
> >>> From: Minda Berbeco [mberb...@gmail.com]
> >>> Sent: Wednesday, October 26, 2011 8:32 AM
> >>> To: ECOLOG-L@LISTSERV.UMD.EDU
> >>> Subject: [ECOLOG-L] AIC scores
> >>>
> >>> Hello,
> >>>
> >>> I am looking for recommendations for programs to use for calculating AIC
> >>> scores. I've looked into the AICcmodavg package with R, but the
> >>> associated
> >>> instructional material is not clear and I have not been able to get
> >>> it to
> >>> work. I hear that SAS is good as well, but have not found a good book
> >>> that
> >>> tells me how to create AIC scores (recommendations would be
> >>> appreciated).
> >>> I've also looked into SPSS, which according to IBM can create AIC
> >>> scores,
> >>> but have had no success.
> >>>
> >>> Any recommendations for programs and clear associated instructional
> >>> material
> >>> with information on how to run the program, write the code etc. would be
> >>> greatly appreciated.
> >>>
> >>> Thanks,
> >>>
> >>> Minda Berbeco
> >>> Viticulture and Enology, UC Davis
> >>> mrberb...@ucdavis.edu
>
--
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
Dr. Gavin Simpson [t] +44 (0)20 7679 0522
ECRC, UCL Geography, [f] +44 (0)20 7679 0565
Pearson Building, [e] gavin.simpsonATNOSPAMucl.ac.uk
Gower Street, London [w] http://www.ucl.ac.uk/~ucfagls/
UK. WC1E 6BT. [w] http://www.freshwaters.org.uk
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
