[R-sig-eco] Select distribution and link in GAM

[Gavin Simpson]

On Mon, 2011-04-11 at 16:02 -0500, ARISTIDES LOPEZ wrote:
> Hello Gavin,
> 
> Firts that all forgive my English. Thank you very much for you explanation.
> I decide to utilise GAM (and not GLM) by their ability to deal with non
> linear relationships between data. Zuur et al (2007) advise to use  poisson
> distribution when utilize cuont data, binomial distribution when use
> presence -absence data and you advise me for gamma distribution. At this
> point my question is how can evaluate which distribution is better to my
> model? exist any test or plot ?
> 
> Can anyone help me?

You need to learn more about GLMs. You choose the error distribution on
the basis of what type of response and therefore errors you have.

Poisson and Bionomial are for discrete data, and length is *not*
discrete, so these would seem inappropriate. As I mentioned in my last
message, a Gamma GLM/GAM would be a reasonable starting point as your
response variable, length, is continuous and strictly positive (you
can't have a zero length fish).

G

> thanks a lot!
> 
> 
> 2011/4/11 Gavin Simpson <gavin.simpson at ucl.ac.uk>
> 
> > On Mon, 2011-04-11 at 11:18 -0500, ARISTIDES LOPEZ wrote:
> > > Hello,
> > >
> > > I want make a GAM to describe the factors that influence the length
> > > distribution of  demersal fishes in the Colombian Caribbean Sea. The
> > model
> > > is like this:
> > > gam(length~s(Lat)+s(Long)+s(depth)+s(temperature)+as.factor(rayni
> > > season)+as.factor(botton),data=B. capriscus). My question is how can
> > select
> > > the type of distribution and the link of my model?
> >
> > You *did* read ?gam, yes?
> >
> > >From there we have:
> >
> >  family: This is a family object specifying the distribution and link
> >          to use in fitting etc. See ‘glm’ and ‘family’ for more
> >          details. A negative binomial family is provided: see
> >          ‘negbin’.
> >
> > And that points us to ?glm and ?family, so from ?glm:
> >
> >  family: a description of the error distribution and link function to
> >          be used in the model. This can be a character string naming a
> >          family function, a family function or the result of a call to
> >          a family function.  (See ‘family’ for details of family
> >          functions.)
> >
> > which also tells us to look at ?family for details, so we do, and we
> > find all the information we need.
> >
> > Even a look at the examples for gam() in ?gam would have shown you what
> > was involved.
> >
> > Or did you mean something else by "select"? In which case, I think you
> > should think about what form the response takes: it is continuous but
> > bound above zero as you can't have a zero length fish. In other words
> > the response is strictly positive and continuous. A Gamma GLM would seem
> > to be an appropriate starting point in that case.
> >
> > HTH
> >
> > G
> >
> > --
> > %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
> >  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
> > %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
> >
> >
> 
> 

-- 
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
 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
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%