[R-sig-eco] parametric estimators for species richness in R

[Jari Oksanen]

On 2/12/10 11:36 AM, "yangwenjing" <yangwenjing at ibcas.ac.cn> wrote:

> 
> Dear all,
> 
> I am doing some work about species richness estimation. Nonparametric
> estimation (such as Chao1, Jackknife1) can be done just using function
> "specpool()" and "estimateR()" in package "vegan". The problem is that I can
> not find any function for parametric estimation (such as MMMeans, MMruns,
> Michaelis-Menten). Do you know any function for doing this? Thanks a lot.
>
Howdy Wenjing,

I don't know what these are except Michaelis-Menten that is used in enzyme
kinetics. From this I gather that you may want to fit a regression on
species accumulation against number of individuals or sampling units.  Some
of these non-linear regression models may have a parameter that can be
interpreted as the asymptotic (total, extrapolated) number of species. If
this is the case, you can either directly use nonlinear regression models
(function nls) which has inbuilt selfStart models for some cases, like
Michaelis-Menten (SSmicmen). See appropriate help pages ?nls, ?selfStart,
?SSmicmen. 

The development version of vegan in http://vegan.r-forge.r-project.org/ has
function fitspecaccum() which is able to fit some of those models of species
accumulation against number of sampling units. The R-Forge version of vegan
has also some selfStart models used in species--area modelling (Arrhenius,
Gleason, Gitay, Lomolino) which can be used in fitspecaccum() alternatively
to the standard R models, or which can be used in nls(). I am just preparing
a new release of vegan (1.17-5), but I am not sure if these models will be
included in the release. However, you can try them in the devel version.

I am not too keen to release them, since I really do not like the idea of
nonlinear model shopping. Nonlinear regression models are used when you must
use them, and you must use them when you know that the real physical model
is nonlinear, and you want to estimate the parameters of that physical model
(like in enzyme kinetics). There must be a theoretical reason to use a
specific nonlinear model. There seems not be such a reason in species--area
or species--sample-size models nor a theory on which a particular model is
based. It seems that people just shop around and try various arbitrary
models developed for other problems and then pick one for random reasons
(like happening to be cute with this particular data set). Therefore I'd
like to think a bit before I release these functions: there is a danger
people may use them...

Cheers, Jari Oksanen