GO: Unconstrained Gaussian Maximum Likelihood Ordination

In jarioksa/GO: Gaussian Ordination and Community Simulation

Unconstrained Gaussian Maximum Likelihood Ordination

Description

The functions fit unconstrained maximum likelihood ordination with unit-width Gaussian response models.

Usage

GO1(comm, tot = max(comm), freqlim = 5, parallel = 1, trace = TRUE, ...)

GO(comm, k = 1, tot = max(comm), freqlim = 5, family = c("poisson",
  "binomial"), far = 10, init, trace = TRUE, ...)

metaGO(comm, k = 1, trymax = 3, firstOK = TRUE, trace = FALSE, ...)

## S3 method for class 'GO'
plot(x, choices = 1, label = FALSE, marginal = FALSE,
  cex = 0.7, col = 1:6, ...)

## S3 method for class 'GO'
anova(object, ...)

spanodev(mod1, mod2 = NULL, ...)

## S3 method for class 'GO'
predict(object, newdata, type = c("response", "link"), ...)

## S3 method for class 'GO'
calibrate(object, newdata, ...)

Arguments

comm	Community data frame.
tot	Total abundance used in Binomial models. This can be either a single value for all data, or a vector with value for each row of comm. The default is to use the maximum value in matrix.
freqlim	Minimum number of occurrence for analysed species.
parallel	Number of parallel processes.
trace	logical; print information on the progress of the analysis.
k	Number of estimated gradients (axes).
family	Error distribution. Can be either "poisson" for quasi-Poisson or "binomial" for quasi-Binomial (and must be quoted).
far	Threshold distance for species optima regarded as alien and frozen in fitting.
init	Initial configuration to start the iterations. This should be a matrix, and number of rows should match the community data, and number coluns the number of gradients (k). The default is to use scores from decorana.
trymax	Maximum number of random starts.
firstOK	Do not launch random starts if default start succeeds.
x	Fitted model.
choices	The axis or axes plotted.
label	Label species responses.
marginal	Plot marginal responses or slice through origin of other dimensions.
cex	Character size for labels.
col	Colours of response curves.
object	Ordination result object.
mod1, mod2	Compared result objects
newdata	New gradient locations to predict species responses or new community data to calibrate gradient locations.
type	Predictions in the scale of responses or in the scale of link function.
...	Other parameters passed to functions. In GO these are passed to nlm and can include, e.g., iterlim (which often must be set to higher value than the default 100).

Details

Function is under development and unreleased. It will be released under different name in vegan. The current version is only provided for review purposes. The function and its support functions require vegan, although this requirements is not explicitly made. The optimization is based on nlm function, and passes arguments to this function.

Function anova can analyse two nested models or a single model against null model of flat responses using parametric tests based on quasi-Likelihood. Function spanodev performs similar test split by species. Function predict returns estimated response curves, and newdata can be gradient locations. Function calibrate returns estimated gradient locations, and newdata can be community data.

The plot function displays fitted respose curves against one ordination axis. In principle, the ordination can be rotated using vegan function MDSrotate, but this requires a version that agrees to analyse GO results. Traditional ordination plots of SU scores and species optima can be displayed with ordiplot (vegan package). The function is written so that several other vegan and standard R functions can handle results.

Functions

• GO1: Alternating estimation of species parameters and gradient locations in one dimension.

• GO: Simultaneous estimation of species parameters and gradient locations.

• metaGO: Start GO several times from random configurations if default start fails or optionally always

• spanodev: Comparison of goodness of fit for individual species.

Author(s)

Jari Oksanen

See Also

cgo in VGAM package.

Examples

library(vegan) ## *must* be given before using the function
data(varespec)
mod <- GO(varespec, k=2, far=5, tot=100, family="binomial", iterlim=1000)
plot(mod, label=TRUE)
ordiplot(mod, type="t")
ordiplot(mod, dis="si", type="t")
anova(mod)
mod1 <- update(mod, k=1)
anova(mod1, mod)
spanodev(mod1)
spanodev(mod1, mod)

jarioksa/GO documentation built on April 2, 2024, 8:37 p.m.