HOF: Hierarchical logistic regression

In eHOF: Extended HOF (Huisman-Olff-Fresco) Models

Hierarchical logistic regression

Description

Fit seven hierarchical logistic regression models and select the most appropriate model by information criteria and a bootstrap approach to guarantee model stability. The first five shapes are known as Huisman-Olff-Fresco (HOF) models in ecology (Huisman et al. 1993). Additionally the package provides two bimodal shapes.

Usage

	## Default S3 method:
HOF(occ, grad, M = max(occ), y.name, family=binomial, lim=100, 
  bootstrap=100, test=c('AICc', 'BIC', 'AIC','Dev'), 
  modeltypes=eHOF.modelnames, ...)
	## S3 method for class 'data.frame'
HOF(veg, grad, M, freq.limit = 10, ...)
	## S3 method for class 'HOF'
pick.model(object, level = 0.95, test = c("AICc", "BIC", "AIC", "Dev"), 
  modeltypes, penal = 'df', gam = FALSE,   selectMethod =   c('bootselect.lower', 
  'bootselect.always', 'IC.weight', 'pick.model'),  silent = FALSE, ...)
	## S3 method for class 'HOF'
fitted(object, model, ...)
	## S3 method for class 'HOF'
predict(object, model, newdata, ...)

Arguments

occ	species occurrences (=response) vector.
veg	vegetation data frame. Either as matrix with species in columns and plots in rows, or in Turboveg format.
grad	gradient data vector.
M	maximum attainable value in the HOF model, similar to binomial denominator.
y.name	name of the species.
family	error distribution. Alternatives are binomial, poisson and gaussian.
lim	limiting criterion for minimization function, see details.
bootstrap	number of bootstrap samplings to check model robustness, see details.
freq.limit	lowest frequency of species analysed.
object	HOF model object, results from function HOF. Either for a single species or a list for several species.
level	probability for model selection (1-P) for F or Chisq tests.
test	information criterion for model selection. Alternatives are "AICc", "BIC", "AIC" or just "Dev"iance.
modeltypes	vector of model types, when only a subset of the seven implemented modeltypes should be considered.
penal	penalty term for model types, default is the number of model parameter.
gam	calculate AIC of GAM model and compare it with HOF models.
selectMethod	choose the model type selection method in case of divergence of the most frequent bootstrapped model or the one with highest information criterion weight from the primary chosen type.
silent	messages about selectMethod
newdata	vector of gradient values for prediction.
model	You can specify the HOF model type to be used, otherwise it is selected through function pick.model.
...	further arguments passed to or from other methods

Details

The minimization function changed compared to package gravy (Oksanen 2002) from nlm to nlminb to be able to set a limit for estimated parameters (default= -100 to 100). The old models III and V have been often too sharp, lim=Inf will produce results similar to gravy. Function pick.model finds the most adequate modeltype according to the chosen Information Criterion (AICc is default). Function fitted returns the fitted values for the used gradient, and predict for any values in newdata.

To improve and check model stability a bootstrapping mechanism is implemented in function HOF. If the initially chosen model type is different from the most frequent one, the latter will be chosen by default. Bootstrapping is done with sample(length(grad), replace = TRUE).

Value

HOF.fit returns an object of class "HOF" which contains the parameters of the call, the fitting results for every model type and a vector of chosen model types from bootstrapping.

Author(s)

Florian Jansen, Jari Oksanen

References

Jansen, F. & Oksanen, J. (2013) What shape are species responses along ecological gradients? - Huisman-Olf-Fresco models revisited. Journal of Vegetation Science, DOI: 10.1111/jvs.12050

Oksanen, J. & Minchin, P.R. (2002). Continuum theory revisited: what shape are species responses along ecological gradients? Ecological Modelling 157, 119-129.

Huisman, J., Olff, H. & Fresco, L.F.M. (1993). A hierarchical set of models for species response analysis. Journal of Vegetation Science 4, 37-46.

See Also

plot.HOF provides advanced plotting schemes for HOF models. Para derives model parameters like optimum, niche (width), slope etc.

Examples

	data(acre)
	sel <- c('MATRREC', 'RUMEACT', 'SILENOC', 'APHAARV', 'MYOSARV', 'DESUSOP', 'ARTE#VU')
	mo <- HOF(acre[,match(sel, names(acre))], acre.env$PH_KCL, M=1, bootstrap=NULL)
	mo

eHOF documentation built on April 23, 2022, 1:05 a.m.