get_starve_distributions: List of starve model options

In lawlerem/staRVe: Spatio-Temporal Analysis of Research surVEy data

List of starve model options

Description

Print a list of implemented response distributions, link functions, and covariance functions.

Usage

get_starve_distributions(which = c("distribution", "link", "covariance"))

Arguments

which

A character vector containing "distribution", "link", and/or "covariance". Defaults to all.

Value

A named character vector giving the implemented distributions, link functions, and/or covariance functions, depending on the value of the 'which' parameter.

Response Distributions

gaussian -

Linear predictor determines the mean, has a variance parameter.

t -

Linear predictor determines the mean, has a degrees of freedom parameter.

gamma -

Used for model strictly positive continuous data such as biomass, linear predictor determines the mean, has a variance parameter.

lognormal -

Used for modelling strictly positive continuous data such as biomass, linear predictor determines the mean, has a variance parameter.

tweedie -

Used to model non-negative continuous data with point mass at 0, linear predictor determines the response mean before applying the offset. The offset term is a multiplier to the response mean and can be supplied in the model formula through the sample.size(...) term.

poisson -

Typically used for count data, linear predictor determines the intensity parameter.

negative binomial -

Typically used for over-dispersed count data, linear predictor determines the mean, has an overdispersion parameter (>1 results in overdispersion).

compois -

Used to model over- and under-dispersed count data, linear predictor determines the mean, has a dispersion parameter. A dispersion <1 results in under-dispersion, while a dispersion >1 results in over-dispersion.

bernoulli -

Used to model yes/no or presence/absence data, linear predictor determines the probability of a yes.

binomial -

Used to model the # of yesses in k yes/no trials, linear predictor determines the probability of a yes in a single trial. The sample size k for each observation can be supplied in the model formula through the sample.size(...) term.

atLeastOneBinomial -

Used to model the probability of observing at least one yes in k yes/no trials, linear predictor determines the probability of a yes in a single trial. The sample size k for each observation can be supplied in the model formula through the sample.size(...) term.

Link Functions

identity

log

logit

Covariance Functions

exponential -

Matern covariance with smoothness nu = 0.5. Results in a non-differentiable Gaussian random field.

gaussian -

Limiting function of the Matern covariance as smoothness nu approaches infinity. Results in smooth (infinitely differentiable) Gaussian random field. The variance parameter in this covariance function gives the marginal variance of the spatial field.

matern -

Matern covariance with arbitrary smoothness nu. Results in Gaussian random fields that are floor(nu) times differentiable. The smoothness parameter is hard to estimate, so fixing it to a set value is recommended. This option will run much slower than other specific values of the Matern covariance function (e.g. exponential).

matern32 -

Matern covariance with smoothness nu = 1.5. Results in a smooth (once differentiable) Gaussian random field.

lawlerem/staRVe documentation built on Oct. 9, 2024, 4:48 a.m.