bootstrap: Bootstrapping for GAI models

In calliste-fagard-jenkin/GAI: Generalised Abundance Index for Seasonal Invertibrates

Bootstrapping for GAI models

Description

Produces either a non-parameteric bootstrap by refitting the GAI at each iteration, or produces a parametric resampling of the MLEs at each iteration using an estimate of their asymptotic normal distribution.

Usage

bootstrap(
  GAI_fit,
  R = 100,
  refit = T,
  alpha = 0.05,
  parallel = T,
  cores = NULL,
  transform = T
)

Arguments

GAI_fit	An object produced by using fit_GAI for model fitting
R	The number of resamples to produce for the bootstrap
refit	If TRUE, resamples the observations from the sites and occasions uniformly, and refits the model at each bootsrap iteration. If FALSE, the bootstrap simply resamples the fitted parameter values from their asymptotic normal distribution (therefore this option requires a Hessian to have been produced during the model fitting stage).
alpha	1 - alpha gives the coverage the bootstrap confidence intervals aim to produce.
parallel	if TRUE, calculates the bootstraps in parallel, using the maximum number of available cores.
cores	If not NULL, this specifies the number of cores to use, if the default of using all available cores is undesirable.
transform	if TRUE, will return a bootstrap on the transformed parameters, rather than on the link scale. It should be noted that when covariates are present in the model, transformed outputs are averaged across all covariate values in the data, to avoid erroneous interpretation of covariate effects on the transformed scale.

Value

A named list with entries "EC", "N", "A" and "par" giving the c(alpha / 2, 1 - alpha / 2) confidence interval for the expected observed count at each site on each occasion, the estimated site super-population, the seasonal component, and the estimated parameter values, respectively.

calliste-fagard-jenkin/GAI documentation built on July 26, 2022, 11:20 a.m.