nimble.occ: Estimation of Single-Season Occupancy Models
In clewisbe/nimbleEco: Fit Ecological Models Using NIMBLE

Description Usage Arguments Details Value Author(s) Examples

View source: R/nim_occupancy.R

nimble.occ is used to fit single species single-season occupancy models using Bayesian estimation. The function returns model summary statistics, posterior draws, and the BUGS code used to estimate the model.

nimble.occ(siteformula = NULL, obsformula = NULL, y = NULL,
  sitevars = NULL, obsvars = NULL, priors = c("Normal", "t", "Uniform",
  "Gamma"), dropbase = TRUE, droplast = FALSE, niter = 10000,
  burnin = 1000, initmcmc = c(0.5, 1, 5, 10), chains = 4,
  returncode = TRUE, returnsamp = TRUE, ...)

`siteformula`	A linear model formula for the latent state: log(λ_{i}). Formula must start with a tilde (~). Add -1 to remove intercept. Random effects are specified with a pipe (see lmer). If model contains an intercept random effects will be centered at 0. Interactions can be specified with * (e.g., x*y).
`obsformula`	A linear model formula for the observation process: logit(p_{i,j}).
`y`	Detection and non detection data. Must be provided as either a matrix or dataframe. Rows are sites and columns are replicates. 1 is detection and 0 is non detection.
`sitevars`	A dataframe of site-specific covariates. Must be a dataframe with named columns corresponding to variables. Number of rows should equal the number of sites.
`obsvars`	A named list with site and survey specific covariates. Each list element is a matrix with rows corresponding to site and columns to replicate. Variables specified in the model must match the variable names in the list.
`priors`	One prior is specified for all model parameters in the siteformula and obsformula model statements. The default prior is a Normal(mu = 0, sd = 1000). Other options are t(mu = 0, tau = 1000, df = 5), Uniform(0, 1000), and Gamma(shape = .001, rate = .001).
`dropbase`	Drops the first level of a factor variable. Default is TRUE.
`droplast`	Drops the last level of a factor variable. Default is FALSE.
`niter`	Number of iterations to run the MCMC. Default is 10,000 iterations.
`burnin`	Number of burnin iterations for MCMC. Default is 1000 iterations.
`initmcmc`	Vector of initial values for MCMC. Length of vector must equal the number of chains. The same initial values are used for all parameters. Default is c(1, 5, 10, 15)
`chains`	Number of MCMC chains to run. Default is 4.
`returncode`	Option to return the BUGS code passed to NIMBLE. Default is TRUE.
`returnsamp`	Option to return all posterior draws. Default is TRUE
`...`	Additional arguments can be passed to specify different parameters for the priors. Options are location, scale, lb, ub, df, shape, and rate.

This is a function for fitting Single Season-Occupany Models using a Bayeisan framework. The model structure is:

z_{i} ~ Bernoulli(ψ_{i})

y_{i,j} | z_{i} ~ Bernoulli(z_{i} * p_{i,j})

Where i is the number of sites and j the number of temporal replicates. z is a binary latent variable for the true observed state (occupied or not occupied). Both ψ and p can be modeled as a linear function of covariates via a linked GLM. The link for ψ and p is the logit link. Estimation is performed using NIMBLE. In addition to fitting the model, generated BUGS code for the full model, as well as posterior draws, are returned.

Output is a named list with the following elements: Summary, BUGScode, and Samples. In the Summary statement, variables from the siteformula will start with "s.", and variables in the obsformula will start with "o." Note: parameters are on the transformed scale (log for siteformula variables; logit for obsformula variables). In addition to quantiles, the effective sample size effectiveSize and Gelman Rubin diagnoistic gelman.diag are provided from the coda package.

Colin Lewis-Beck

# Simualte Occupancy Data
R <- 200
T <- 3
X <- sort(runif(n = R, min = -1, max = 1))
Y <- sort(rgamma(n = R, shape = 1, rate = 1))
sim.covariates <- data.frame("X" = X, "Y" = Y)
psi <- plogis(-1 + 2*X)
p <- plogis(1 - 3*X + 4*Y + 4*X*Y)
z <- rbinom(n = R, size = 1, prob = psi)
y <- matrix(NA, nrow = R, ncol = T)
for (i in 1:T){y[,i] <- rbinom(n = R, size = 1, prob = z*p)}


#Fit Model
model <- nimble.occ(siteformula = ~ 1 + X, obsformula = ~ 1 + X +
Y + X*Y, y = y, sitevars = sim.covariates, priors = "t", initmcmc = c(.5,
1.5, 3, 5), chains = 4)