Functions to fit various spatio-temporal GLMMs in Template Model Builder (TMB), and produce fully Bayesian benchmarked estimates. Additional functionality includes various benchmarking approaches applied to posterior samples.
stbench is an open-source R package for producing fully Bayesian benchmarked estimates for hierarchical models in TMB. The package currently supports spatial and spatio-temporal modeling of binomial count data of under-5 mortality (U5MR), spatial models for generic binomial count data, and spatial models for normally distributed outcomes.
Current model implementations:
-
U5MR
- Single-survey
- Space-only
- Binomial benched/unbenched
- Time-only
- Binomial benched/unbenched
- Space-time
- Binomial benched/unbenched
- BetaBinomial benched/unbenched
- Space-only
- Multi-survey
- Space-time
- Binomial benched/unbenched
- Space-time
- Single-survey
-
Generic binary outcomes
- Single survey
- Space-only
- Binomial benched/unbenched
- Space-only
- Single survey
-
Normally distributed outcomes
- Single survey
- Space-only
- benched/unbenched
- Space-only
- Single survey
Details can be found in the functions fit_u5mr, fit_binary, and fit_normal, respectively.
Current functions available that apply benchmarking methods to samples from a distribution include:
-
benchmark_sampler: Fully Bayesian benchmarking via a rejection sampler, described in Okonek and Wakefield, 2022 -
benchmark_bayesest: Constrained Bayes estimate approach to benchmarking, described in Datta et al., 2011 -
benchmark_mh: Fully Bayesian benchmarking via a Metropolis-Hastings algorithm.
The current development version can be installed using devtools::install_github():
devtools::install_github(repo="taylorokonek/stbench")