editor_options: markdown: wrap: 72

bayescoveragedeploy

Precompiled Stan models for fitting local country-level Bayesian hierarchical transition models for health coverage indicators.

Overview

bayescoveragedeploy is a deployment-focused package that provides precompiled Stan models for users without C++ compilers. It is a thin wrapper around the bayescoveragemodel package, which contains all the core functionality for fitting Bayesian hierarchical transition models to health coverage indicators (ANC4, institutional delivery, vaccination).

Key Features

No C++ compiler required: Stan models are precompiled during package installation
Fast startup: No compilation wait time when running models
Simple interface: One focused function for fitting local country models
Full functionality: All data processing and analysis from bayescoveragemodel available

Binary packages are available for Windows and macOS.

Installation

Installation instructions are on https://github.com/AlkemaLab/bayescoverage_app.

To use the deploy package, install the following Bayescoverage-related packages from github (now also available via R universe)

devtools::install_github("AlkemaLab/bayescoveragemodel") devtools::install_github("AlkemaLab/localhierarchy")

Install the deploy-related packages from R universe:

install.packages('bayescoveragedeploy', repos = c('https://alkemalab.r-universe.dev'))

Also install the following package from CRAN:

install.packages(c("dplyr", "haven", "ggplot2", "shiny", "posterior", "readr", "here", "stringr", "tibble"))

For using routine data, brms is used as well, install using

install.packages("brms")

Usage

Quick Start

library(bayescoveragedeploy)
library(haven)
library(readr)

# Load data
dat0 <- read_dta("data_raw/ICEH_national.dta")
regions_dat <- read_csv("data_raw/regions_updated.csv")

# Process data using parent package functions
data <- bayescoveragemodel::process_data(
  dat = dat0,
  regions_dat = regions_dat,
  indicator = "anc4"
)

# Fit local model with precompiled Stan models
fit <- fit_local_model(
  survey_df = data,
  iso_select = "KEN",
  indicator = "anc4",
  chains = 4,
  iter_sampling = 200
)

# Plot results using parent package functions
bayescoveragemodel::plot_estimates_local_all(fit)

Available indicators

The package includes global fit objects for:

anc4 - Antenatal care (4+ visits)
ideliv - Institutional delivery
vdpt - DPT vaccination
anc1trimester - Antenatal care (first trimester)
ancq8 - Antenatal care quality (8+ contacts)
bfexcl0_5 - Exclusive breastfeeding (0-5 months)
cci - Comprehensive coverage index
sba - Skilled birth attendance
vmsl - Measles vaccination

With routine data

# Fit model with routine data
fit <- fit_local_model(
  survey_df = data,
  routine_df = routine_data,
  iso_select = "KEN",
  indicator = "anc4"
)

Package Architecture

This package contains:

Precompiled Stan models: 4 model variants (fpem, fpem_routine, fpem_aggregates, fpem_routine_aggregates)
Thin wrapper function: fit_local_model() delegates to bayescoveragemodel::fit_model()

All core functionality (data processing, model fitting logic, plotting, etc.) comes from the bayescoveragemodel package.

For Developers

Updating the package

When bayescoveragemodel is updated:

Check if Stan models changed (in bayescoveragemodel/inst/stan/)
If changed, copy updated Stan files to bayescoveragedeploy/inst/stan/
Update include paths if needed (should use chunks/ prefix)
Rebuild and test.

License

MIT License - see LICENSE file

References

Stan Development Team. RStan: the R interface to Stan. https://mc-stan.org