alexvantwisk/bayesianICSimulations: BMC Medical Research Methodology Release

v1.0.0 — Initial Release
This is the first stable release of bayesianICSimulations, developed as part of an MSc Biostatistics dissertation at Stellenbosch University.
What this package does
Implements a large-scale simulation study comparing two Bayesian MCMC samplers for fitting log-logistic AFT models to interval-censored survival data with survey weights:

• HMC via Stan / CmdStanR
• Metropolis-Hastings via JAGS / rjags

The study covers 5,400 simulated datasets across three sample sizes (n = 200, 2,000, 10,000) and evaluates sampler performance under the ADEMP framework (bias, RMSE, coverage, computational efficiency).

Included

Model fitting

• fit_logistic_hmc() — HMC fitting with Stan; 4 parallel chains, reproducible seed, standardized output
• fit_logistic_mh() — MH fitting with JAGS; zeros trick weighted likelihood, reproducible per-dataset seeds

Analysis pipeline

• combine_results() — Aggregates thousands of .rds outputs; parses filenames, computes error metrics, flags convergence
• perform_statistical_analysis() — ADEMP-framework analysis: bias, RMSE, coverage, CI width, all with Monte Carlo standard errors
• mcse_mean(), mcse_prop(), calc_bias_rmse(), cohen_d() — MCSE and effect size utilities

Visualisation (13 figures)

• R-hat ECDFs, ESS ridge plots, coverage plots, bias/RMSE comparisons, ESS-per-second, runtime, posterior means agreement, credible interval matrices

HPC support

• PBS job array scripts for Stellenbosch University cluster
• One-time setup script (installs JAGS from source, CmdStan, all R dependencies)
• Supports 10,800 model fits in parallel

Real-data application

• ZIMPHIA (Zimbabwe Population-based HIV Impact Assessment) analysis functions and tables

Requirements

• R ≥ 4.0.0
• CmdStan (via cmdstanr::install_cmdstan())
• JAGS 4.x (brew install jags / apt-get install jags)

▎ Developed as part of an MSc Biostatistics research project, Stellenbosch University, 2025.