Bayesian Analysis Workshop

Bayesian Analysis for Applied Social Science and Education Research is a two-day hands-on workshop for researchers who already use frequentist methods and want to learn Bayesian analysis as a practical research workflow.

The materials are packaged as a Quarto website with participant notebooks, teaching datasets, facilitator notes, slides, and reporting templates.

Live Site

The rendered workshop is hosted on AWS at:

https://d1etjt8tsng564.cloudfront.net/

Participants can read the full notebooks online without installing anything. The local installation steps below remain required for running the R chunks interactively.

Quick Start

Follow these steps to run the workshop materials locally.

1. Clone the Repository

Windows users: clone to a short path outside OneDrive, for example C:\workshops\ or $HOME\ (not $HOME\Documents\ if your Documents folder is synced to OneDrive). OneDrive sync locks and long paths can break Stan model compilation later.

git clone https://github.com/moeketsims/BayesianAnalysis.git
cd BayesianAnalysis

If you already have the repository locally, move into the folder where you saved or cloned it:

cd "path\to\BayesianAnalysis"

For example:

cd "C:\workshops\BayesianAnalysis"

2. Install Required Software

Install in this order:

1. R: https://cran.r-project.org/
2. RTools (Windows only): https://cran.r-project.org/bin/windows/Rtools/
   • RTools is the C++ toolchain Stan needs to compile Bayesian models. Without it, model fitting fails on Windows.
   • The RTools version must match your R version: RTools43 for R 4.3.x, RTools44 for R 4.4.x, RTools45 for R 4.5.x. Installing the wrong version will not work.
3. Quarto: https://quarto.org/docs/get-started/
4. RStudio or Positron: recommended for working with the notebooks.
5. Node.js: optional, only needed if you want to regenerate or validate the synthetic datasets.

After installing the above on Windows, close and reopen PowerShell so the new programs are available on your PATH. Then check:

Rscript --version
quarto --version

3. Install R Packages

From the project folder, run:

Rscript setup/packages.R

This installs the main packages used in the workshop, including:

• brms
• cmdstanr (installed from the Stan r-universe repository — not on CRAN)
• tidyverse
• tidybayes
• bayesplot
• posterior
• loo

The script then runs cmdstanr::install_cmdstan() to install the underlying Stan toolchain. On Windows this step requires RTools (Step 2). Bayesian modeling with brms uses Stan, so the full install can take 10–20 minutes.

4. Test the R Setup

Run:

Rscript setup/test_installation.R

If the test reports missing packages, rerun:

Rscript setup/packages.R

If CmdStan is missing inside R, run:

cmdstanr::install_cmdstan()

More setup detail is available in setup/install_guide.md.

5. Validate the Workshop Files

If Node.js is installed, run:

node scripts/validate_structure.mjs

Expected output:

Workshop package structure validated.

6. Regenerate the Synthetic Datasets, Optional

The datasets are already included. To regenerate them:

node scripts/generate_synthetic_data.mjs

This rewrites the files in data/ using a fixed seed.

7. Preview the Workshop Website

Run:

quarto preview

Quarto will start a local server and print a URL such as:

http://localhost:4200

Open that URL in your browser.

8. Open Individual Notebooks

You can also open the notebooks directly in RStudio or Positron:

notebooks/01_bayesian_reasoning.qmd
notebooks/02_priors.qmd
notebooks/03_bayesian_regression.qmd
notebooks/04_binary_and_ordinal_models.qmd
notebooks/05_hierarchical_models.qmd
notebooks/06_model_checking.qmd
notebooks/07_reporting.qmd

If Something Does Not Work

Rscript is not recognized

R is either not installed or not on your system path. Install R, restart PowerShell, and try again.

quarto is not recognized

Install Quarto from https://quarto.org/docs/get-started/ and restart PowerShell.

Stan or CmdStan fails

Stan needs a working C++ toolchain.

On Windows, install RTools for your R version. Then restart R and run:

cmdstanr::install_cmdstan()

Bayesian models take too long

The notebooks can still be used for teaching and interpretation. The facilitator should render the notebooks in advance and keep expected outputs available.

Repository Structure

.
  README.md
  WORKSHOP_DEVELOPMENT_PLAN.md
  _quarto.yml
  index.qmd
  setup/
    install_guide.md
    packages.R
    test_installation.R
  data/
    education_intervention.csv
    teacher_adoption_survey.csv
    student_engagement_ordinal.csv
    data_dictionary_*.md
  notebooks/
    00_notebook_template.qmd
    01_bayesian_reasoning.qmd
    02_priors.qmd
    03_bayesian_regression.qmd
    04_binary_and_ordinal_models.qmd
    05_hierarchical_models.qmd
    06_model_checking.qmd
    07_reporting.qmd
  facilitator-guide/
  reporting-templates/
  slides/
  solutions/
  scripts/

Audience

This workshop assumes participants already understand common applied methods:

• p-values and confidence intervals;
• linear regression;
• logistic regression at a basic level;
• interpretation of coefficients;
• applied research design and reporting.

It does not assume prior Bayesian training or Stan programming experience.

Main Workflow Taught

research question
  -> likelihood
  -> prior
  -> prior predictive check
  -> model fitting
  -> posterior interpretation
  -> model checking
  -> sensitivity analysis
  -> reporting

Suggested Teaching Sequence

Day 1:

1. Bayesian reasoning for frequentist researchers.
2. Priors and prior predictive checks.
3. Bayesian linear regression.

Day 2:

1. Bayesian hierarchical models.
2. Model checking and model comparison.
3. Binary and ordinal model extensions.
4. Reporting Bayesian results.
5. Applied planning for participants' own research.

Main Materials

• Workshop development plan
• Setup guide
• Participant notebooks
• Facilitator guide
• Participant analysis plan
• Reporting templates
• Slides

Deployment

The site is hosted on AWS: a private S3 bucket (bayesian-analysis-workshop-208509455458 in us-east-1) fronted by a CloudFront distribution (EOET36EGUHLUZ). The bucket policy grants read access only to that distribution via Origin Access Control, so the bucket itself stays unreachable.

Re-publish after edits

Pushes to main automatically render and publish via GitHub Actions (.github/workflows/deploy.yml), so the usual workflow is just to commit and push. The Action authenticates to AWS via OIDC — no static credentials are stored in the repo.

For local previews or out-of-band publishing, run from the project root:

scripts/deploy.sh

The script renders the Quarto site, syncs _site/ to S3, and invalidates CloudFront so visitors see the latest version within a few minutes. Pass --skip-render to publish without re-rendering.

Required for local use: quarto, aws CLI configured under the default profile, and R with the workshop packages installed.

Costs at workshop traffic volumes sit under $1/month: CloudFront's free tier covers the first 1 TB/month of egress, and S3 storage at 11 MiB is negligible.

Development Status

This repository contains the first complete source package and is live online via the URL above. The next recommended step is to pilot Notebook 01 with a small group.