Setting up R, JAGS, and Nimble on Windows and Mac

What you need

By the end of this page you should have all of the following working:

1. Install R and RStudio

2. Install the compiler toolchain (needed by Nimble)

Nimble writes C++ and compiles it, so it needs a compiler. This is the step people forget.

3. Install JAGS

brew install jags

4. Install the R packages

In the RStudio Console (same on Windows and Mac):

install.packages(c(
  "coda", "rjags", "R2jags",
  "nimble", "MCMCvis", "HDInterval",
  "ggplot2", "dplyr"
))

Notes:

• rjags must be installed after JAGS, or it will fail to load.
• nimble is large and may take several minutes; it does not fully compile until you build your first model (that's the verification below).

5. Verify everything works

Run each block. If all three succeed, you are ready for the Bayesian labs.

a. JAGS + rjags

library(rjags)
cat("model { p ~ dbeta(1,1); y ~ dbin(p, 10) }", file = "m.bug")
m <- jags.model("m.bug", data = list(y = 7), n.chains = 3, quiet = TRUE)
s <- coda.samples(m, "p", n.iter = 2000)
summary(s)$statistics       # posterior mean of p should be near 0.67

b. Nimble (this is the real compiler test)

library(nimble)
code <- nimbleCode({
  p ~ dbeta(1, 1)
  y ~ dbinom(p, size = 10)
})
fit <- nimbleMCMC(code = code, data = list(y = 7),
                  inits = list(p = 0.5),
                  monitors = "p", niter = 2000, nburnin = 500)
mean(fit[, "p"])          # near 0.67; if this compiles, your toolchain works

c. MCMCvis (we use it to summarize posteriors)

library(MCMCvis)
MCMCsummary(s)             # tidy table with mean, sd, 95% CI, Rhat, n.eff

       mean        sd      2.5%       50%     97.5% Rhat n.eff
p 0.6691019 0.1272098 0.4078615 0.6802727 0.8862891    1  4003

Your digits will differ slightly — the samplers are stochastic and the blocks above set no seed. What matters is that the table prints: the mean sits near 0.67 and Rhat is about 1.00. The three chains in block (a) are what make Rhat computable; with a single chain it comes back NA and MCMCvis warns.

6. Troubleshooting

"Error: .onLoad failed ... cannot find JAGS" (rjags)

• You installed rjags before JAGS. Reinstall JAGS, then run install.packages("rjags", type = "source").
• Windows: if it still can't find JAGS, set the path once per session with Sys.setenv(JAGS_HOME = "C:/Program Files/JAGS/JAGS-4.3.1") (adjust the version), then reload library(rjags).
• Mac (Homebrew on Apple Silicon): JAGS lands under /opt/homebrew. If rjags can't find it, install rjags from source: install.packages("rjags", type = "source").

Nimble compile errors ("make: command not found", "clang: error", "cannot compile")

• Windows: Rtools is missing or the wrong version. Reinstall the Rtools that matches your R version and restart RStudio. Confirm with pkgbuild::has_build_tools(debug = TRUE).
• Mac: run xcode-select --install again; if it says "already installed" but compiling still fails, reset it with sudo xcode-select --reset.
• Restart R after installing any toolchain (Session → Restart R).

Package won't install / binary vs source

• If a binary install fails, try install.packages("nimble", type = "source").
• Keep R and RStudio up to date; very old R versions may not have current package binaries.

Still stuck?

Email the instructor before Day 1 with: your operating system and chip, your R.version.string, and the exact error text. Setup problems are much faster to fix ahead of time than during a lab.

Final checklist