Bugs Fundamentals

BUGS/JAGS Fundamentals

When to Use This Skill

• Writing new WinBUGS or JAGS models
• Understanding BUGS declarative syntax
• Converting between BUGS and Stan
• Integrating with R via R2jags or R2WinBUGS

Model Structure

BUGS uses a single declarative block where order doesn't matter:

model {
  # Likelihood (order doesn't matter)
  for (i in 1:N) {
    y[i] ~ dnorm(mu[i], tau)
    mu[i] = threshold
eq <- equals(y, 0)           # 1 if y == 0

Common Priors

# Vague normal (variance = 1000)
alpha ~ dnorm(0, 0.001)

# Half-Cauchy on SD (via uniform)
sigma ~ dunif(0, 100)
tau <- pow(sigma, -2)

# Vague gamma on precision
tau ~ dgamma(0.001, 0.001)

# Correlation matrix
Omega ~ dwish(I[,], K + 1)

R Integration

R2jags (Recommended)

library(R2jags)

jags.data <- list(N = 100, y = y, x = x)
jags.params <- c("alpha", "beta", "sigma")
jags.inits <- function() {
  list(alpha = 0, beta = 0, tau = 1)
}

fit <- jags(
  data = jags.data,
  inits = jags.inits,
  parameters.to.save = jags.params,
  model.file = "model.txt",
  n.chains = 4,
  n.iter = 10000,
  n.burnin = 5000
)

print(fit)
fit$BUGSoutput$summary

R2WinBUGS (Windows)

library(R2WinBUGS)

fit <- bugs(
  data = bugs.data,
  inits = bugs.inits,
  parameters.to.save = bugs.params,
  model.file = "model.txt",
  n.chains = 3,
  n.iter = 10000,
  bugs.directory = "C:/WinBUGS14/"
)

Key Differences from Stan

| Feature | BUGS/JAGS | Stan | |---------|-----------|------| | Normal | dnorm(mu, tau) precision | normal(mu, sigma) SD | | MVN | dmnorm(mu, Omega) precision | multi_normal(mu, Sigma) cov | | Syntax | Declarative (DAG) | Imperative (sequential) | | Blocks | Single model{} | 7 optional blocks | | Sampling | Gibbs + Metropolis | HMC/NUTS | | Discrete | Direct sampling | Marginalization required |

Common Errors

1. Using SD instead of precision: dnorm(0, 1) means variance=1, NOT SD=1
2. Wrong binomial order: dbin(p, n) not dbin(n, p)
3. Missing initial values: Provide inits for complex models
4. Invalid parent values: Check for NA/NaN in data

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: choxos
• Source: choxos/BiostatAgent
• License: MIT

Install and usage instructions live in the source repository linked above.