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] <- alpha + beta * x[i] } # Priors alpha ~ dnorm(0, 0.001) beta ~ dnorm(0, 0.001) tau ~ dgamma(0.001, 0.001) # Derived quantities sigma <- 1 / sqrt(tau) } ``` ## CRITICAL: Precision Parameterization **BUGS uses PRECISION (tau = 1/variance), NOT standard deviation:** | Distribution | BUGS Syntax | Meaning | |-------------|-------------|---------| | Normal | `dnorm(mu, tau)` | tau = 1/sigma² | | MVN | `dmnorm(mu[], Omega[,])` | Omega = inverse(Sigma) | ### Converting SD ↔ Precision ``` # Precision from SD tau <- pow(sigma, -2) # SD from precision sigma <- 1 / sqrt(tau) ``` ## Distribution Reference ### Continuous (All use precision!) ``` y ~ dnorm(mu, tau) # Normal: tau = 1/sigma² y ~ dlnorm(mu, tau) # Log-normal (log-scale) y ~ dt(mu, tau, df) # Student-t y ~ dunif(lower, upper) # Uniform y ~ dgamma(shape, rate) # Gamma y ~ dbeta(a, b) # Beta y ~ dexp(lambda) # Exponential (rate) y ~ dweib(shape, lambda) # Weibull y ~ ddexp(mu, tau) # Double exponential ``` ### Discrete ``` y ~ dbern(p) # Bernoulli y ~ dbin(p, n) # Binomial (p first!) y ~ dpois(lambda) # Poisson y ~ dnegbin(p, r) # Negative binomial y ~ dcat(p[]) # Categorical y ~ dmulti(p[], n) # Multinomial ``` ### Multivariate ``` y[1:K] ~ dmnorm(mu[], Omega[,]) # MVN (precision matrix!) Omega[1:K,1:K] ~ dwish(R[,], df) # Wishart (for precision) p[1:K] ~ ddirch(alpha[]) # Dirichlet ``` ## Syntax Essentials ### Stochastic vs Deterministic ``` # Stochastic (random variable) y ~ dnorm(mu, tau) # Deterministic (function) mu <- alpha + beta * x ``` ### Loops ``` for (i in 1:N) { y[i] ~ dnorm(mu[i], tau) } ``` ### Truncation (JAGS) ``` y ~ dnorm(mu, tau) T(lower, upper) y ~ dnorm(mu, tau) T(0, ) # Lower only ``` ### Logical Functions (JAGS) ``` ind <- step(y - threshold) # 1 if y >= 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) ```r 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) ```r 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