HSI
From: Bayesian Models for Astrophysical Data, Cambridge Univ. Press
(c) 2017, Joseph M. Hilbe, Rafael S. de Souza and Emille E. O. Ishida
you are kindly asked to include the complete citation if you used this material in a publication
Code 6.24 Zero Truncated Negative binomial with 0-trick using JAGS - direct
=============================================================
require(MASS)
require(R2jags)
require(VGAM)
set.seed(123579)
nobs <- 1000
x1 <- rbinom(nobs,size=1,0.7)
x2 <- runif(nobs)
xb <- 1 + 2*x1 - 4*x2
exb <- exp(xb)
alpha = 5
rztnb <- function(n, mu, size){
p <- runif(n, dzinegbin(0, munb=mu, size=size),1)
ztnb <- qzinegbin(p, munb=mu, size=size)
return(ztnb)
}
ztnby <- rztnb(nobs, exb, size=1/alpha)
ztnb.data <-data.frame(ztnby, x1, x2)
X <- model.matrix(~ x1 + x2, data = ztnb.data)
K <- ncol(X)
model.data <- list(Y = ztnb.data$ztnby,
X = X,
K = K, # number of betas
N = nobs,
Zeros = rep(0, nobs)) # sample size
ZTNB <- "
model{
for (i in 1:K) {beta[i] ~ dnorm(0, 1e-4)}
alpha ~ dgamma(1e-3,1e-3)
# Likelihood with zero trick
C <- 10000
for (i in 1:N) {
# Log likelihood function using zero trick:
Zeros[i] ~ dpois(Zeros.mean[i])
Zeros.mean[i] <- -L[i] + C
l1[i] <- 1/alpha * log(u[i])
l2[i] <- Y[i] * log(1 - u[i])
l3[i] <- loggam(Y[i] + 1/alpha)
l4[i] <- loggam(1/alpha)
l5[i] <- loggam(Y[i] + 1)
l6[i] <- log(1 - (1 + alpha * mu[i])^(-1/alpha))
L[i] <- l1[i] + l2[i] + l3[i] - l4[i] - l5[i] - l6[i]
u[i] <- 1/(1 + alpha * mu[i])
log(mu[i]) <- inprod(X[i,], beta[])
}
}"
inits <- function () {
list(beta = rnorm(K, 0, 0.1))}
params <- c("beta","alpha")
ZTNB1 <- jags(data = model.data,
inits = inits,
parameters = params,
model = textConnection(ZTNB),
n.thin = 1,
n.chains = 3,
n.burnin = 2500,
n.iter = 5000)
print(ZTNB1, intervals=c(0.025, 0.975), digits=3)
=============================================================
Output on screen:
Inference for Bugs model at "3", fit using jags, 3 chains,
each with 5000 iterations (first 2500 discarded)
n.sims = 7500 iterations saved
mu.vect sd.vect 2.5% 97.5% Rhat n.eff
alpha 4.619 1.198 2.936 7.580 1.032 81
beta[1] 1.099 0.199 0.664 1.445 1.016 150
beta[2] 1.823 0.131 1.564 2.074 1.004 600
beta[3] -3.957 0.199 -4.345 -3.571 1.008 290
deviance 20004596.833 2.861 20004593.165 20004603.978 1.000 1
For each parameter, n.eff is a crude measure of effective sample size,
and Rhat is the potential scale reduction factor (at convergence, Rhat=1).
DIC info (using the rule, pD = var(deviance)/2)
pD = 4.1 and DIC = 20004600.9
DIC is an estimate of expected predictive error (lower deviance is better).