I am relatively new to Bayesian statistics and am trying to apply a non-linear hierarchical model using R2winBUGS on some tree stocking density data. I am hoping someone may be able to help me find the reason why R2winBUGS is giving me the following error:
gen.inits cannot be executed (is greyed out)
Even though I get this error, the code still produces output. For parameters that converge, the model produces output that seem reasonable but there are two parameters (mean_N0 sigma_N0 in below code) that are not mixing well (not converging) when I use two chains. The chains seem to start around the initial value (i.e. mean_N0 starts at 4800.5 and 4799.5, whereas sigma_N0 starts at 800.5 and 799.5) but don’t move far from those values. The mean values for both parameters are about 0.5 off from the initial set values. I am not sure whether the above error is causing this convergence problem.
I have exhausted my investigations into this problem and am now hoping someone may be able to see what is causing my problem in the below winBUGS or R code. I would greatly appreciate your time if you could help.
Kind Regards
Dom
WINBUGS CODE
model {
## loop over data for likelihood
for(i in 1:Ntotal){
N[i] <- log(N0[P_ID_Bug[i]] - 25)-(Age[i]/(Beta_0 + Beta_1*Age[i]))
Y[i] ~ dnorm(N[i],tauY)
}
tauY ~ dgamma(1.0E-3, 1.0E-3)
Beta_0 ~ dnorm(9,0.25)
Beta_1 ~ dnorm(0.16,400)
## hierarchical model for each Plots intercept & slope
for (p in 1:P_ID_Length) {
N0[p] ~ dgamma(r_N0, lambda_N0)
}
mean_N0 ~ dnorm(5000,1.0E-6)
sigma_N0 ~ dnorm(5000,0.25E-6)
lambda_N0 <- mean_N0/(sigma_N0*sigma_N0)
r_N0 <- mean_N0 * lambda_N0
}
R CODE
data <- list(P_ID_Length = length(P_ID),
P_ID_Bug = P_ID_Bug,
Age=Grouped_SDen$Age,
Y =Grouped_SDen$LogSDen_Ha,
Ntotal=nrow(Grouped_SDen))
inits1 <- list(N0= rep(coef(NLS_SDen_Log_1)[[1]], P_ID_Length),
Beta_0 = coef(NLS_SDen_Log_1)[[2]],
Beta_1 = coef(NLS_SDen_Log_1)[[3]],
tauY = 20,
mean_N0 = 4800,
sigma_N0 = 800
)
inits2 <- list(N0= rep(coef(NLS_SDen_Log_1)[[1]], P_ID_Length),
Beta_0 = coef(NLS_SDen_Log_1)[[2]],
Beta_1 = coef(NLS_SDen_Log_1)[[3]],
tauY = 20,
mean_N0 = 4800,
sigma_N0 = 800
)
inits <- list(inits1, inits2)
parameters <- c("N0", "Beta_0", "Beta_1", "tauY", "mean_N0", "sigma_N0")
sims <- bugs(model.file= "C:/WS/Post-Doc/TINNR/PAPER_4_WinBugs/Stocking_Growth.bug",
data = data,
parameters = parameters,
inits = inits,
n.chains = 2,
n.iter = 1000,
n.burnin = 500,
n.thin = 2,
debug=TRUE,
bugs.directory = "C:/Program Files/WinBUGS14/")
I have also tried using the following format for inits:
inits <- function(){
list(N0= rep(coef(NLS_SDen_Log_1)[[1]], P_ID_Length),
Beta_0 = coef(NLS_SDen_Log_1)[[2]],
tauY = 1,
mean_N0 = 4800,
sigma_N0 = 800
)
}
Note that the first and second inits are the same and below I show you what the second list looks like. Thanks again for finding the time to help.
[[2]]
[[2]]$N0
[1] 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531
[13] 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531
[25] 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531
[37] 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531
[49] 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531
[61] 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531
[73] 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531
[85] 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531
[97] 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531 4920.531
[[2]]$Beta_0
[1] 8.710965
[[2]]$Beta_1
[1] 0.1623536
[[2]]$tauY
[1] 20
[[2]]$mean_N0
[1] 4800
[[2]]$sigma_N0
[1] 800