Hi,

I'm trying to estimate a random effects probit with missing response (y) using a latent Normal approach (Albert and Chib 1993).

The model is below:

# Model

model{

for (i in 1:N){ # N - No. of individuals

z[i] ~ dnorm(mu[i],1)T(low[y[i]+1],high[y[i]+1]) mu[i] <- beta[1] + u[cluster[i]] + beta[2]*treat[i] + beta[3]*x3[i] + beta[4]*x4[i] }

low[1]<--20 ; low[2]<-0 ; high[1]<-0 ;high[2]<-20

# Level-2 random effects

for (j in 1:J){ # J - No. of studies

u[j] ~ dnorm(zero, tau.u)

}

# Prior on level-2 variance

tau.u ~ gamma(0.001, 0.001)

sigma2.u <-inverse(tau.u)

# Prior on betas

beta[1:4] ~ dmnorm(b0[1:4] , B0[1:4,1:4])

}

When I have fully-observed outcome, the model runs fine, but with missing outcome, it fails to compile (even when I provide starting values for the missing 'y'):

RUNTIME ERROR:

Compilation error on line 4.

Unable to resolve node y[102]

This may be due to an undefined ancestor node or a directed cycle in the graph

Any help would be much appreciated.

Manny