Dimension mismatch in subset expression

[O.mykiss]

I am fairly new to RJAGS and I am crawling my way through writing model code and data formatting to perform some hierarchical modeling but I have hit some road bumps. I am trying to run the model

model <- function() {
  for(i in 1:z){
    for(j in 1:y){
      m[i,j] ~ dbin(p[i,j],n[i,j]) 
      logit(p[i,j]) <- etaP[i,j]
      etaP[i,j] ~ dnorm(etaP1[i],tauP)

      u[i,j] ~ dbin(p[i,j],U[i,j])
      U[i,j] <- round(exp(etaU[i,j])) 
      etaU[i,j]~ dnorm(etaU1[i],tauU)

    }
    etaP1[i] ~ dnorm(np,tp)

    etaU1[i] ~ dnorm(nu,tu)
  }
  tauP ~ dgamma(.001,.001)
  tauU ~ dgamma(.001,.001)

  np ~ dnorm(-2,1.22)
  tp ~ dgamma(.001,.001)

  nu ~ dnorm(-2,1.22)
  tu ~ dgamma(.001,.001)
}

##########Example data
DATA
z = 20
y = 10
m = vector of 200 integers
n = vector of 200 integers
u = vector of 200 integers

But I receive the error

"Error in jags.model(model.file, data = data, inits = init.values, n.chains = n.chains,  : 
  RUNTIME ERROR:
Compilation error on line 5.
Dimension mismatch in subset expression of m"

I have a hunch it has to do with my data classification more than my model (though my model might be off as well) but I am not sure. Any help would be hugely appreciated!

[Martyn Plummer]

In your model, m, n, and u are matrices with z rows and y columns. When you supply them as data they should also be matrices with matching dimensions (i.e. z=20 rows and y=10 columns).

[O.mykiss]

Thanks a ton Martyn. I reformatted the data for n, m, & u into 20x10 matrices and it cleared up the dimension mismatch. Now onto model errors (I haven't changed the model code from the original post);

Error in jags.model(model.file, data = data, inits = init.values, n.chains = n.chains,  : 
  Error in node u[1,1]
Observed node inconsistent with unobserved parents at initialization.
Try setting appropriate initial values.

I don't have any initial values set because I haven't needed them with my simpler models. I suspect the error has more to do with the model structure than the initial values but I am might be wrong.

[Martyn Plummer]

I think this is straightforward: The unobserved size parameter U[i,j] must be at least as big as the observed count u[i,j] so set initial values for U[i,j] accordingly.

[O.mykiss]

I haven't had to set initial values before so I tried

model.inits <- function(){
    list("p" = runif(z*y, 0.1, 0.7) , "U" = round(runif(z*y, 2000, 4000))) 
}

and received

Error in setParameters(init.values[[i]], i) : RUNTIME ERROR:
Dimension mismatch in values supplied for U

This should probably be straightforward but I am a JAGS novice and could use some advice.

Also, thanks again for your help so far Martyn.

[Martyn Plummer]

It's the same issue as with the data. The variable U is a z by y matrix, so when you supply initial values from R, these have to be in the form of a matrix with the same dimensions. Currently you are supplying a vector.

Note that you cannot supply initial values for p because it is not a random variable. In your model, p is calculated deterministically from etaP.

logit(p[i,j]) <- etaP[i,j]

You can supply initial values for etaP however, since this is a random variable:

etaU[i,j]~ dnorm(etaU1[i],tauU)