model{ 
		y[1:s] ~ dmulti(th[1 : s] , n)
		sum.g <- sum(g[])
	# smoothed frequencies
		for (i in 1 : s) {     
			Sm[i] <- n * th[i]
			g[i] <- exp(gam[i])    
			th[i]  <- g[i] / sum.g
		}
	# prior on elements of AR Precision Matrix  
		rho ~ dunif(0, 1)
		tau ~ dunif(0.5, 10)
	# MVN for logit parameters
		gam[1 : s] ~ dmnorm(mu[], T[ , ])
		for (j in 1:s) { 
			mu[j] <- -log(s)
		}
	# Define Precision Matrix
		for (j in 2 : s - 1) {
			T[j, j] <- tau * (1 + pow(rho, 2))
		}
		T[1, 1] <- tau T[s, s] <- tau
		for (j in 1 : s-1) { 
			T[j, j + 1] <- -tau * rho
			T[j + 1, j] <- T[j, j + 1]
		}
		for (i in 1 : s - 1) {
			for (j in 2 + i: s) {
				T[i, j] <- 0; T[j, i] <- 0 
			}
		}
	# Or Could do in terms of covariance, which is simpler to write but much slower
#		for (i in 1 :s) {
#			for (j in 1 : s) {
#				cov[i, j] <- pow(rho, abs(i - j)) / tau
#			}
#		}
#		T[1:s, 1: s] <- inverse(cov[,])
	}
