model {
####################################################################
 ## Attribution accross reservoirs 
  mu.m <- 0
  mu.epsilon <- 0
  prec.resvr <- 0.01
  prec.else <- 0.01

  for(k in 2:stud.n_MODEL){
    m.MODEL[k] ~ dnorm(mu.m, prec.else);
    epsilon.MODEL[k] ~ dnorm(mu.epsilon, prec.else);
    P.MODEL[k] ~ dnorm(m.MODEL[j],exp(epsilon.MODEL[k]))
  }
  P.MODEL[1] <- 0
    
  for(j in 2:stud.n_METHOD){
    m.METHOD[j] ~ dnorm(mu.m, prec.else);
    epsilon.METHOD[j] ~ dnorm(mu.epsilon, prec.else);
    P.METHOD[j] ~ dnorm(m.METHOD[j],exp(epsilon.METHOD[j]))
  }
  P.METHOD[1] <- 0
    
  for(i in 1:stud.n_reservoir){
    m.reservoir[i] ~ dnorm(mu.m, prec.resvr);
    epsilon.reservoir[i] ~ dnorm(mu.epsilon, prec.resvr);
    P.reservoir[i] ~ dnorm(m.reservoir[i],exp(epsilon.reservoir[i]))
  }

  for(ide in 1:stud.n_ID) {
  for(i in 1:stud.n_reservoir) {
  for(j in 1:stud.n_METHOD) {
  for(k in 1:stud.n_MODEL) {
    P[i,j,k,ide] <- ilogit(P.reservoir[i] +
                           P.MODEL[k] + 
                           P.METHOD[j] )
    log10.P[i,j,k,ide] <- log(P[i,j,k,ide]) / log(10)
  }
  }
  }
  }
  
  
  ## setting distribution for unknown/missing number of counts
  ## an arbitrtay ramping distribution which favours lower values
  for(l in 1:stud.n_EXPERIMENT){
    lmbd.extra[l] ~ dnorm(0, 0.01)
    n.extra.aux[l] ~ dpois(exp(lmbd.extra[l]))  
    n.extra.counts[l] <- round(n.extra.aux[l]) 
  }
  for(i in 1:stud.n){
    n.eff.counts[i] <- n.total.counts[i] + n.extra.counts[EXPERIMENT[i]]
  }


  for(i in 1:stud.n){
 ### Here fuzz data according to how old it is
    sig.pos.cases[i] <- (max(2020-AVG.YEAR[i] , 1))
    precision.positive.cases[i] <-  1/(sig.pos.cases[i])^3 
    observed.cases[i] ~ dnorm(cases[i],precision.positive.cases[i])
   
 ##################### single binomials for atribution #####################
    cases[i] ~ dbin( P[reservoir[i],
                       METHOD[i],
                       MODEL[i],
                       STUDY[i]
                      ],
                      n.eff.counts[i])
  }


####################################################################################
################ Attribution accross transmission routes ########################
#################################################################################
########## General transmission routes and expert opinion (trms)
  trms.expert.prec.log <- 1
  prec <- 0.01
  for(j in 1:trms.n_exp.Exposure){
    m.trms[j] ~ dnorm(mu.m, prec);
    u.trms[j] <- exp(m.trms[j]) / (1+exp(m.trms[j]))
    
    P.trms[j] <- u.trms[j]/sum.u.trms  
    log10.P.trms[j] <- log(P.trms[j]) / log(10)
  }
  sum.u.trms <- sum(u.trms) 
  
  lambd.vague.cases.trms ~ dlnorm(7,0.0001)
  for(j in 1:trms.n){
      trms.n.cases.total[j] ~ dpois(lambd.vague.cases.trms)  ## for handling absent cases
      trms.reduct.eff[j] <- min( trms.CP.observed[j]/trms.CP.experts[j], 1 ) ## reduction all cases to effective number
      trms.n.cases.eff[j] <- round( trms.reduct.eff[j]*trms.n.cases.total[j] )
      trms.n.observed[j] ~ dbin(P.trms[trms.Exposure[j]],
                                trms.n.cases.eff[j])
                                
      zero.trms[j] ~ dnorm((log(trms.P.Expert.estimates[j])-
                               log(P.trms[trms.Exposure[j]])),
                            trms.expert.prec.log)
  }

#####################################################
#####################################################
############## Foodborne-only transmission routes and expert opinion (fb)
  fb.expert.prec.log <- 1
  for(j in 1:fb.n_fb.Foodborne.exposure){
    w.fb[j] ~ dnorm(mu.m, prec);
    u.fb[j] <- exp(m.fb[j]) / (1+exp(m.fb[j]))
    
    P.fb[j] <- u.fb[j]/sum.u.fb  
    log10.P.fb[j] <- log(P.fb[j]) / log(10)
  }
  sum.u.fb <- sum(u.fb)  

  lambd.vague.cases.fb ~ dlnorm(7,0.0001)
  for(j in 1:fb.n){
      fb.n.cases.total[j] ~ dpois(lambd.vague.cases.fb)
      fb.reduct.eff[j] <- min( fb.CP.observed[j]/fb.CP.experts[j], 1 ) ## reduction all cases to effective number
      fb.n.cases.eff[j] <- round( fb.reduct.eff[j]*fb.n.cases.total[j] )
      fb.n.observed[j] ~ dbin(P.fb[fb.Foodborne.exposure[j]],
                                   fb.n.cases.eff[j])
      
      zero.fb[j] ~ dnorm((log(fb.P.Experts.estimates[j]) -
                               log(P.fb[fb.Foodborne.exposure[j]])),
                            fb.expert.prec.log)
      
  }


}
data {
   for(j in 1:trms.n){
    zero.trms[j] <- 0
  }
  for(j in 1:fb.n){
    zero.fb[j] <- 0
  }
}