models/goodbad_models/test_ac3.R

In soniamitchell/SpARKjags:

# ac1 and ac2
model {
  # Define likelihood model for data:
  for (p in 1:N_patients)
  {
    # probability of belonging to the bad group is different for clinical /
    # carriage samples
    bad.p[p] ~ dbern(prob.of.bad.hosp[hospital_clinical[p]])
    # index the bad group in a.prob
    index.bad.p[p] <- bad.p[p] + 1

    for(a in 1:antibiotic_classes)
    {
      # Response is different for each antibiotic and depending on
      # which pop it's from
      hospital_response[p,a] ~
        dbern(ac.prob[a,
                      index.bad.p[p],
                      hospital_clinical[p]])
    }
  }

  for (gp in 1:N_gp)
  {
    # probability of belonging to the bad group
    bad.gp[gp] ~ dbern(prob.of.bad.gp)
    # index the bad group in a.prob
    index.bad.gp[gp] <- bad.gp[gp] + 1

    for(a in 1:antibiotic_classes)
    {
      # Response is different for each antibiotic and depending on
      # which pop it's from
      gp_response[gp,a] ~ dbern(a.prob[a, index.bad.gp[gp]])
    }
  }

  for (v in 1:N_volunteers)
  {
    # probability of belonging to the bad group
    bad.v[v] ~ dbern(prob.of.bad.vol)
    # index the bad group in a.prob
    index.bad.v[v] <- bad.v[v] + 1

    for(a in 1:antibiotic_classes)
    {
      # Response is different for each antibiotic and depending on
      # which pop it's from
      vol_response[v,a] ~ dbern(a.prob[a, index.bad.v[v]])
    }
  }

  for (o in 1:N_outpatients)
  {
    # probability of belonging to the bad group
    bad.o[o] ~ dbern(prob.of.bad.out)
    # index the bad group in a.prob
    index.bad.o[o] <- bad.o[o] + 1

    for(a in 1:antibiotic_classes)
    {
      # Response is different for each antibiotic and depending on
      # which pop it's from
      out_response[o,a] ~ dbern(a.prob[a, index.bad.o[o]])
    }
  }

  # ------------------------

  # Define the priors:
  clin.effect[ncarr] ~ dnorm(0, 0.001)
  clin.effect[nclin] <- -clin.effect[ncarr]

  # Define the priors:
  for(a in 1:antibiotic_classes)
  {
    # probability of being resistant in the good group (less resistances)
    antibiotic.class.effect[a, 1] ~ dnorm(intercept, tau.class)
    logit(a.prob[a,1]) <- antibiotic.class.effect[a, 1]

    # probability of being resistant in the bad group (many resistances) will
    # always be higher than the good group
    antibiotic.class.effect[a, 2] ~ dnorm(intercept.plus, tau.class)
    logit(a.prob[a,2]) <- antibiotic.class.effect[a, 2]

    for(b in 1:2)
    { # good bad
      for (c in c(ncarr, nclin)) # 1, 2!
      {
        ac.effect[a,b,c] <- antibiotic.class.effect[a,b] + clin.effect[c]
        logit(ac.prob[a,b,c]) <- ac.effect[a,b,c]
      }

      a.gp.effect[a,b] <- ac.effect[a,b,gp_clinical]
      a.v.effect[a,b] <- ac.effect[a,b,v_clinical]
      a.o.effect[a,b] <- ac.effect[a,b,o_clinical]
      logit(a.gp.prob[a,b]) <- a.gp.effect[a,b]
      logit(a.v.prob[a,b]) <- a.v.effect[a,b]
      logit(a.o.prob[a,b]) <- a.o.effect[a,b]
    }
  }

  # Prior value for intercept
  intercept ~ dnorm(0, 0.001)

  # Difference between distribution means of "good" and "bad" groups
  diff ~ dgamma(0.001, 0.001)
  intercept.plus <- intercept + diff

  # Probability of being in the bad group is dependent on clinical state
  prob.of.bad.hosp[1] ~ dbeta(1, 1) # Carriage
  prob.of.bad.hosp[2] ~ dbeta(1, 1) # Clinical
  # Probability of being in the bad group
  prob.of.bad.gp ~ dbeta(1, 1)
  prob.of.bad.vol ~ dbeta(1, 1)
  prob.of.bad.out ~ dbeta(1, 1)

  # Prior values for precision
  tau.class ~ dgamma(0.001, 0.001)

  # Convert precisions to sd
  sd.class <- sqrt(1/tau.class)

  #monitor# full.pd, dic, deviance, a.prob, ac.prob, a.gp.prob, a.v.prob, a.o.prob, prob.of.bad.hosp, prob.of.bad.gp, prob.of.bad.vol, prob.of.bad.out, bad.p, bad.gp, bad.v, bad.o, intercept, sd.class
}