R/network.prior.R

In MikeJSeo/bnma: Bayesian Network Meta-Analysis using 'JAGS'

hy.prior.update <- function(network, hy.prior.Eta, hy.prior.bl, hy.prior.cov, hy.prior){

  with(network, {

    if(baseline.risk == "exchangeable"){
      if(is.null(hy.prior.Eta)){
        hy.prior.Eta <- hy.prior.default(network)
      }
    }
        
    if(baseline == "exchangeable"){
      if(is.null(hy.prior.bl)){
        hy.prior.bl <- hy.prior.default(network)
      }
    }
    
    if(!is.null(covariate)){
      if(covariate.model == "exchangeable"){
        if(is.null(hy.prior.cov)){
          hy.prior.cov <- hy.prior.default(network)
        }
      }  
    }
    
    if(type == "random"){
      if(is.null(hy.prior)){
        hy.prior <- hy.prior.default(network)  
      }
    }
    return(list(hy.prior.Eta = hy.prior.Eta, hy.prior.bl = hy.prior.bl, hy.prior.cov = hy.prior.cov, hy.prior = hy.prior))
  })
}
    

network.prior.default <- function(network, mean.d, prec.d, mean.Eta, prec.Eta, hy.prior.Eta, mean.bl, prec.bl, hy.prior.bl, mean.cov, prec.cov, hy.prior.cov, hy.prior) {
  
  with(network, {
    
    if(response == "binomial" || response == "normal"){
      if(is.null(mean.d)){
        mean.d = 0
      }
      if(is.null(prec.d)){
        prec.d = 0.0001
      }
      if(is.null(mean.Eta)){
        mean.Eta = 0
      }
      if(is.null(prec.Eta)){
        prec.Eta = 0.0001
      }
      prior.data = list(mean.d = mean.d, prec.d = prec.d, mean.Eta = mean.Eta, prec.Eta = prec.Eta)
      
      if(type == "random"){
        prior.data$hy.prior.1 <- hy.prior[[2]]
        prior.data$hy.prior.2 <- hy.prior[[3]]
      }
      
      if(baseline != "none"){
        if(is.null(mean.bl)){
          mean.bl = 0
        }
        if(is.null(prec.bl)){
          prec.bl = 0.0001
        }
        prior.data$mean.bl = mean.bl
        prior.data$prec.bl = prec.bl
        if(baseline == "exchangeable"){
          prior.data$hy.prior.bl.1 <- hy.prior.bl[[2]]
          prior.data$hy.prior.bl.2 <- hy.prior.bl[[3]]
        }
      }
      
      if(baseline.risk == "exchangeable"){
        prior.data$hy.prior.Eta.1 <- hy.prior.Eta[[2]]
        prior.data$hy.prior.Eta.2 <- hy.prior.Eta[[3]]
      }
      
      if(!is.null(covariate)){
        if(is.null(mean.cov)){
          mean.cov = 0
        }
        if(is.null(prec.cov)){
          prec.cov = 0.0001
        }
        prior.data$mean.cov = mean.cov
        prior.data$prec.cov = prec.cov
        
        if(covariate.model == "exchangeable"){
          prior.data$hy.prior.cov.1 <- hy.prior.cov[[2]]
          prior.data$hy.prior.cov.2 <- hy.prior.cov[[3]]
        }
      }
    } else if(response == "multinomial"){
      if(is.null(mean.d)){
        mean.d = rep(0, ncat - 1)
      }
      if(is.null(prec.d)){
        prec.d = diag(0.0001, ncat - 1)
      }
      if(is.null(mean.Eta)){
        mean.Eta = rep(0, ncat - 1)
      }
      if(is.null(prec.Eta)){
        prec.Eta = diag(0.25, ncat - 1)
      }
      prior.data = list(mean.d = mean.d, prec.d = prec.d, mean.Eta = mean.Eta, prec.Eta = prec.Eta)
      
      if(type == "random"){
        prior.data$hy.prior.1 <- hy.prior[[2]]
        prior.data$hy.prior.2 <- hy.prior[[3]]
      }
      
      if(baseline != "none"){
        if(is.null(mean.bl)){
          mean.bl = rep(0, network$ncat - 1)
        }
        if(is.null(prec.bl)){
          prec.bl = diag(0.0001, network$ncat - 1)
        }
        prior.data$mean.bl = mean.bl
        prior.data$prec.bl = prec.bl
        if(baseline == "exchangeable"){
          prior.data$hy.prior.bl.1 <- hy.prior.bl[[2]]
          prior.data$hy.prior.bl.2 <- hy.prior.bl[[3]]
        }
      }
      
      if(baseline.risk == "exchangeable"){
        prior.data$hy.prior.Eta.1 <- hy.prior.Eta[[2]]
        prior.data$hy.prior.Eta.2 <- hy.prior.Eta[[3]] 
      }
      
      if(!is.null(network$covariate)){
        if(is.null(mean.cov)){
          mean.cov = rep(0, network$ncat - 1)
        }
        if(is.null(prec.cov)){
          prec.cov = diag(0.0001, network$ncat - 1)
        }
        prior.data$mean.cov = mean.cov
        prior.data$prec.cov = prec.cov
        
        if(covariate.model == "exchangeable"){
          prior.data$hy.prior.cov.1 <- hy.prior.cov[[2]]
          prior.data$hy.prior.cov.2 <- hy.prior.cov[[3]]
        }
      }
    }
    return(prior.data)
  })
}


hy.prior.default <- function(network){
  
  with(network,{
    
    hy.prior <- if(response == "binomial"){
      list("dunif", 0, 5)
    } else if(response == "normal"){
      list("dunif", 0, 100)
    } else if(response == "multinomial"){
      list("dwish", diag(ncat - 1), ncat -1)
    }
    return(hy.prior)
  })
}