Partition sampling in Bayesian RaCE-NMA models (internal use only)

This function implements a reversible jump MCMC procedure for updating the parameter partition in Bayesian Rank-Clustered Estimation for Network Meta-Analysis models in the case when we assume correlation among treatment effects. For internal use only.

sample_partition_correlation(
  mu_hat,
  J,
  nu,
  g,
  K,
  mu0,
  sigma0,
  cov,
  tau = tau,
  b_g = 0.5,
  d_g = 0.5,
  logdmvn = NULL
)

Arguments

mu_hat

A vector of estimated average relative intervention effects based on a previous NMA. The jth entry is the effect of intervention j.

J

A numeric indicating the total number of interventions being compared.

nu

A vector indicating current values for nu in the Gibbs sampler.

g

A vector indicating current values for g in the Gibbs sampler.

K

A vector indicating current values for K in the Gibbs sampler.

mu0

The hyperparameter mu0, usually specified as the grand mean of the average intervention effects.

sigma0

The hyperparameter sigma_0, usually a large number as to be minimally informative.

cov

A variance covariance matrix of relative intervention effects based on a previous NMA. The (i,j) entry is the covariane between intervention i and j's effects.

tau

The standard deviation of the Metropolis Hastings proposal distribution.

b_g

The probability of "birth"ing a new partition cluster, if possible. Default is 0.5.

d_g

The probability of "death"ing an existing partition cluster, if possible. Default is 0.5.

logdmvn

An optional function of (x, mean) returning the multivariate normal log-density with covariance cov, as built by make_logdmvnorm. Supplying it avoids re-factorising cov on every call. If NULL, one is built from cov.

Value

A list containing updated values for g, nu, and K.

Examples

sample_partition_correlation(mu_hat=c(1,1,1),J=3,nu=1,g=c(1,1,1),K=1,mu0=1,
 sigma0=1,cov=diag(rep(.1,3)),tau=0.1)
#> $g
#> [1] 1 1 1
#> 
#> $nu
#> [1] 1
#> 
#> $K
#> [1] 1
#>