Performs reversible-jump Markov chain Monte Carlo (Green, 1995)
, specifically the restriction introduced by
Barker & Link (2013) . By utilising
a 'universal parameter' space, RJMCMC is treated as a Gibbs sampling
problem. Previously-calculated posterior distributions are used to
quickly estimate posterior model probabilities. Jacobian matrices are
found using automatic differentiation. For a detailed description of
the package, see Gelling, Schofield & Barker (2019)
.

Performs reversible-jump MCMC, a Bayesian multimodel inference method. The process is simpler than a manual implementation; for instance, all Jacobian matrices are automatically calculated using the madness package. The effort required to find Bayes factors and posterior model probabilities is reduced.

Usage

For each model considered, the user requires a posterior distribution obtained via MCMC or the like. They then define a bijection between its parameter space and the universal parameter space; the likelihood model on the data; and the priors on the parameters. The rjmcmcpost function uses a post-processing algorithm to estimate posterior model probabilities. See ?rjmcmcpost for a simple example using binomial likelihoods.