powder: Marginal likelihood estimation with thermodynamic integration and steppingstone sampling. power posteriors via differential evolution in r. When presented with several competing formal models, one is required to select between these different explanations: a process commonly known as model selection. One of the more robust methods of performing model selection is through marginal likelihoods, which can be used to compute Bayes factors. powder estimates marginal likelihoods via thermodynamic integration (Friel & Pettitt, 2008; Lartillot & Philippe, 2006) and steppingstone sampling (Xie, Lewis, Fan, Kuo, & Chen, 2011) by sampling from power posteriors using differential evolution MCMC.