Title: Modular modelling: Joining and splitting models with Markov melding
Authors: Robert Goudie - University of Cambridge (United Kingdom) [presenting]
Abstract: Integrating multiple sources of data into a joint analysis yields more precise estimates and reduces the risk of biases introduced by using only partial data. However, it can be difficult to conduct a joint analysis. Often it is only feasible in practice to take a modular approach, with each data source modelled separately, but this leads to information and uncertainty not being propagated. We propose to address this problem using a simple, general method, which requires only small changes to existing models and software. We first form a joint model based upon the original submodels using a generic approach called Markov melding. We then show that this model can be fit in stages, rather than as a single, monolithic model. The approach also enables splitting of large joint models into smaller submodels, allowing inference for the original joint model to be conducted via our multi-stage algorithm. We demonstrate how the approach can be used to integrate longitudinal latent class models with Dirichlet process-based clustering models; to integrate intensive care unit A/H1N1 influenza data and other information sources; and jointly model ecological census and mark-recapture-recovery data.