Model diagnostics for infectious epidemics using data augmentation Seminar
- Time:
- 15:45
- Date:
- 4 December 2014
- Venue:
- To be confirmed
Event details
S3RI Seminar
The dynamics of disease spread in epidemics, or system change in ecology, can be described through spatio-temporal compartmental models where important characteristics, such as the distribution of sojourn time in a particular infection state, or the spatial transmission kernel, can be represented using a range of different settings. This leads to issues of model criticism and choice within a certain class of models. The analysis and estimation in such systems is typically hindered by the incomplete nature of observed data and the presence of dependencies in the involved processes. Inference can be carried out using Markov chain Monte Carlo under a Bayesian approach, but model assessment often suffers from issues related to prior choice in Bayes factors, or missing data in DIC-type methods. Here we perform model evaluation using methodology based on the properties of the so-called Bayesian latent residuals, which become available through data augmentation and whose sampling distribution is known under the hypothesis of model adequacy. Comparison between two candidate models is also considered using a latent likelihood ratio–type test that avoids problems mentioned above. We illustrate the methods using simulations and also two applications involving an epidemic outbreak and ecological data.
Speaker information
George Streftaris , Heriot-Watt University. Senior Lecturer