Likelihood-based Inference for Stochastic Epidemic Models
summary
Claims an exact likelihood route for continuous‑time stochastic SIR by treating missing event times as latent variables and exploring them with data-augmented MCMC. Uses CTMC likelihoods, gamma conjugate updates, Gibbs steps and a fast branching‑process proposal to impute infection/recovery times; also supports observed dynamic Bluetooth contact networks. Implements conditional simulation of latent events and runs on a laptop in minutes for datasets of order hundreds (college Bluetooth flu study), returning parameter estimates with uncertainty and scalable performance.
AI-written summary · about
chapters (11)
- 0:00Introduction to SIR models
- 2:50Stochastic model mechanics
- 7:02Network dynamics and data
- 10:28Inference challenges
- 13:22Likelihood for Markov chains
- 17:12Mathematical approaches
- 22:57Data augmentation MCMC
- 26:48Dynamic network integration
- 38:47Fitting observed data
- 49:21Branching process extensions
- 56:15Discussion and Q&A