Math Biology Seminar: Jessica Stockdale

A major challenge in mathematical analysis of infectious diseases is that the epidemic process is usually only partially observed. Although we might be able to identify when an individual became symptomatic, rarely can we observe when infection began or from whom it was transmitted. This means that the likelihood of the observed data is computationally intractable for any more than a handful of infected cases. Although data-augmented Markov Chain Monte Carlo methods are generally considered the gold standard for analysis of partial epidemic data since they employ a tractable augmented likelihood, they also often struggle for large population sizes. I will describe a new approach which seeks to instead approximate the likelihood by exploiting the underlying structure of the epidemic model, without the need for augmentation. We regard this approach as a useful and adaptable addition to the toolkit for analysing infectious disease data, and I will provide examples of applications to real outbreaks and a variety of disease transmission models.