The paper introduces Poisson Approximate Likelihood (PAL) methods for fitting stochastic compartmental models. Compartmental modelling is one of the most widespread methods for quantifying the dynamics of infectious diseases in populations; in this paradigm a population is compartmentalised into disease states, the model describes the rates of transitions of individuals between these compartments. The estimation of epidemiological parameters associated with these models, such as the R number, is vital for informing decisions impacting public health.
However, the computational cost of fitting such models to data is substantial and grows with the number of disease compartments and population size; subsequentially, inference often relies on computationally complex and bespoke algorithms. In response to this challenge, PAL methods are:
• Computationally Simple and easy to implement,
• Applicable to a highly flexible class of models,
• Justified by rigorous statistical guarantees in the form of a large population consistency result,
• Orders of magnitude faster and able to provide an overwhelming improvement in model fit to real data in comparison to state-of-the art methods.
Michael Whitehouse is funded by a studentship from Compass – the EPSRC Centre for Doctoral Training in Computational Statistics and Data Science. This work is part of an ongoing collaboration with Prof Nick Whiteley and Dr Lorenzo Rimella; the latter is supported by EPSRC Grant EP/R018561/1 (Bayes for Health, https://www.lancaster.ac.uk/bayes-for-health/).
Whitehouse, Whiteley & Rimella's paper can be read here.