Models and muddles in the COVID-19 pandemic

Abstract

The COVID-19 crisis is an opportunity for scientists to showcase their skill and the impact that good science can have on society. However, not all scientists have risen to the occasion with the sense of responsibility and accountability
that their work deserves. Scientists worldwide have shown, and continue to show, great enthusiasm regarding the
use of specific scientific tools, mainly modelling and predictive analytics, to estimate how the virus spreads and
behaves and to assess interventions against counterfactual scenarios. In this Commentary, we question whether
the application of these tools has always been appropriately managed by discussing the underlying elements of modelling which need to be understood and evaluated for results to be meaningful and credible. A mathematical model must capture the principles that dictate the dynamics of what is being modelled: assumptions, constraints and relevant natural laws, for example. These principles serve as the ‘rules’ for understanding the results obtained and provide the context within which the model has meaning. Within this context, the model goes beyond being a mere collection of mathematical operations and represents – albeit in idealised or imperfect form – some feature of the actual world. Here we argue that these rules have often been ignored when engaging with the results obtained from mathematical models used for predictive purposes in the COVID-19 pandemic (including policy purposes) and from data-driven models designed via machine-learning methods.
To make our case, we first provide some context of the origins of disease modelling and then offer a ‘current day’
frame of reference which illustrates why caution is needed when employing models for prediction.