Public news sentiment for reducing risk when modelling the COVID-19 growth rate curve

What is it about?

In this work, we explore the statistical concept of model risk, which manifests in the inability to adequately capture the behaviour of the COVID-19 progression growth rate curve. Model risk is mathematically characterised as having two components: the dispersion of the observation distribution, and the structure of a function over time that describes the force of infection (the “intensity function”). Furthermore, we investigated how to include in these population models the effect governmental interventions have had on the number of infected people. This was achieved through the development of an exposure adjustment to the force of infection comprised of a tailored sentiment index constructed from various authoritative public health news reporting, namely major global circulation newspapers, including The New York Times, The Guardian, The Telegraph, and Reuters global blog, as well as international, acknowledged health authorities, i.e. the European Centre for Disease Prevention and Control, the United Nations Economic Commission for Europe, the United States Centres for Disease Control and Prevention, and the World Health Organisation. Our research revealed that the baseline Gompertz model is unable to adequately capture the pandemic evolution for all countries throughout the period of study, and, in addition, models that incorporated the proposed sentiment adjustment are better able to calibrate to the infection spread in all countries under investigation, particularly during the early stages of the pandemic.

Featured Image

Photo by Martin Sanchez on Unsplash

Photo by Martin Sanchez on Unsplash

Why is it important?

Throughout the COVID-19 pandemic, severe contact restriction measures and social mobility limitations had to be put in effect by governments all over the world, to limit the population's exposure to the novel coronavirus. These public health policy decisions were made by considering the output of statistical models for infection rates in national populations. To assist in the public effort, we conducted research where we modelled the temporal evolution of national-level infection counts for the United Kingdom, Germany, Italy, Spain, Japan, Australia and the United States for the period January 2020 to January 2021, in order to better understand the most reliable model structure for the COVID-19 epidemic growth curve. We achieved this by exploring a variety of stochastic population growth models and comparing their calibration, with and without an exposure adjustment, to the most widely used growth model, the Gompertz population model, often referred to in the public health policy discourse during the pandemic. Our analysis showed that policymakers ought to pay particular attention to the per-country characteristics of the growth curves of the number of infections, and if necessary, employ more advanced population growth models such that they include the identified curve features. In this way, model risk will be reduced, provided also that decisions on the selected models are reviewed frequently while continuously incorporating new data that express the pandemic’s evolution. Furthermore, we contributed a novel sentiment index that was obtained from news articles and institutional announcements about the COVID-19 pandemic, as communicated by institutions and national Centres for Disease Control. The inclusion of the sentiment index in the population growth models via an exposure adjustment revealed that at the onset of the pandemic, the in-sample model fit is much better, namely the sentiment adjustment significantly helps the model capture the growth rate of infected cases. This is especially important for model assessment, and assessment of the effectiveness of the applied pandemic countermeasures, protective policies, and the way they were communicated to the public. Our results prove that this research work is particularly useful for designing, communicating and evaluating protective policies during extreme events, as well as scenario generation to better prepare for crises in the future.

Perspectives