Towards More Reliable Predictions: Multi-model Ensembles for Simulating the Corona Pandemic
- Contact person:
- Paul Davidsson
- Malmö University
- Responsible at Malmö University:
- Paul Davidsson
- Project members:
- Delft University of Technology
- Stockholm University
- Umeå University
- Utrecht University
- Time frame:
- 01 August 2020 - 31 March 2021
- Research environment :
The corona pandemic has hit most countries unprepared, affects the health of many people, determines our everyday life, and leads to a high burden on the health sector. Also, the economic sector is affected by the crisis which results in financial distress or bankruptcy leading to short time working or termination of the employees. To cope with this situation, governments need to make drastic decisions in the short term and implement different measures to counteract the spreading. Yet, decision-makers lack information and experience to make these decisions.
The underlying mechanisms of the ongoing crisis are only partially known as well as potential consequences of different measures and how they might mutually reinforce or cancel out each other. Simulation can be used to systematically investigate different scenarios and measures before implementing them in the society. A variety of prediction models has been used since the start of the pandemic, which have different strengths and weaknesses. They focus on different scenarios (countries, cities, populations), make different assumptions (infection processes, R0), follow different paradigms (macro, micro) and implement different measures (home-work, closing of schools, social distancing, lockdown). A common problem is the researchers’ overconfidence in their models, which makes it difficult for decision-makers to know who to trust.
As a consequence, authorities have started to use different models in parallel and compare their results. It has been shown in other contexts that different models can be combined into multi-model ensembles to provide better results compared to each individual model. We argue that this could be a useful approach also for the prediction of pandemics. The mathematical models used to predict the corona pandemic do not take into account peoples’ individual needs and their behavior in different situations. Moreover, they do not consider the interplay between different parts of society and economy. Agent-based Social Simulation (ABSS) models are able to consider these factors, which allow for more detailed investigations of social phenomena.
The long-term goal of this research is to investigate how ensemble-based simulation can be achieved to combine the strengths of different models.
Research questions that will be investigated include:
- Which simulation models have been developed for the corona pandemic?
- Which characteristics do they have and which assumption do they make?
- How do the models have to be calibrated and adapted to each other?
- How can different models be combined for different purposes?
- What are the prospects and limitations of ABSS for investigating the effects of different measures to manage pandemics?