Chatham House
Printable version

Predictions and Policymaking: Complex Modelling Beyond COVID-19

The COVID-19 pandemic has highlighted the potential of complex systems modelling for policymaking but it is crucial to also understand its limitations.

A member of the media wearing a protective face mask works in Downing Street where Britain's Prime Minister Boris Johnson is self-isolating in central London, 27 March 2020. Photo by TOLGA AKMEN/AFP via Getty Images.

A member of the media wearing a protective face mask works in Downing Street where Britain's Prime Minister Boris Johnson is self-isolating in central London, 27 March 2020. Photo by TOLGA AKMEN/AFP via Getty Images.

Complex systems models have played a significant role in informing and shaping the public health measures adopted by governments in the context of the COVID-19 pandemic. For instance, modelling carried out by a team at Imperial College London is widely reported to have driven the approach in the UK from a strategy of mitigation to one of suppression.

Complex systems modelling will increasingly feed into policymaking by predicting a range of potential correlations, results and outcomes based on a set of parameters, assumptions, data and pre-defined interactions. It is already instrumental in developing risk mitigation and resilience measures to address and prepare for existential crises such as pandemics, prospects of a nuclear war, as well as climate change.

The Human Factor

In the end, model-driven approaches must stand up to the test of real-life data. Modelling for policymaking must take into account a number of caveats and limitations. Models are developed to help answer specific questions, and their predictions will depend on the hypotheses and definitions set by the modellers, which are subject to their individual and collective biases and assumptions. For instance, the models developed by Imperial College(opens in new window) came with the caveated assumption that a policy of social distancing for people over 70 will have a 75 per cent compliance rate. This assumption is based on the modellers’ own perceptions of demographics and society, and may not reflect all societal factors that could impact this compliance rate in real life, such as gender, age, ethnicity, genetic diversity, economic stability, as well as access to food, supplies and healthcare. This is why modelling benefits from a cognitively diverse team who bring a wide range of knowledge and understanding to the early creation of a model.

The Potential of Artificial Intelligence

Machine learning, or artificial intelligence (AI), has the potential to advance the capacity and accuracy of modelling techniques by identifying new patterns and interactions, and overcoming some of the limitations resulting from human assumptions and bias. Yet, increasing reliance on these techniques raises the issue of explainability. Policymakers need to be fully aware and understand the model, assumptions and input data behind any predictions and must be able to communicate this aspect of modelling in order to uphold democratic accountability and transparency in public decision-making.   

In addition, models using machine learning techniques require extensive amounts of data, which must also be of high quality and as free from bias as possible to ensure accuracy and address the issues at stake. Although technology may be used in the process (i.e. automated extraction and processing of information with big data), data is ultimately created, collected, aggregated and analysed by and for human users. Datasets will reflect the individual and collective biases and assumptions of those creating, collecting, processing and analysing this data. Algorithmic bias is inevitable, and it is essential that policy- and decision-makers are fully aware of how reliable the systems are, as well as their potential social implications.

The Age of Distrust

Increasing use of emerging technologies for data- and evidence-based policymaking is taking place, paradoxically, in an era of growing mistrust towards expertise and experts, as infamously surmised by Michael Gove. Policymakers and subject-matter experts have faced increased public scrutiny of their findings and the resultant policies that they have been used to justify.

This distrust and scepticism within public discourse has only been fuelled by an ever-increasing availability of diffuse sources of information, not all of which are verifiable and robust. This has caused tension between experts, policymakers and public, which has led to conflicts and uncertainty over what data and predictions can be trusted, and to what degree. This dynamic is exacerbated when considering that certain individuals may purposefully misappropriate, or simply misinterpret, data to support their argument or policies. Politicians are presently considered the least trusted professionals by the UK public, highlighting the importance of better and more effective communication between the scientific community, policymakers and the populations affected by policy decisions.

Acknowledging Limitations

While measures can and should be built in to improve the transparency and robustness of scientific models in order to counteract these common criticisms, it is important to acknowledge that there are limitations to the steps that can be taken. This is particularly the case when dealing with predictions of future events, which inherently involve degrees of uncertainty that cannot be fully accounted for by human or machine. As a result, if not carefully considered and communicated, the increased use of complex modelling in policymaking holds the potential to undermine and obfuscate the policymaking process, which may contribute towards significant mistakes being made, increased uncertainty, lack of trust in the models and in the political process and further disaffection of citizens.

The potential contribution of complexity modelling to the work of policymakers is undeniable. However, it is imperative to appreciate the inner workings and limitations of these models, such as the biases that underpin their functioning and the uncertainties that they will not be fully capable of accounting for, in spite of their immense power. They must be tested against the data, again and again, as new information becomes available or there is a risk of scientific models becoming embroiled in partisan politicization and potentially weaponized for political purposes. It is therefore important not to consider these models as oracles, but instead as one of many contributions to the process of policymaking.


Channel website:

Original article link:

Share this article


Latest News from
Chatham House

On-Demand Webinar: Better Understand and Manage your Natural Capital