BLOG: Modelling human behaviour and social interactions can help evaluations to explain the unexpected

Modelling human behaviour and social interactions can help evaluations to explain the unexpected

 Zaid Chalabi and Theo Lorenc discuss why simulations based on known human behaviour could help us to understand complex interventions like Glasgow’s urban regeneration programme.

When big sports stadiums are built, designers must ask an important question: ‘How are people going to get out if there’s a fire?’ It’s not enough just to include plenty of exits. Designers need to understand how individuals are likely to act during a large-scale fire, to ensure that exits would not become blocked and the evacuation plan would actually work.

This question can be understood using a method called ‘agent-based modelling’. It involves simulating real-life situations in a ‘bottom-up’ way, with agents who can respond to their environment and to each other, based on what is known about individual behaviours. It’s a more intricate form of modelling than, for example, chronic disease modelling in which people are simply passive victims of environmental exposures and risk factors. Crucially, agent-based modelling sees individuals as decision-makers and tries to create social simulations of their behaviours.

It’s an approach that can’t predict every eventuality, but it gets you closer to likely outcomes than simply relying on hunches or intuition. Like other forms of simulation that harness available evidence, it is cheaper than realising, too late, that you’ve made a big mistake. It helps to avoid the unintended consequences of change in complex systems involving multiple actors.

However, relatively few researchers have tried to apply this approach to large-scale social programmes, such as urban regeneration. For example, Glasgow is currently demolishing vast neighbourhoods and rehousing people in regenerated environments. The process is likely to take up to 20 years, the entire childhood of some residents and a third of the lifetime of many from these deprived communities.  LSHTM’s Matt Egan is part of the 10-year GoWell longitudinal study of this programme. Matt has already detailed initial evaluations in a previous blog and SPHR@L seminar (read Matt’s blog here).

We know that improving the physical fabric of people’s lives can be good for health. But the demolitions and regeneration of the 1960s demonstrated that maintaining and even strengthening the social fabric can be more problematic. As Matt argues, if you destroy a neighbourhood, you need a strategy to support those who rely on it most.

That’s where we believe there is a place for agent-based modelling. An agent-based model simulating the population of the regenerated areas could help us to anticipate the behaviour of different types of residents over time.  We could also investigate how the model matches up with and explains what has been observed so far in the Glasgow regeneration project. Testing the model against the reality would help to validate it and also refine it for use elsewhere.

How might agent-based modelling help researchers? It could provide an important tool to support evaluation. For example, the GoWell evaluation of the demolition programme expected to find a negative health impact on residents in the initial phase, resulting from remaining in a deteriorating environment and being left behind by friends, family and neighbours. It didn’t. As Matt explains in his blog, the evaluation team found virtually no change in the mental and physical health of residents living for two years in an area that was being destroyed around them.

There are lots of possible reasons for this surprising finding; perhaps things had already got so bad on the estates that the demolition programme did not make matters much worse.  However, agent-based modelling, which builds in known evidence about human actions and interactions in certain circumstances, might help to disentangle and inform these findings. Such modelling could also potentially help policy strategists to plan such big projects far better in the future, by simulating how large-scale changes to the environment might impact on residents’ behaviour and wellbeing.

The need for such insights is clear for the later stages of the Glasgow regeneration project. Research is beginning to suggest that the mental health of white Scottish residents seems largely unthreatened by the disruption of relocation. This may be because these residents may already have had weak community links. In contrast, immigrant communities, whose cohesiveness probably protected their health from the adversity of living on slum estates, may now be heading for trouble as their networks risk being  broken up in new housing configurations.  A good agent-based model might be able to simulate such potential downsides for different populations and help policy-makers to build in alleviating measures.

Modelling cannot answer all questions. The world is complex, particularly when you are modelling a society and trying to build in all the different decisions that people make in various circumstances. Modelling is also only as good as the evidence base. But if we start to build models, they will gradually get better and more informative as researchers gather the missing data.

Dr Zaid Chalabi is Senior Lecturer in Health Impact Analysis and Mathematical Modelling at LSHTM. Dr Theo Lorenc is Honorary Lecturer at LSHTM and Provost Fellow at the Department of Science, Technology, Engineering, and Public Policy (STEaPP), University College London. On May 20, they will present a seminar at the NIHR School for Public Health Research at LSHTM, entitled ‘Can Agent-Based Models Inform the Evaluation of Complex Natural Experiments?’