Skip to content

Big shifts for behavioural science: working with complexity and systems

Comment & Opinion 10th Dec 2024

This is the first piece in our series explaining the four big shifts in behavioural science that BIT is aiming to achieve.

In 2023, we published our manifesto, which set out ten proposals for where the field of behavioural science should go next. Now, BIT and Nesta have taken things further. We’ve identified four main ways we can advance these proposals through our own work. The first of these ‘big shifts’ concerns working with complexity.

“Be more capable of working with complexity and at the level of systems. Consider how behaviours play out in societies and other networks.” 

Behavioural science has the potential to help address some of the world’s toughest problems. But the nature of those problems challenges the main assumptions underlying existing behavioural science approaches: a tight focus on a target behaviour, predictable effects, and stability over time. 

Our manifesto makes the case for behavioural scientists to go beyond this narrow thinking and ‘see the system’ as a whole. However, talking about complexity and the importance of systems is not simply a statement that issues are hard to address. Rather, it’s an approach that can offer new, credible and practical ways of doing things differently.

And much is being done – at BIT and beyond – to adopt this approach. For one, we’ve created a nonprofit entity, System 2, to do deep thinking on how to achieve systemic change. Organizations such as Busara, the Centre for Public Impact and Sistema Futura have all been looking at how to fuse behavioural science with complexity and systems thinking as well. None other than the co-creator of Nudge, Richard Thaler, has set the ambition that behavioural scientists should be understanding and designing systems, rather than just nudging them. 

But everyone would admit that this task is still in progress – and that means we aren’t offering a completely defined toolkit for the future. Rather, we’re showing how we’ve started to work in this area, the progress made to date, and how we hope to learn as we go. There are three main behavioural science activities that can benefit: understand, intervene, and evaluate.

Understand     

Behavioural scientists can ask many questions when they first look at an issue, like ‘What is the issue? Who decides?’ But often, it is essential to understand who is acting and what factors are influencing their behaviour. If there are many interactions between actors and factors, that requires you to see the system. A powerful way of achieving this vision is to create a visual map of these relationships.     

This is what BIT did recently when considering how to improve the hospital discharge process in East London. In January 2023, 1 in 7 inpatient hospital beds were occupied by people fit for discharge, causing delays for patients who needed to be admitted and increasing the risk of hospital-acquired infections

We recognised that the discharge process was a complex system, comprising many decisions by individuals and organisations connected with the hospital. That means applying a single nudge may not have made much difference, unless it was well-targeted. Working with Homerton University Hospital, we mapped the connections to show how they reinforced or mitigated each other. This analysis revealed several harmful backfires that could have occurred in the system if we had launched interventions that looked otherwise promising. The map we produced is shown below.  

 

Flowchart visualizing factors causing discharge delays in a hospital, highlighting interconnections between communication, staffing, record-keeping, and leadership issues.
 

Download a high res version of this diagram.

This kind of in-depth analysis builds on earlier work BIT has done to map systems, like what drives the use of single-use plastics in the Solomon Islands. It’s a more formal approach to the kind of systemic analysis we’ve done to understand how to get to net zero emissions, what causes patterns of violence or the barriers to overcoming disadvantage.

Our hospital discharge work revealed something else. The map itself was not the only benefit that emerged from the process – the very act of running this exercise brought the participants a deeper understanding of the work they do. 

We believe that helping people understand complex systems should be a priority for behavioural scientists. These systems challenge our intuitive understanding of the way the world works. That means we need to create new heuristics and habits to help people visualise how, for example, a big initiative can have little impact, while a throwaway comment can transform a national conversation. 

We’re still testing the best ways of building this understanding, but it should be part of a wider set of tools that help people see that multiple outcomes are possible – and how an intervention could change the nature of the problem itself. These tools can include:

  • Using pre-mortems, ‘dark logic’ exercises, and other frameworks to challenge assumptions that an intervention will have the desired effects.  
  • Making predictions about the likely effects of interventions and, crucially, feeding back results to predictors when they are known.
  • Using other foresight techniques, including scenarios, to consider a broader range of likely adaptations and possible changes. 

These changes reflect the reality that, when dealing with a complex system, the best option may be to produce something that is ‘roughly right’ and which can adapt to possible changes. 

While it may be tempting to adopt a rigid plan that aims for narrow and precise goals, this may result in brittleness and disappointment instead. Yet, this way of thinking cuts against the way that most policies and products are made. Behavioural scientists should focus on how to change those assumptions and build new skills instead.

Intervene 

Complexity does not mean you cannot act. Sometimes a map that shows a bewildering array of connections can be disheartening – and sometimes actors can use complexity as an excuse to do nothing

Instead, understanding complex adaptive systems can offer a new way of seeing how to intervene. It can show how focusing on a specific part of a system (for example, the adoption of heat pumps) can drive wider changes. It can also show that trying to focus only on systemic factors can also be ineffective, since collections of individual behaviours can self-organise and create new structures themselves. Here are three possible ways forward:

Spreading behaviours using the power of networks 

Recent years have seen increasing interest in how to ‘seed’ behaviours with individuals or small groups in a network so that they spread widely. For example, one study in the US found that encouraging a small set of students to take an anti-bullying stance reduced school conflict by 30%. The impact was greater when the group had a higher number of ‘social referent’ students who were influential in the school network. 

BIT is now adapting and evaluating this approach in the UK, through its Grassroots programme. In participating schools, all students aged 11-15 complete a survey to identify a seed group of around 30 pupils that represent the student body. They attend around ten sessions every two weeks to create activities to reduce conflict in schools. The idea is that having nominated students promote these activities will increase the chance that they spread through the school network.     

Using leverage points   

Sometimes seeing the system can reveal certain points where intervening could have a major effect on the desired outcome. The change may seem minor when considered in isolation, but the structure of the network means its effects are large – a new spin on how ‘small changes can have a big impact’

In our hospital discharge project, we identified three leverage points that would have the greatest impact on the system surrounding discharges: 

  • Providing timely information relevant to the discharge process
  • Following delayed discharge, making performance data accessible and providing feedback 
  • Changing staff perceptions about the relative priority of discharge-related tasks 

Of course, systems can exist at various levels, including businesses, cities, regions, national institutions, or global trade. 

System 2 analysed the Employment Services System, which operates at the national level in Australia. It identified leverage points, such as changing the procurement system and success metrics for service providers, and using intermediaries to help people transition into work (rather than relying on employers). After a co-design process, the second leverage point was developed into a proposed intervention called the Jobs and Training Deal.        

Targeted changes to the ‘rules of the game’

Finally, behavioural scientists can identify a change to a key parameter or variable in the system that will result in desired behavioural outcomes. A good example here might be the UK’s tax on sugared drinks, which identified the tapering of the tax by sugar levels as a key aspect that would encourage manufacturers to reduce sugar levels. If that happened, consumers would not be forced to switch drinks – and inertia would lead to reduced sugar consumption on its own. 

This example shows that the insights into the psychology of individuals and organisationals are important inputs to designing or tweaking systems, rather than being distractions from the task. 

For example, BIT has been measuring how far consumers are affected by online ‘dark patterns’, by specific features in sports betting apps, and by gamification ploys in investment platforms. These insights can be assembled into a coherent picture of consumer psychology that informs more targeted and sophisticated regulatory strategies.    

Regulation is not the only way that systems are managed. When it comes to the economy, for instance, institutions also have to fashion narratives and manage expectations. Our work has shown that the way these institutions act and speak is an important input to the system. 

Working with the Bank of England, we found people had higher trust and comprehension when economic information was presented in ways they could easily relate to their own lives – like the cost of trips abroad. These kinds of insights could be combined with new studies on how households notice and understand financial signals from others, and how the heuristics they use can have systemic impacts.     

More generally, behavioural scientists should move away from thinking about standalone interventions when dealing with complex systems. 

A better perspective is to see the overall impact of collections of policies, not just the specific indicators that have been preselected. That points towards system stewardship as a goal, where behavioural scientists try to shape the way a system works, and respond to how it adapts, rather than focusing on trying to change narrow, prespecified behaviours.   

Evaluate

BIT has promoted the use of randomised controlled trials (RCTs) throughout its existence. After running more than 800 of them, we know that they can be done more cheaply and quickly than people realise. When dealing with stable situations and simpler problems, they should be used more often than they are – our TESTS report shows how. But they struggle to deal with complex systems

That doesn’t mean we should abandon them. Cluster RCTs have been central to evaluating how behaviours spread. These trials test different approaches in many separate systems, like cities or online communities, and then compare the results. 

For example, a series of cluster RCTs in villages in Honduras have found that behaviours spread more effectively if they are seeded with the nominated friends of randomly selected individuals, rather than with the randomly selected individuals themselves.

RCTs can also be set up to measure systemic effects, within reason. For instance, we tested a text message intervention aimed at parents, and found that it shifted the intended outcome of encouraging more conversations about science. 

But the RCT also found that it reduced other beneficial parental behaviours, like checking that children were studying or making them go to sleep on time. These conclusions were only possible because these substitution effects were anticipated in the ‘understanding’ phase. 

Embracing theory-based evaluation       

However, dealing with complex systems also requires other approaches, especially in contexts where RCTs aren’t possible. At BIT and Nesta, we think there is particular promise in the work we’re doing using theory-based evaluation.  

Theory-based evaluation (TBE) is a powerful way of understanding not just whether something worked, but how and why it worked. It helps us to develop an overall sense of the contribution that a set of changes made, and to trace the journey of how that impact happened. It provides a way of handling the many interacting factors that are present in systems. 

If you take the example of our Grassroots programme, a TBE approach might also examine each of the connections we imagine are necessary for success. Did the nominated students attend the sessions? Did they create activities? Did other students notice these activities? Did attitudes towards conflict change? What are the most convincing explanations for what we’ve observed?

By looking at these connections, TBE helps us see which aspects of an intervention are working as expected and which need adjustment. It can also reveal unexpected pathways to success. For instance, we might find that some schools reduced conflict not through the planned activities, but because the programme prompted teachers to change how they handled disputes.

Since TBE considers multiple pathways, looks at the conditions for success and the role of context, and shows unintended consequences, it can deal well with complex systems. In other words, dealing with complexity needs more flexible ways of evaluating impact. That shift does not mean abandoning the scientific method, but rather strengthening it. 

Making connections

This big shift around working with complexity has strong links to the other three. 

The shift towards being more iterative and adaptive responds to the unexpected changes that complex systems can produce. A powerful way to understand systems is to explore people’s experiences of them – in other words, to be more participatory.

And finally, many of the systems we will consider are social networks, supporting the shift to be more attentive to the social drivers of behaviour. We will cover this dimension next in our series.

Authors

Talk to us

Get in touch to see how our deep understanding of human behaviour and evidence-led problem solving can help you.

Get in touch

Design and development by Soapbox.