Scope 02

Build Behavioral Science into Organizations

There has been too little focus on using behavioral science to shape organizations themselves, as opposed to increasing how much an organization uses behavioral science to achieve its goals. We need to talk about how to set up a dedicated behavioral team, and more about how behavioral science can be integrated into an organization’s standard processes. For example, as well as trying to ensure that a departmental budget includes provisions for behavioral science, why not use behavioral science to improve the way this budget is created (e.g., are managers anchored to outdated spending assumptions)?

But we need to understand this new way of thinking maps against existing debates about how to set up a behavioral function in organizations. We propose that doing so reveals six main scenarios, as shown in the diagram below.

Behavioral Science Incorporated into Organizational Processes

In “Call For The Experts”, an organization has concentrated behavioral expertise, but there are also prompts and resources that allow this expertise to be integrated more into “business as usual”. Expertise is not widespread, but access to it is. This setup could mean that processes stimulate demand for behavioral expertise that the central team can fulfill. That team may also have the institutional support to proactively monitor activities and respond quickly to specific crises.

In “Behavioral Entrepreneurs”, there is behavioral science capacity distributed throughout the organization, either through direct capacity building or recruitment. The problem is that organizational processes do not support these individual pockets of knowledge. Therefore, those with expertise find it hard to apply ideas in practice, evaluate their effects, share findings, and build learning.

Finally, a “Behaviorally-Enabled Organization” is one where everyone has knowledge of behavioral science and its use throughout the organization, which supports the integration of this knowledge and use in policies, processes, and the behavioral science is not seen as separate from business as usual; it is the business as usual. While this scenario is most desirable, it also requires the most resources.

Most discussions make it seem like the meaningful choice is between the different columns in the table above – how to organize dedicated behavioral science resources. Instead, the more important move is from the top row to the bottom row: moving from projects to processes, from commissions to culture. A useful way of thinking about this task is about building or upgrading the “choice infrastructure” of the organization.

Working out how best to build the choice infrastructure in organizations should be a major priority for behavioral science. One advantage to this approach is that it can help organizations address problems with scaling interventions. Already we can see some features will be crucial: reducing the costs of experimentation; creating a system that can learn from its actions; and developing new and better ways of using behavioral science principles to analyze the behavioral effects of organizational processes, rules, incentives, metrics, and guidelines.

The Behavioural Insights Team / A Manifesto for Applying Behavioral Science

METHODS

04 PUT RCTs IN THEIR PLACE

Randomized Controlled Trials (RCTs) have been a core part of applied behavioral science, and they work very well in relatively simple and stable contexts. However, they can fare worse in complex adaptive systems, where many shifting connections can make it difficult to keep a control group isolated, or where a narrow focus on predetermined outcomes may neglect others that are important but difficult to predict.

We can strengthen RCTs to deal better with complexity by gaining a better understanding of system interactions and anticipating how they may play out, perhaps through “dark logic” exercises that trace potential harms rather than benefits. We can set up RCTs to measure diffusion and contagion in networks, either by creating separate online environments or by randomizing real-world clusters, like separate villages.

Finally, we can build feedback and adaptation into the RCT design, allowing adjustments to changing conditions. Options include:

• Using two-stage trial protocols
• Evolutionary RCTs
• Sequential multiple assignment randomized (SMART) trials
• Bandit algorithms that identify high-performing interventions and allocate more people to them

We can also use behavioral science to enhance alternative ways of measuring impact, particularly agent-based modeling, which simulates interactions between different actors in a system. The agents in these models are mostly assumed to operate on rational choice principles, presenting an opportunity to incorporate more evidence about the drivers of behavior, such as habits and social comparisons.

Full detail on page 45

05 REPLICATION, VARIATION, ADAPTATION

The “replication crisis” of the last decade has sparked intense debate and concern about the reliability of behavioral science findings. Poor research practices were a major cause of this crisis; however, many have improved as a result.

We need to secure and build on these advances, moving towards a future where meta-analyses of high-quality studies, including deliberate replications, are used to identify the most reliable interventions, develop an accurate sense of their likely effects, and provide weaker options. We have a responsibility to discard ideas if solid evidence shows they are shaky and to offer a realistic view of what behavioral science can accomplish.

This responsibility also requires a hard conversation about heterogeneity in results. The complexity of human behavior creates so much statistical noise that it is often hard to detect consistent signals and patterns. The main drivers of heterogeneity are:

1. Contexts influence results
2. The effect of an intervention may vary greatly between groups within a population

These factors complicate the idea of replication itself: a “failed” replication may not indicate that a finding was false, but rather how it exists under some conditions and not others.

Applied behavioral scientists need to set a much higher bar for claiming that an effect holds true across many unspecified settings. There is a growing sense that interventions should be discussed as hypotheses that were true in one place and may need adapting to be true elsewhere.

We need specific proposals as well as normative changes. The first concerns data collection: expand studies to include a wider range of contexts and participants, and gather richer data about them. To date, only a small minority of behavioral studies have provided enough information to see how effects vary. Coordinated multi-site studies will be needed to collect enough data to explore heterogeneity systematically; “crowdsourced” studies offer particular promise for testing context and methods.

Behavioral scientists also need to improve their judgment regarding how much an intervention’s results were linked to its context and how much adaptation it may need. We should use and modify frameworks from implementation science to develop such judgment. Finally, we need to codify and cultivate the practical skills that successfully adapt interventions to new contexts; expertise in behavioral science should not be seen as simply knowing about concepts and findings in the abstract.

Full detail on page 51

06 BEYOND LISTS OF BIASES

The heterogeneity in behavioral science findings indicates that our underlying theories need improvement. We lack good explanations for why findings vary so much, which is part of a wider “theory crisis” in psychology.

The first concern is that theories of behavior often try to explain complex and wide-ranging phenomena. This variability can lead to vague and weak theories that generate many different hypotheses, some of which may contradict each other, making them hard to test.

The second concern is that while theories can make specific predictions, they are often disconnected from each other and from a deeper, general framework that can provide broader explanations. This issue affects behavioral science through heuristics and biases, which are often presented as lists of standalone curiosities, leading to overconfident thinking that targeting a specific bias in isolation will achieve a certain outcome.

Focusing on lists of biases distracts us from answering core underlying questions, such as when one or another bias applies and which are widely applicable versus highly specific. These are practical questions when taking an intervention to new places.

A priority for responding to this challenge is to develop practical theories that:

• Fill the gap between day-to-day working hypotheses and comprehensive attempts to find universal explanations
• Are based on data rather than pure theorizing
• Generate testable hypotheses that can be disproved
• Specify the conditions under which a prediction applies or does not
• Offer actionable steps toward solving real-world problems

Resource rationality is a good example of a practical theory, starting from the premise that people make rational use of their limited cognitive resources. Given the cost of thinking, people seek solutions that balance choice quality with effort. These principles provide a systematic framework for building useful models of human behavior.

A recent study has shown how these models can predict responses to different nudges in certain contexts and can be integrated with machine learning to create an automated method for constructing optimal nudges, demonstrating practical benefits from applying a particular theory.

Full detail on page 61

07 PREDICT AND ADJUST

Hindsight bias occurs when people feel “I knew it all along,” even if they did not. When experiment results come in, hindsight bias may lead behavioral scientists to believe they predicted them or quickly find ways to explain why they occurred. This bias is problematic as it breeds overconfidence, impedes learning, dissuades innovation, and prevents understanding of what is truly unexpected.

In response, behavioral scientists should establish a standard practice of predicting experiment results and receiving feedback on their predictions. Hindsight bias can flourish if we have opaque expectations about study results or if we do not check or remember the state of knowledge before an experiment. Making predictions provides clear feedback that is more likely to trigger surprise and reassessment rather than hindsight bias.

More studies are explicitly integrating predictions, but barriers remain. People may resist challenges to their self-image, predicting may seem like an additional task, and the benefits are often future-oriented.

We propose to make predicting easy by incorporating it into standard organizational processes, minimizing threats to predictors’ self-image by making predictions anonymous, providing concrete prompts for learning and reflection, and building learning from prediction within and between institutions.

Full detail on page 69

VALUES

08 BE HUMBLE, EXPLORE AND ENABLE

Behavioral scientists, like other experts, may overconfidently rely on decontextualized principles that do not match real-world settings for behavior. Deeper inquiry can reveal reasonable explanations for what seem to be behavioral biases. In response, those applying behavioral science should:

• Avoid using the term “irrationality,” which can limit attempts to understand actions in context
• Acknowledge that diagnoses of behavior are provisional and incomplete (epistemic humility)
• Design processes and institutions to counteract overconfidence

A Common Theme in Behavioral Science

A common theme through these ideas is the need for more and better inquiry into behaviors in context, rather than making assumptions. Open-ended qualitative exploration of the context and drivers for behaviors is not new to the behavioral sciences. However, three areas demand particular focus in the future:

1. Pay greater attention to people’s goals and strategies, and their own interpretations of their beliefs, feelings, and behaviors.
2. Reach a wider range of experiences, including marginalized voices and communities, understanding how structural inequalities can lead to expectations and experiences varying greatly by group and geography.
3. Recognize how apparently universal cognitive processes are shaped by specific contexts, thereby unlocking new ways for behavioral science to engage with values and culture.

Level of Involvement in Initiative

Examples of Initiatives

• Healthy foods placed prominently in canteen.
• Employees deciding they want healthy foods to be prominent and developing arrangement.
• Creating regular automatic bank transfers to savings account.
• Gyms introducing commitment devices to increase usage.
• Users working with gyms to develop commitment devices that actively remind users of their exercise goals.
• Initiator decides that the best option is to teach people heuristics to increase savings.
• Individual reads about heuristics and constructs own heuristics for savings goals.
• Placing letters to be posted on door handle.

In addition, more can and should be done to broaden ownership of behavioral science approaches. Many behavioral science applications have been quite top-down, with a “choice architect” enabling certain outcomes. One route is to enable people to become more involved in designing interventions themselves. “Nudge plus”, “self nudges”, and “boosts” have been proposed as ways of doing this. Reliable criteria are needed to decide when enabling approaches may be appropriate, including:

• Whether the opportunity to use an enabling approach exists.
• Ability and motivation.
• Preferences.
• Learning and setup costs.
• Equity impacts.
• Effectiveness (recognizing evidence on this point is still emerging).

These new approaches should not be seen simplistically as “enabling” alternatives to “disempowering” nudges. Instead, we need to consider:

1. How far the person performing the behavior is involved in shaping the initiative itself.
2. The level and nature of any capacity created by the intervention.

People may be heavily engaged in selecting and developing a nudge intervention that nonetheless does not trigger any reflection or build any skills. Alternatively, a policy maker may have paternalistically assumed that people want to build up their capacity to perform an action, when in fact they do not. This is the real choice to be made.

A final piece missing from current thinking is that enabling people can lead to a major decentering of the use of behavioral science. If more people are enabled to use behavioral science, they may decide to introduce interventions that influence others. Rather than just creating self-nudges through altering their immediate environments, they may decide that wider system changes are needed instead. A range of people could be enabled to create nudges that generate positive societal change, as happened for the “Fair Tax Mark” in the UK.

Full detail on page 75

SCOPE

01 USE BEHAVIORAL SCIENCE AS A LENS

We need to see behavioral science as a lens that improves the view of any public and private issue, rather than as a tool that we sometimes pick up. Making this change will help break the self-sustaining pattern whereby demand for behavioral science, and the tools we have developed, has pushed work towards downstream interventions and away from structural changes.

The recent surge in applying behavioral science to practical issues has made a measurable difference across many domains. The approach has been adopted by public sector bodies at the local, national, and supra-national level, and by private companies large and small. They have improved outcomes in health, education, sustainability, transport, and financial behavior, among many areas. Many of these improvements have come at relatively low cost.

Despite these achievements, objections have emerged. A common one is that there’s been a focus on tractable and easy-to-measure changes, at the expense of bigger impact on major issues. Behavioral science, it’s claimed, has just been tinkering around the edges of fundamental problems.

We agree with the challenge that behavioral science can and should do more. Every day, new policies cut against well-established evidence of how people behave. Services are shaped in ways that people cannot navigate. Products are launched with fundamental misconceptions about how people are likely to approach them. There are fewer prominent examples that clearly show how governing policies and systems have been designed using concepts from behavioral science, as opposed to specific aspects of how those policies were presented or structured.

So, how to move forward? Step one is to realize that the strengths that have brought success may also be holding behavioral science back. To explain, let’s go back to the start of the current phase of applied behavioral science (around 2008-2012), when there was a pressing need to demonstrate clear results and build credibility. That pressure led us and others to form standard ways of applying behavioral insights. These approaches generally have a common set of features. The standard series of actions looks something like this:

• scoping the issue and exploring drivers of behavior
• defining a specific target behavior that can be measured reliably
• generating evidence-based interventions to change the target behavior
• creating a robust experimental design to test the intervention’s effects
• if desired, taking the intervention to new places (e.g., scaling)

These actions are usually presented in a step-by-step (linear) way, although most guides stress that people can loop between stages.

Over the past decade, this kind of approach has tended to produce:

• downstream interventions that concern how specific aspects of a policy, product or service are designed
• a focus on discrete behaviors by actors (e.g. people, businesses), considered mostly in isolation.

Perhaps a good example is BIT’s project to reduce missed hospital appointments in the UK. This work identified the wording of text message reminders as an opportunity for improvement. We ran two randomized controlled trials in London, which found that a message referring to the cost of a missed appointment for the health system reduced no-shows from 11.1% to 8.5% – a 25% relative change. This low-cost change was then taken up by other health providers around the world.

As this example shows, the approach is a strong one. There’s a neatly-defined problem, a specific intervention, and a measurable outcome. However, there’s a clear limit to what this approach can achieve. These strengths mean that there are still so many improvements that this approach could achieve. For instance, one priority should be to clear the vast administrative burdens (or sludge) that prevent people – particularly those with fewest resources – from understanding or accessing government services.

However, the justified focus on these clear and credible results about downstream impact also becomes self-reinforcing. People start to think that this is the sole way that behavioral science can be applied. In turn, this perception shapes demand: only certain kinds of problems are seen as ones suitable for behavioral scientists.

Opportunities, skills, and ambitions have been constricted as a result. In general, practitioners have focused more on expanding the application of some tried and tested interventions to new areas, and less on exploring new ones or getting a deeper understanding of familiar ones.

We need a rebalancing.

Behavioral science also has much to say about broader, larger issues in society like discrimination, pollution, or economic mobility, and the structures that produce them. Behavioral science has the potential to fundamentally change how we understand the factors shaping behavior and therefore how we constitute an issue and what is possible.

Take the economy: behavioral science can show how to regulate markets differently; how to design taxes to drive wider behavioral changes; and even offer a vision for the future behavioral economy as a whole.

As this list shows, there are examples of how behavioral science has tackled more complex, structural, upstream issues. But these examples are harder to communicate because they often deal with the fluid, murky, and fractured narratives of politics and policy-making. Unlike the clear, linear stories of the approach outlined above, the contribution of behavioral science may be difficult to trace or may play out over a long timeframe. It’s easier to talk about the neat narratives instead, particularly since many people are aware of and curious about the idea of nudging, which is often associated (inaccurately) with small presentational tweaks only.

The Behavioural Insights Team / A Manifesto for Applying Behavioral Science

The wide potential scope of applied behavioral science is an idea that BIT has promoted consistently since its creation. But the self-reinforcing limiting factors we outlined have proved strong. Now the increasingly urgent question is: how can we successfully change behavioral science itself?

Our answer is to consider the proposals in this manifesto, which aim to create a package that can achieve that change. We can start by trying to switch the metaphors or frames through which we perceive behavioral science itself.

Behavioral science should be understood as a lens that enhances the view of any public and private issue, rather than as a tool that we sometimes pick up.

Changing the Metaphor

The trends we highlight above have tended to reinforce this tool metaphor, which encourages this way of thinking:

• Behavioral science is a specialist tool that is applied to certain kinds of problems – and not others. Often these are defined as delivery issues (e.g., “How do we structure this message?”), but sometimes it can be used to solve a behavioral problem as an alternative to more traditional approaches like rules and incentives.

These implications lead us down the wrong path. Instead, behavioral science should be understood as a lens that can be applied to any public and private action. This change offers several advantages:

• A lens metaphor shows that behavioral insights can enhance the way we see policy options (for example, revealing new ways of structuring taxes), rather than just acting as an alternative tool.
• A lens metaphor conveys that the uses of behavioral science are not limited to creating new interventions, but also include creating new ways of understanding and diagnosing effects. It emphasizes the behavioral diagnosis of a situation or issue, rather than pushing too soon to define a precise target outcome and intervention.
• Specifying that this lens can be applied to any action conveys the error of separating out behavioral and non-behavioral issues: most of the goals of private and public action depend on certain behaviors happening (or not). Behavioral science should therefore be integrated into an organization’s core activities, rather than acting as an optional specialist tool.

In one sense, this proposal is about returning to first principles. Back in 2010 we emphasized that civil servants need to better understand the behavioral dimension of their policies and actions, and also stressed how behavioral science powerfully complements and improves conventional policy tools. But, for all the reasons above, this aspect has been less prominent over the last decade.

Other metaphors apart from a lens would be powerful as well. For example, moving from choice architecture to choice infrastructure effectively highlights the broader, embedded nature of our behaviors. The point is that behavioral science itself shows us the power of framing: the metaphors we use shape the way we behave, and therefore can be agents of change.

Metaphors are particularly important because the task of broadening the use of behavioral science requires making a compelling case to decision makers. Behavioral science practitioners need to understand their audience and then shape their offers accordingly.

In a way, the metaphor of behavioral science as a tool that produces clear results has served the field well over the last decade – it has established credibility and acceptance in a defined area. The challenge now is to expand beyond that area, allowing behavioral science to fulfill its potential before the self-reinforcing cycle becomes too hard to break.

02 BUILD BEHAVIORAL SCIENCE INTO ORGANIZATIONS

There has been too little focus on using behavioral science to shape organizations themselves, as opposed to increasing how much an organization uses behavioral science to achieve its goals. We need to talk less on how to set up a dedicated behavioral function, and more about how behavioral science can be integrated into an organization’s standard processes.

For example, as well as trying to ensure that a departmental budget includes provisions for behavioral science, why not use behavioral science to improve the way this budget is created (e.g., are managers anchored to outdated spending assumptions)?

But we need to understand how this new way of thinking maps against the existing debate about how to set up a behavioral function in organizations. We propose doing so reveals six main scenarios.

Scenarios for Integrating Behavioral Science

Resilience

The goal would be to produce behaviorally-informed standard or business as usual processes, rather than the continued application of behavioral science explicitly. That approach is relevant to changes in the demand for behavioral science solutions as such in the future.

BIT’s 2018 Behavioral Government report proposes many practical changes to organizational processes to mitigate biases in government. But we need to understand the range of options for implementing such changes. How do we think about them alongside the desire to create a dedicated behavioral insights team, for example?

We think that the diagram above offers a useful way of mapping the options for building behavioral science into organizations.

The vertical axis represents whether behavioral science has been used to shape the organization’s own structures or processes, using a crude yes/no distinction to make the diagram manageable. We will bring this distinction to life with examples in the following sections, before defining it in more detail.

The horizontal axis deals with the extent and form of behavioral science knowledge and capacity in an organization. In the Baseline scenario, there is very little or no awareness of behavioral science concepts in the organization. Concentrated refers to the setup where there is a dedicated team or resource that applies behavioral science to organizational priorities. In the Diffused scenario, people or teams with competence in behavioral science are spread throughout the organization. Deciding between concentrated and diffused setups is generally seen as a central choice for organizations looking to build a behavioral science function.

Document Section 7 of 9

Here, levels of behavioral science awareness are still low, but its principles have been used to redesign processes to create better outcomes for staff or service users. For example, the pervasive optimism bias in organizations’ plans can be reduced by mandatory “pre-mortems”, where decision makers imagine the future failure of their project and then work back to identify why things went wrong. Group reinforcement (or “groupthink”) could be minimized by creating various futures and reviewing them anonymously before and after decisions are made. And the pressure to conform could be reduced by more diverse teams. To reduce the administrative burdens, organizations can choose services and processes that are easier to access.

This is often described as a “nudged organization” because no explicit behavioral science knowledge or capacity is created or needed. Like for nudging, it is the choice architecture (or choice infrastructure) that produces the outcomes, and there is no neutral choice in the way that an organization’s processes are set up. That means no behavioral team or unit is created; the change or goals may not even be framed in terms of behavioral science (as for “administrative burdens”).

For this reason, the best starting point is to understand how the existing setup is influencing behavior. Where is the choice architecture currently working well, through accident or design? How can existing processes be amended easily to draw on these practices? Who are the people who oversee the rules, incentives, metrics, and guidelines that influence people throughout the organization?

To give a concrete example, human resource leaders profoundly shape what organizations permit and reward. Yet, there has been relatively little focus on “behavioral HR”. Recent studies have shown that cognitive biases such as decay effects, framing effects, anchoring and halo effects can be created in practical decisions such as procurement and performance appraisal. They can also be countered: when considering the purchase of an email software, framing effects like saying 20% of users were dissatisfied significantly affected intentions (versus saying 80% were satisfied), but these effects were eliminated if both percentages were shown (in a random order).

The big outstanding question in this scenario is who introduces the nudges, since the organization has little internal capacity. Perhaps these could be one-off changes introduced from outside? Answering this question feels important, since the return on investment here could be large – and, for that reason, this model feels like a neglected opportunity that needs more attention.

Proactive Consultancy

In this situation, leaders may have set up a dedicated behavioral team, but perhaps not given much thought to supportive organizational changes. The result is that the team has to work in an enterprising way, going to look for opportunities and having to prove its worth.

This situation reflects the reality for many teams, who are “looking to develop networks, positions, and tactics that establish their authority and credibility among decision makers.” As a result, much of the discussion has focused on how best to set up these teams. The better contributions have recognized that this question is fundamentally political, rather than technocratic – how do the people leading such a resource build relationships and present their team as useful to their organizations?

The problem with this scenario is that teams may not be in a resilient position, since they lack ways to be grafted onto the standard processes of an organization. For example, leaders may neglect to support and resource evidence-gathering and experimentation. At the same time, they may have unrealistic expectations because they know only the highlights of previous behavioral science success stories.

Call for Experts

In the Call for Experts scenario, an organization has similarly concentrated behavioral expertise, but there are also prompts and resources that allow this expertise to be integrated more into ‘business as usual’. At its simplest, this might mean that standard procedures prompt staff to recognize the expertise and capacity held by any new behavioral team and what support needs to be in place for them to succeed (and be seen). Expertise and resources are also shared.

A new behavioral team, for example, may have invested in the development of advisory systems, and have built up a network of key stakeholders who they are called on. If working well, this setup would mean that processes stimulate demand for behavioral expertise that the central team can fulfill. That team may also have the institutional support to proactively monitor activities and respond quickly to specific crises.

One benefit to this kind of setup is that it allows teams to select the most promising collaborations, rather than taking whatever is on offer. For example, the team in Employment and Social Development Canada’s Innovation Lab claims that a ‘careful selection process is critical to the success of incorporating behavioral insights into an organization’, since it identifies partners who are open and willing to innovate, which makes it more likely that the subsequent project demonstrates the true value behavioral science can add.

Behavioral Entrepreneurs

In this situation, there is behavioral science capacity distributed throughout the organization, either through direct capacity building or recruitment. The distribution of expertise can work if there are effective support networks and efforts at coordination.

The problem with the behavioral entrepreneurs scenario is that organizational processes do not support these individual pockets of knowledge. Therefore those with expertise find it hard to apply ideas in practice, evaluate their effects, share findings, and build learning. For example, reducing “sludge” often requires coordination among a number of teams in the organization, which is a problem when teams work in silos and it is not acceptable in the organization’s culture to interfere with other teams’ affairs.

These are not just hypotheticals. A review of the Dutch policy landscape found that ‘most behavioral policy practices have not been deeply institutionalized,’ and their advancement ‘depends on the ambition of individual enthusiasts’ (or, as we’ve called them, ‘behavioral entrepreneurs’). While they can achieve some successes, their lack of institutional grounding can mean that they become jaded and start looking for other options instead.

Behaviorally-Enabled Organization

We see a behaviorally-enabled organization as one where the knowledge of behavioral science is diffused throughout the organization, which also has processes that reflect this knowledge and support its deployment. This is the most resilient setup, since staff will be applying behavioral science in a deliberate way as part of “business as usual”, rather than through special projects.

A behaviorally-enabled organization would bring together some of our previous proposals. It would embed the behavioral lens mentioned earlier into its core functions, including strategy and operations. For example, behavioral science could be integrated into cross-cutting frameworks like the WHO’s ‘Health in All Policies’ to prompt inter-departmental working. It would address the need to “see the system,” recognizing that sustainable outcomes are difficult to achieve through isolated changes. Such an organization would be self-reflective, and carefully explore the varying perspectives and experiences of its staff and service users.

While this setup has the greatest opportunity for scale and sustainability, it also requires the greatest investment. We conclude by talking about what kinds of investments are needed.

Choices and Priorities for the Build

Most discussions make it seem like the meaningful choice is between the different columns in our framework – how to organize your dedicated behavioral science resources. We argue that the more important move is from the top row to the bottom row: moving from projects to processes, from commissions to culture.

A useful way of thinking about this task is about building or upgrading the “choice infrastructure” of the organization, defined as the institutional conditions and mechanics of systems – the structures, processes, and capabilities – that directly underpin and shape the organization’s ability to help people make better choices.

In other words, we should place greater focus on the institutional conditions and processes that support building and using the best behavioral science knowledge and capacity. As the image below highlights, there are choices to be made about how this is done, based on ambitions and resources.

Working out how best to build the choice infrastructure in organizations should be a major priority for behavioral science. As with many systems, the best option may be to focus on creating the conditions for desired behaviors to emerge, rather than over specifying solutions. But already we can see some features will be crucial.

Dilip Soman and Katherine Yeung argue for the importance of reducing the costs of experimentation, including cheaper data collection, creating an experimental mindset, reducing institutional impatience, and building agility so that organizations can easily adapt to learning. Others have promoted the importance of sharing learning itself, pointing towards the crucial role of Singapore’s Civil Service College in ‘curating and facilitating an ecosystem of learning opportunities’ for behavioral science.

To this list, we want to add new and better ways of using behavioral science knowledge to analyze the behavioral effects of processes, rules, incentives, metrics, and guidelines. Such work has surged recently under the labels of ‘behavioral public administration’ and ‘behavioral operations management’, building on a longer tradition of organizational behavior research. We need to ensure that this agenda produces work that has practical value (and not just for the public sector), as in the proposal of “sludge audits” to reduce administrative burdens. Doing so will mean that the behavioral levers just proposed can be used by an organization’s members – and, ideally, by its leaders, who are in a position to achieve broader, systemic change.

The Behavioural Insights Team / A Manifesto for Applying Behavioral Science

03 SEE THE SYSTEM

Many big policy challenges emerge from complex adaptive systems, which present major challenges to the dominant way that behavioral science has been applied. However, we can adapt behavioral science to deal with complexity better, and use it to:

• Identify “leverage points” where a specific shift in behavior will produce wider system effects;
• Understand the collective implications of individuals using simple heuristics to navigate a system;
• Change the rules of that system to make it more likely that desired behaviors will emerge.

Of course, not every problem will involve a complex adaptive system. So behavioral scientists should first develop the skills to recognize the type of system that they are facing (“see the system”), and then choose their approach accordingly. Fulfilling the broader promise of behavioral science also requires us to expand the ways we tackle problems. The process of identifying targets, exploring drivers, developing solutions, and testing them is strong. The problem is that it contains several assumptions that do not hold when confronting some of the biggest challenges societies face. That’s because these challenges often consist of behaviors in complex adaptive systems (CAS).

The risk of talking about ‘complexity’ is that it may seem like just another way of saying problems are difficult (indeed some use the term as an excuse to do nothing). In fact, we mean applying a particular way of analyzing the world. Our proposal is that combining behavioral science with complexity thinking offers new, credible, practical ways of doing things differently.

Understanding Complex Adaptive Systems

A complex adaptive system is a dynamic network of many agents who each act according to individual strategies or routines and have many connections with each other. They are constantly both acting and reacting to what others are doing, while also adapting to the environment they find themselves in. Because actors are so interrelated, changes are not linear or straightforward: Small changes can cascade into big consequences; equally, major efforts can produce little apparent change. An important point is that coherent behavior can emerge from these interactions—the system as a whole can produce something more than the sum of its parts.

CAS consist of many different causes, actors, and goals. There are many examples of them in human societies, including cities, markets, criminal justice systems, and political movements. They often create what have been called ‘wicked problems,’ which are difficult or impossible to solve because of incomplete, contradictory, and changing requirements that are often difficult to recognize.

CONTESTED IDEAS OF SUCCESS

Throughout the pandemic there has been disagreement between individuals, organizations, and governments about whether overall policy goals should be to reduce deaths, avoid overwhelming the health system, or protect the economy. These goals are not mutually exclusive, but they do have varying implications.

NON-LINEARITY

The initial coronavirus variant appears to have been ‘over-dispersed’ – outbreaks were seeded by a handful of super-spreading events. It is estimated that around 80% of infections were caused by just 10% of individuals. These properties can explain why some nascent outbreaks fizzle and others take off: viral spread is a non-linear process.

Understanding these features could target policy responses effectively. They might indicate that stopping many repeated contacts within a small set of individuals has little effect on spread, in contrast to reducing random contacts at events and restaurants. In other words, the way the virus interacts with social systems means preventing many social contacts is unlikely to reduce viral spread in a linear way, whereas preventing a few super-spreading events may have an outsized impact.

UNINTENDED CONSEQUENCES

The wide-ranging, intensive actions taken to mitigate the spread of Covid-19 have had many indirect effects. These may have included increases in domestic violence; shortages of hydroxychloroquine in West Africa; reductions in carbon emissions; and changes in healthcare access with shifts to telehealth. Our point is not that these actions should not have been taken per se, but that solutions may change the nature of a problem or create many new ones.

CHALLENGES TO BEHAVIORAL SCIENCE

The issue is that there are fewer examples of behavioral insights applied to understand behavior in complex change processes. Why? Because the realities of complex adaptive systems challenge the main assumptions underlying the dominant behavioral science approach: tight focus on a target behavior, linear effects, and stability. We outline each of these before offering a way forward.

TIGHT FOCUS ON A TARGET BEHAVIOR

People can agree on a specific, measurable target behavior. Interventions that shift this behavior are successful – wider effects are minor and may not be considered, unless pre-specified.

Some behavioral science organizations focus on breaking problems down into their constituent parts to understand the desired behaviors. For some issues, there is value in doing this to identify a core target behavior to drive improvement. But you cannot understand a complex system by breaking it down into parts and mapping it – the way its connection function is key.

We cannot assume that changing a particular part of the system will have the desired overall outcome. For a start, there may be intense disagreement between parties about how a behavior contributes to an issue: some people may see pre-school provision as key to regenerating an urban area; others may view that as an unimportant contributor, compared with reducing crime.

LINEAR EFFECTS

The linear effect assumes participants in a direct and predictable way, as if they follow a linear theory of change.

In a CAS, actors adapt to the behavior of others, so there is often not a simple relationship between inputs and outputs. Actors in a system may adopt to “buffer” the effect of an attempted change and keep things operationally stable – i.e., making it seem that the intervention had no effects.

However, repeated efforts may weaken these stabilizing factors, and then a minor subsequent event produces a tipping point, where change happens suddenly and the system flips into a new state. An example might be repeated challenges that weaken the commitment of a country’s armed forces to democracy, which do not translate into action but which create the conditions for an apparently minor event to trigger a coup.

STABILITY

You can measure the pre-specified target behavior between point A and point B. The system will remain stable over that time, and people will not adapt in response to the intervention.

Since actors adapt to new conditions, and influence each other in doing so, the nature of the problem may be changed by the introduction of an apparent solution itself. A snapshot of behaviors at one point in time is not enough to claim victory. Perhaps the best example is regulation: market players experiment with and gradually adapt to a regulatory regime, working out how to evade its provisions, until a new policy is needed. Therefore success may actually lie in how well behavioral scientists adapt to the unanticipated effects their own actions produce.

DEVELOP BEHAVIORAL SCIENCE THAT CAN TACKLE PROBLEMS IN COMPLEX SYSTEMS

We have outlined the criticisms. Now, we want to offer hope and a way forward. There is an opportunity to develop behavioral science so it can tackle the aspects of complexity that are common to major policy issues. The starting point is to show how behavioral science can shed new light on well-known features of CAS: the fact that small changes can have big impacts, and the way that actors often use a simple set of rules to navigate a system.

Small Changes and Big Impacts

The idea that small changes can have a big impact is often used to describe how minor features of choice design or presentation can significantly affect subsequent behavior. This concept aligns with the standard approach of applying specific, targeted nudges to change precise behaviors.

However, Complex Adaptive Systems (CASs) demonstrate that small changes, when viewed differently, can influence higher-level features of a system, which in turn shape lower-level behaviors, leading to the emergence of new aspects and ongoing fluctuations.

Examples from the Covid-19 Pandemic

During the Covid-19 pandemic, individuals sought to achieve their goals within a broad set of rules while responding to changing events. These adaptive behaviors interacted with the virus’s adaptive abilities, resulting in the emergence of new variants. Some variants quickly became widespread, altering the nature of the pandemic, particularly through increased transmissibility or vaccine resistance. Policymakers had to develop new strategies to address the evolving situation.

Random Fluctuations and Societal Divides

Experiments have shown that initial random fluctuations can solidify into stark divides that shape societies. For instance, a recent study in the US revealed that partisan policy divisions might arise from these random fluctuations. The experiment created online “worlds” where self-identified Democrats or Republicans were asked to agree or disagree with various statements that did not reflect pre-existing partisan positions.

In eight of these worlds, participants could see whether mainly Democrats or Republicans agreed with a proposal. This visibility led to strong partisan alignment. Notably, the proposals that fell into the Democratic or Republican camp varied significantly between worlds, demonstrating initial fluctuations driven by chance and context before a sudden non-linear alignment occurred.

Emergence of Norms and Behaviors

More fundamentally, norms, rules, practices, and culture can emerge from aggregated social interactions, which then shape cognition and behavioral patterns. This understanding challenges the simplistic distinctions between upstream versus downstream or high-level versus low-level policies. Instead, we observe cross-scale behaviors, where behaviors embedded in specific contexts, influenced by the overall system functioning, can self-organize and emerge to shape the system itself.

New Possibilities in Behavioral Science

This perspective opens new possibilities for behavioral science to identify leverage points where behavior can be nudged to produce wider system effects. Interventions could be targeted at the stage of random fluctuations, where contextual features can determine which behaviors become locked in. For example, a recent study indicated that presenting individuals with a random selection of opinions, termed a “random dynamical nudge,” could prevent the formation of segregated echo chambers in online environments.