“But what‘s the behaviour you‘re trying to change?”

What applied behavioural science gets wrong under complexity

New problems, new approaches, new publicity: there is little doubt that behavioural science has grown up significantly over the last decade. From language focused on irrational behaviours, nudging and solution-focused frameworks towards an approach that places value on context and diagnostic frameworks such as COM-B; the field has changed.

This change is not just about ‘how’ we contribute to solving a problem but surely signifies a deeper transformation in the types of problems we address.

And here, a key ambition has been for behavioural science to move further ‘upstream’, to a place where applied behavioural science means to define and analyse problems before potential solutions become too constrained. So this allows us to avoid situations where we might prematurely design a text message without considering whether this intervention is the right approach or if there might be alternatives with greater impact and less risk.

As well as the shift to considering problems more ‘upstream’, another push has been to more complex problems. Specifically, we hear about behavioural science being applied to so-called ‘wicked problems’: those problems that are difficult to define, with a web of interconnected factors and no straightforward solutions. To us as applied behavioural scientists it seems obvious that when it comes to addressing some of these wicked problems, be they climate change, poverty, education reform or matters of global security, behavioural science should have a seat at the table. After all, almost all problems can be broken down into behaviours and to work with these problems means to understand the behaviours that constitute them.

Behavioural science often advertises being able to change behaviours. However, have our methods and approaches grown as fast as our ambitions? We believe that when it comes to moving upstream, applied behavioural science has not considered the type of system we are working with, and whether our current methods are suitable for these kinds of problems.

In this article we explore why that might be the case, provide a way of thinking first about the type of problem we are working with and finally explore what methodologies might be more suitable.

Frozen in time: behaviour, barriers, design

Explaining to a client or a colleague what behavioural scientists ‘actually do’ is a challenge many of us know (not that this is necessarily fair: the world seems to have accepted much more readily that an economist well…does ‘economicsy’ things, for better or worse). Together with the need for easy-to-digest, tangible and sellable information, we often revert back to simple ‘step by step’ approaches. Different consulting firms and organisations have their own acronyms and names to describe them, but fundamentally they often boil down to:

1.      Understand the behaviours that constitute the problem

2.      Define the behaviour you want to change

3.      Analyse barriers to achieving the change

4.      Design interventions that overcome barriers

Whilst the process described above likely originated as a communication tool more than an actual ‘step by step’ process to follow, it can be observed to be implemented as such. And for some challenges, this -linear, strategic – way of entry can indeed be making a big difference, and bring clarity and logic. For some problems however this approach is way too static and brings unrealistic expectations around how much clarity behavioural science can actually bring in this moment of time.

One way in which this process of applying behavioural science was advanced in recent years was through a greater appreciation and integration of systems thinking.  In systems thinking one often observes the current system with its complexities, projecting an ideal future system and considers how to transition from the current to the future state. The applied behavioural science process described above works in a similar way, in that we understand behaviours in their current context, imagine ‘better’ behaviours, analyse why they don‘t currently occur and then try to close the gap with behavioural design. Thus, with the integration of systems thinking approaches behavioural science has largely expanded the above 4 steps into taking account of multiple actors and multiple behaviours in a system. However, even with an expansion into systems thinking approaches of behavioural science there is little appreciation about the kind of system we are dealing with. And this is important: complex systems have characteristics that current, linear approaches are not necessarily suited for.

Complex systems and behavioural science: the issues

Behavioural science in complexity has, according to Matti Heino and colleagues, several features which pose challenges to more linear approaches.

The first is that there is a complex, interconnected system with multiple independent variables across different timescales that are interdependent and interacting. Important here is that how a system is organized, the relationship between component parts, can be more influential than individual components. Variables such as beliefs and social norms are linked, with their relationships changing across time. Deconstructing them into singular time snap-shots omits critical insights into the organization of the system in its entirety. The regression-dominant approaches that focus on individual components (e.g. barriers) might therefore fail to capture the dynamics within that system.

The second feature emphasizes the individuality of behaviour change: each person‘s behaviour change trajectory is unique and evolves over time. Psychological processes vary from person to person and are dynamic: the trends observed at the group level do not necessarily translate to even a single individual over time. The conclusions drawn about individuals based on group data can therefore be inaccurate, inferring how an individual will behave based on aggregate data can result in the ecological fallacy - leading to conclusions that might not be applicable to any given person, despite holding true at a population level.

The third aspect pertains to the non-linear nature of dynamics within complex systems, where the relationship between cause and effect is not straightforward. Small inputs can unexpectedly lead to substantial changes, while substantial efforts may yield little to no effect. Such systems may also experience sudden shifts after prolonged stability, complicating the forecast of behavioural changes.

In contrast to linear models, where outcomes are simply the aggregate of individual influences, non-linear dynamics involve disproportionate responses to inputs. Effects may remain dormant, only to amplify rapidly or shift abruptly once a threshold is reached.

The issues arising from these characteristics are manifold for behavioural science:

• First, there is the challenge of prediction and control in a system where change can be disproportionate and emergent.

• Second, there is the difficulty in designing interventions that are flexible enough to be effective for individuals with unique, evolving trajectories.

• Lastly, there is the inherent complexity in capturing and understanding the non-linear, dynamic interplay of variables over time.

Each of these issues necessitates a departure from traditional approaches and calls for innovation in how behavioural science conceptualizes, studies, and seeks to influence human behaviour within complex systems.

The question is: how can we get better at identifying where to use what kind of approaches?

The Cynefin framework

One way of stepping back and examining what kind of problem we are currently working with and what kind of action to take is by viewing the problem through the lens of the Cynefin framework (as shown below).

The Cynefin framework is a ‘sense-making’ framework – where sense-making is defined by the author David Snowden as “making sense of the world in order to act in it”. It distinguishes between 3 primary systems: ordered, complex, chaotic, which are defined by the type of constraints (or absence of constraints) in that system. Each type of system is described not just how it is constrained, but also describes how to best take action.

An ordered system has constraints, and in these systems future outcomes are predictable as long as constraints can be sustained. Ordered systems are split into “clear” and “complicated”.

In a clear system, constraints are evident and there is an obvious relationship between cause and effect. Here, the recommended approach is to Sense - Categorize - Respond. This means you assess the facts (sense), categorize them into the best fit from your known categories, and respond using a best practice or established procedure. Here there are known knowns, with  repeatable patterns and consistent results. An example might be increasing hand hygiene in a hospital setting. The problem (low compliance with hand hygiene) and the solution (implementation of hand hygiene stations with clear signage and staff education) are straightforward, supported by a wealth of behavioural research. This is a domain where solution-frameworks such as EAST can provide input to tried and tested approaches.

In complicated systems there are constraints and a relationship between cause and effect, but this relationship is not self-evident and may require analysis or some form of investigation to understand. Here, the approach shifts to Sense - Analyse - Respond. You need to sense the situation, analyse to understand the various interacting components, and then respond with a plan or solution that takes these complexities into account. These are the known unknowns, where expertise and specialized knowledge are required to discern patterns and provide solutions. Here, there are only good practice approaches where different experts may have different ways of engaging with the problem. This is a domain where behavioural scientists may use different frameworks (e.g. COM-B vs protection motivation theory). An example would be the problem of improving medication adherence among a diverse patient population. The issue is multifaceted, involving factors such as patient knowledge, belief systems, and forgetfulness. While research provides various strategies (e.g., reminder systems, educational programs), the challenge lies in analysing which combination of these strategies will be effective for a specific patient demographic.

In complex systems constraints exist but they are flexible and may change or adapt. The cause-and-effect relationship can only be perceived in retrospect, not in advance. The approach here is to Probe - Sense - Respond. This means conducting experiments or probes that are safe to fail, sensing how the system reacts, and then responding by scaling successful probes and dampening ones that fail. It is a space of unknown unknowns, requiring patterns of emergent practice.

For example, tackling the problem of drug-related deaths is a challenge that sits within the complex domain. The contributing factors — such as mental health, socio-economic status, network effects, and substance availability — can interact in unpredictable ways, making a straightforward cause-and-effect analysis difficult. A behavioural scientist might approach this by initiating community engagement programs, policy changes, or harm reduction strategies as probes. By carefully observing the outcomes of these interventions, insights can be gained about the patterns that lead to reduced drug deaths. Successful strategies are then amplified, while ineffective ones are modified or ceased. The complexity of the issue means that the interventions need to be adaptable and responsive to the unique and changing dynamics of the environment. The overarching goal is to iteratively move towards a systemic change that reduces drug deaths, understanding that the path to this goal is one of learning and adaptation, rather than linear progression.

In the complex domain, the standard behavioural science process begins to falter: at best, impacts are smaller than behavioural science markets itself. Most likely, current approaches lead to either no impacts or those that don‘t necessarily sustain over time. At worst, behavioural science interventions may lead to unintended behavioural consequences (without even noticing).

While understanding current behaviours is still possible, especially with more dynamic research methods, defining new behaviours is challenging because of the unpredictability and non-linearity of outcomes. Barriers to behaviour change are not fixed and can evolve, making them difficult to analyse definitively. Designing solutions in this domain involves creating multiple small-scale interventions and monitoring the system's response to them, adapting as necessary.

In chaotic systems the typical behavioural science process does not apply. In chaos, the priority is to act to stabilize the situation rather than to understand it deeply. New behaviours cannot be defined in a chaotic context until some order is restored. Barriers are not just unknown; they may not even exist in any stable form.

For instance, during a health crisis like a pandemic, normal behavioural patterns might be completely disrupted. The usual constraints (public health guidelines, government policies) may no longer be effective, and behaviour can become unpredictable. In this domain, swift, decisive action is needed to stabilize the situation and create new constraints (like emergency laws or health guidelines) to guide behaviour. Here, behavioural scientists may naturally be hesitant to bring concrete behaviours to the decision-making table, due to the learned approaches of analysing first – and may therefore be effectively excluding themselves from providing valuable input.

What can behavioural scientists do differently when working on complex problems?

Given the need for differentiated methods for different kinds of systems, and particular caution about existing approaches for complex and chaotic domains, applied behavioural scientists should be considering the appropriateness of the cornerstones of applied work, such as defining target behaviours. early on in strategic development, as shown in the table below.

In particular, when working in complex systems it is not helpful to adopt the typical process of define target behaviours, devise an intervention that removes barriers and measure this target behaviour again. Instead, good practice in these situations is surely to observe the kinds of behaviours that emerge following ‘safe to fail’ experiments (this highlights the importance of qualitative data specifically in understanding the impact of an intervention).

Note that here, safe to fail is an important component where behavioural scientists can bring particular value in conducting behavioural risk analyses (see e.g. the IN CASE framework) to reduce the chance of causing harm. This is to say that in complex systems we need to interact with them to learn about them – not just strive to influence and promise behavioural outcomes.

Here, positive outcomes can also be building the right kind of connections within this system (e.g. between people), an area of outcomes that are not yet frequently considered in the applied behavioural science space: the focus should be on building connections within the system, facilitating networks and relationships that can support positive change, rather than solely aiming to change end-behaviours.

In addition, rather than single-time quantitative data collection to understand barriers, in complex systems so-called N=1 studies, focusing on a single individual to derive deeper insights, can help us understand behaviours within their native contexts (see this article for more detail) and help avoid ecological fallacies.

Moreover, agent-based modelling, which may become more accessible with advancements in large language models, offers a way to simulate the interactions of individuals within a system to go beyond individual analyses of behaviours and to understand emergent patterns.

Finally, when in working with complex systems, there's a need for humility and a recognition of the system's inherent unpredictability. Behavioural scientists must manage expectations, communicating to clients and colleagues that the process of engagement with the system can yield valuable insights and incremental improvements, rather than the promise of ‘achieving behaviour change’.