Katelyn Stenger, Doctoral Fellow, Convergent Behavioral Science Initiative, University of Virginia
Climate Action. Equality. Peace. Today’s most ambitious goals involve recognizing or understanding complex systems. As many recognize,behavioral science also plays a significant role in achieving these goals. Yet, few seem to appreciate how behavioral science and complex systems intersect. To offer a rough sketch, this article will explore some of these intersections using examples from research and practice, and make the case for much more proactive integration.
Behavioral scientists have identified interventions that are effective in behavioral change. Defaults, like opt-out systems for organ donation, for instance, has proven successful in boosting donor registration, and thereby potential organ transplants, as implemented within the UK. Framing infrastructure decisions to better consider the future can extend the life of infrastructure designs (forthcoming). And, putting savings in more salient daily terms increases savings for those with low to middle incomes.
Nevertheless, these are only the beginnings of behavioral design, using small sets of behavioral insights. Behavioral interventions are often founded and designed for stable conditions and specific demographics. These don’t always translate to other contexts. In practice, of course, many have recognized this and adapted their approach to meet needs such as Evidn’s Behavioral Systems Analysis, and Lemonade’s varied insurance offerings.
Stable conditions can happen, yet they assume a lot. Researchers in both complex systems science and strategic design argue that it is essential to consider future conditions when designing in the present. Those practicing or researching behavioral science could benefit from understanding complex systems. Otherwise, they risk falling prey to their own biases of understanding and valorizing causality.
Complex systems, or more specifically complex adaptive systems, are systems without central control that follow simple rules and give rise to complex behavior. Complexity is the study of adaptive matter; whether it be people, plants, or proteins. In complex systems, one thing is clear — circumstances will change; whether it’s a natural cycle or a full-on disruption.
Systems thinking enables a thinker to recognize components as connected and their interactions to form a functioning whole or system. Further, systems thinking helps recognize social, economic, and natural worlds as parts of a system.
Systems thinking, popularized by Donella Meadow’s Thinking in Systems, is associated with holistic thinking. Ballew & colleagues found evidence that systems thinking reveals a vital pathway to belief in global warming. While systems thinking helps people understand complex systems, we need an additional boost to design within complex systems.
Of course, complex systems are inherently challenging to understand because of their interactions – how components of a system engage with each other. This can be friendships in school children or predation in ecosystems. While many understand scales of a complex system (like local, regional, and national for social networks), the more difficult concepts of complex systems need more support, like the system’s decentralized organization and the unpredictable nature of effects.
In behavioral public policy, complexity presents an opportunity to consider design solutions beyond a narrow set of individual behavioral interventions. As Nathalie Spencer discusses, behavioral interventions are just one group of tools in the toolbox that behavioral designers can apply toward broader organization, system, and social change. The National Resident Matching Program, for instance, restructured its matching system to increase preference alignment between medical students and hospitals by using an algorithm and is regularly revised to achieve equitable outcomes.
Behavioral design has emerged as a human activity based on intention, which shapes decision-making environments. The difference between behavioral science and behavioral design is nuanced and still emerging. Scientists often look at the world and ask, “Why?”, seeking to understand the world as it is, while designers look at the world and often ask, “Why not?”. seeking to imagine the world as it could be. These questions often influence the sources of evidence considered and the scope of problems undertaken. One can be in either of these modes at different stages throughout a project, and both offer crucial perspectives to behavioral public policy.
While caution is advised when applying behavioral science to public policy, there is wide acceptance of basic approaches that are effective, for example priming, framing, and defaults. Behavioral designers use such interventions and other tools, including systems design; finding creative ways to improve situations or solve problems. Ruth Schmidt and I have argued for a more systemic approach to behavioral design, that gives greater consideration to ‘the big picture’ (using tools like systems thinking) and ‘the long game’ (using future thinking).
Predicting group behavior intimately intertwines with complex system science — and, more specifically, network science. Social networks are how people interact with each other, whether friendship among school children or citation practices among academics. Considering how networks form —and evolve— provides a complexity lens to understand human behavior.
Predicting behavior from this environment is a challenging but rewarding inquiry. For example, Galesic & colleagues (open-source) reported in a Nature Human Behavior article that collecting a person’s response to “Who are your friends (or family) going to vote for?” and analyzing these results in a social network improved predictions of election voting. Next time a pollster calls, you might be asked an uncomfortable question: “So, who’s your mother voting for?”
Networks offer one lens to understand and predict behavior. You may have heard that you are six handshakes away from the President? This concept stemmed from early research on social networks known as the small-world experiment. Rafaela Bastos at NudgeRio designed a messaging campaign asking receivers of a message to share the message with friends — an easy prompt to leverage networks — about flattening the curve for COVID-19. The message reached hundreds of thousands.
Behavioral scientists challenge the assumptions of homogenous populations. Dilip Soman and Tanjim Hossain advocate that successfully scaled solutions need not be homogenous and need to be contextually appropriate. Even more, designing solutions that make sense in their contexts uses some insights from complex systems: decentralization. As researchers Anirudh Tagat and Hansika Kapoor at Monk Prayogshala point out, decentralized nudging will need to occur more often in highly diverse regions. Designs involving communication in India, for instance, need to accommodate more than 22 official languages.
Behavioral scientists also use big data to overcome challenges associated with heterogeneous populations. Stuart Mills and colleagues offered an insight that more heterogeneous data can improve and craft more personalized nudges. Contexts with an abundance of personalized data, in smart buildings for example, offer a fertile area for personalized behavioral design.
However, using big data comes with a caveat on ethics. This approach neglects those who cannot or will not generate such data, and can embed critical assumptions about who is included for personalized nudges. If behavioral designers select broad populations — relying on averages — their designs likely neglect individuals without data or marginalized by existing data practices. Consideration of who might be excluded or marginalized within big data inquiries is vital before launching personalized behavioral interventions.
Even more, tools to comprehend big data, such as machine learning, require healthy skepticism and critical thinking on findings. Rather than celebrate the 94percent model accuracy, behavioral scientists need to investigate the 6percent inaccuracy and ask: What features does the model classify? Is there something systematic about the model inaccuracy? What examples does the model struggle with? A skeptical approach learns from well-documented machine learning limitations, in order not to replicate them in behavioral science.
Inviting more disciplines into the fold allows behavioral scientists and designers to understand problems in a more nuanced way. But it comes at a high cost. Behavioral public policy has been characterized by bounded interdisciplinarity, favoring some types of evidence and contexts, thereby limiting its complex problem-solving abilities. Costs of interdisciplinary science are not unique to behavioral public policy, and the switching costs of science are high. On the other hand, an interdisciplinary (and even transdisciplinary) approach increases understanding, and as Baruch Fischoff states:
“the key to communicating scientific research is simple: collaborate”
Moreover, creatively engaging citizens and stakeholders when finding solutions to complex problems, sometimes called co-design, increases the project’s complexity and decreases designer control, but better realizes higher quality solutions. This added complexity may also improve the robustness of behavioral public policy. It needs to be realistically planned for at the onset of projects.
Behavioral science is already confronting major titans: Replicability, scalability, generalizability, and heterogeneity. Most certainly not an itty-bitty feat. Complex systems science will help take on these titans, as we’ve learned with heterogeneity and scalability. As the behavioral revolution in public policy brings increasingly complex challenges into its sights, where behavioral scientists struggle to predict or control behavior change, reframing situations within a lens of complexity can reveal new pathways toward efficacy.