No COM-B For Old Men

Before I go into what I think is an unappreciated challenge with the Behavior Change Wheel, it’s important to discuss what we mean when we talk about “theories”, and “models”, as opposed to simply “frameworks”. The Behavior Change Wheel bills itself as a framework for driving theory-based intervention development, and it provides a model of behavior to organize its content – yet each of these concepts is distinct, with its own implications for how scientific knowledge is represented and used to generate insights and design interventions.

In a theory or a model, we don’t just have some facet of human behavior that we’re looking to explain and a collection of high-level constructs that we think can explain it. Instead, what we have is a representation of the world that’s much more granular and, often, more tightly interwoven. In the case of a "theory", we have:

• A set of statements, or “propositions”, about the nature of the relationships between the theory’s constructs, which help us to say things like, “the more there is of A, the more there will be of B, but not vice-versa”
• An additional set of statements that precisely define the constructs that the theory considers important
• Yet other statements that put boundaries on what it is that the theory is looking to explain

These components of a theory can be quite complex. They may specify causal relationships in which the impact of A on B is dialed up or down by some third variable C, causal pathways in which A impacts B through C but there’s no direct pathway from A to B, and so on. They may also define constructs in ways that go well beyond the everyday understanding of the labels that are affixed to them. Either way, they become the key underpinnings of a “model”, which acts as a more formal (i.e., logical or schematic) way of representing a theory’s constructs and their relationships – sometimes quite crisply, sometimes more conceptually, spanning a broad range of findings and leaving room for pockets of uncertainty (and sometimes even standing in for theory itself) – but always with enough detail and connection to granular, theory-like statements as to make them meaningful bases for prediction and testing in concrete, well-defined contexts.

It can be easy to dismiss the process of theory or model building as if it were little more than specifying how many angels might dance on the head of a pin – but it is hardly so. On the contrary, it’s the care given to the specification of models and theories that makes it possible to conduct research that can speak to whether they are correct or not, as well as to whether the behavioral processes they posit are behind the behaviors observed in a specific context of interest. It’s also what makes it possible to make connections between the ideas a theory or model expresses and other pockets of scientific knowledge. And it’s these efforts combined that make it possible to take the learnings associated with a theory or model and extrapolate them to real world applications. 

What’s important to note about the Behavior Change Wheel is that, while it might be "grounded" in theories and models, it doesn’t look to embrace the nuances found in either. That’s because it’s a framework, and, as a framework, it deliberately avoids the details of a causal theory in favor of reducing scientific knowledge to a few manageable taxonomies and simple hard-coded rules for what goes with what. It may tip its hat to theory with the COM-B “model”, but the model’s constructs are painted with far broader strokes than is typically true of most models found in the literature so as to make them applicable across an exceptionally broad range of problems. It then substitutes a behavioral scientist’s way of using theoretical knowledge with guidebooks and tools, as well as with labels that are meant to provide a semblance of similarity between formal scientific ideas and everyday concepts – all with the laudable goal of making the material accessible, and spreading the power of theory-based insight and intervention-building to a broad base of practitioners.

All intuitive, all quite practical. And yet it is exactly these features of the Behavior Change Wheel that can cause its application to go off the rails. 

I don’t know how I could have completed my work on the earlier asset had I not activated a considerable amount of background knowledge about the processes implicated in the behavior the asset was meant to impact. This wasn’t just a matter of knowing something about the processes involved in goal setting, preference formation, self-regulation, learning, coping, and reasoning (to name just a few); it was also a matter of knowing something about how these processes work together and unfold over time, as well as how they connect to other processes that can work to influence them. It was the development of this rich, interconnected representation of behavioral drivers that allowed me to make educated guesses about how the user’s behavior might break down and the tactics that could be incorporated to make the asset more behaviorally meaningful. To get there, I had to step outside of the Behavior Change Wheel and reference back to several literatures so that I could pull in new material, evaluate it, put it all together, and render it to a level of detail where I could use it to support recommendations in which I could feel comfortable – none of which the Behavior Change Wheel alone could have assisted with. That kind of backfilling might be fine for someone with a behavioral science PhD, but it’s not so fine to push onto a team that must rely on the framework as a substitute for knowledge of the literature on which it’s based.

Alas, I’m not the only one to have come up against the Behavior Change Wheel’s unspoken gaps. As both published literature reviews and qualitative interviews have revealed, one of the biggest challenges practitioners experience when they use the Behavior Change Wheel is in figuring out how to interpret COM-B categories and theoretical domains, and how to prioritize relevant BCTs, using just the framework alone [7-10]. These struggles show up in the amount of confusion teams experience when they try to use the framework’s taxonomies to code the results of literature reviews or original research – a key step in figuring out what’s driving current behavior, and a critical input for identifying the BCTs, functions, or policies that should be considered for intervention design. Tellingly, teams have reported solving the problem through what might best be described as “judgment by committee”, in which they set aside the guide and revert to common sense to figure out how to fit findings to the framework’s constructs [10]. These solutions might very well work in a pinch, but they also introduce through the back door the very “it sounded good at the moment” thinking that the Behavior Change Wheel was designed to avoid.