A Spin of the Behavior Change Wheel

In this post, I will take a look at the Behavior Change Wheel (BCW), a newer framework for understanding health behavior. The model has been around for a decade, but it became much better-known after the book Engaged featured it in 2020. The BCW model is sometimes called by the name of one of its components, like "the COM-B model." Technically COM-B is just part of the full model diagram, shown above (the green inner circle), so I'm going to refer to the full framework as "the BCW" in this post. As you will see, it might be possible to utilize or accept just one of the BCW's components without necessarily buying into the whole thing. I'm also going to call the BCW a "framework" or "model" instead of a theory. That part comes directly from its creators, who describe the BCW as "a synthesis of 19 frameworks of behavior change found in the research literature" (Mitchie, Atkins, & West, p. 11). The 19 frameworks are also models rather than theories: e.g., the sociological culture capital framework, the Cochrane collaboration's EPOC taxonomy of health systems interventions, and the MINDSPACE framework for policy-makers in the United Kingdom. The BCW's synthesis of those models is reflected in the red circle shown in the diagram above. There's also a white outer circle focused on policy-level interventions, which is a nice addition for lawmakers or administrators, and which isn't often found in a psychological model of behavior. Unfortunately, as I will show below, these outer layers of the BCW don't map clearly onto one another or onto COM-B. 

What the BCW is not, I will argue, is a theory. That's because a theory of health behavior change has to specify causal relationships and consistent connections between its parts. COM-B is a theory for explaining behavior, although I will argue below that it's a weak one. The BCW does have linkages to more traditional health behavior theories, which connect via Michie's taxonomy of 19 possible influences on behavior called the Theoretical Domains Framework (TDF: yellow wheel in the diagram above). TDF is an atheoretical or integrative list of behavior-change techniques often used in psychotherapy or other interventions focused on an individual person -- I talk more about it in this blog post from 2021. Michie's team has also developed a much longer list of over 100 micro-strategies that might be active ingredients in behavior change interventions, the Behavior Change Taxonomy (BCT). That component is not shown in the diagram, but is also potentially part of the BCW approach. The TDF and BCT do have some level of alignment with one another, although again the relationships between constructs aren't one-to-one. But wait, there's more! The BCW manual also includes some practical criteria for weeding through the list of options in a mnemonic called APEASE -- affordability, practicality, efficacy, acceptability, side effects, and equity -- which is a handy tool but seems like it could be considered entirely separately from the rest of the BCW framework. The BCTs and APEASE tool, although described in the BCW manual, aren't shown in the circle diagram above.

So, in short, there's a lot here. I will take a look at individual components of the BCW below.

The COM-B Model (green inner circle)

The heart of COM-B is a simple model, suggesting that 3 variables predict behavior: capability which is someone's ability to do the behavior, motivation which is their interest in or willingness to do the behavior, and opportunity which describes the environment or circumstances in which they do or don't have the chance to perform the behavior. Notice in the diagram below that almost every variable links to every other one -- interestingly, there's no direct link between capability and opportunity, suggesting that those two things operate completely independent of one another, and each of them has a single-sided arrow pointing to motivation, which means that they can affect motivation but motivation can't affect them. The three predictors have apparently equal weights in producing behavior.

I would argue that the missing arrows are a problem here. In particular, motivation can affect opportunity (people seek out favorable situations), and motivation can affect capability (all of adult learning theory is built on the idea that people learn when they want to learn). So at the very least, the arrows around motivation need to be double-sided in order to accurately reflect the world as we know it. Capability and opportunity are probably related as well: For example, people with greater skill playing the violin will get more invitations to perform in public, and people who have opportunities to practice basketball often (a court behind their apartment building) will get better at it faster than people who didn't have that opportunity. So the three factors probably all inter-relate, which makes this an "everything leads to everything" type of model, and therefore not all that useful for explaining the causes of behavior. 

Where did the three factors of the COM-B model even come from? A 2011 article suggests that they originated in criminal law: "in order to prove that someone is guilty of a crime one has to show three things: means or capability, opportunity, and motive" (Michie et al., Implementation Science, 2011, 6:4). At its heart, the model is just a Venn diagram saying that capability, motivation, opportunity, and behavior all have something to do with one another. Unfortunately, we still aren't quite sure what.

The intervention functions (red circle)

Next I will jump to the red circle, because the yellow circle isn't shown in some versions of the BCW diagram -- in the manual, there are actually 2 different versions of the figure, one with the red-circle items (Figure 1) and a separate adaptation with the yellow-circle items (Figure 1.7). The "intervention functions" in the red circle are the most commonly referenced when the BCW is cited. The BCW manual says that the word "functions" is used because "the same intervention may have more than one function so interventions cannot be classified in this way, only characterized" (p. 19). In other words, there is a one-to-many relationship between interventions to address one of the COM-B factors (green circle) and the specific intervention functions (red circle) that might be included in those interventions. the manual has a matrix showing these many-to-many relationships, to which I have added some other possible intersections. Even if you ignore my scribbles and focus just on the shaded boxes, you can see that a given intervention function could be used to address many different COM-B components, and a given COM-B component could be tackled using many different intervention functions:

Many-to-many relationships are a problem in a theory of health behavior, because they don't allow you to make clear predictions about what's going to happen in any individual case. For example, if I design a smoking-cessation intervention using a peer support group with a weekly speaker, is the active ingredient education? persuasion? modeling? maybe incentivization if there's particularly good pizza at the group meeting each week? coercion based on the embarrassment a participant might suffer if they have to tell fellow group members they had a relapse? A single intervention can include multiple "functions," something that the authors of the BCW readily acknowledge. In an infectious-disease example, the authors of the BCW manual do in fact note that "all intervention functions could potentially bring about the desired change" (p. 194).

One way to think about this is that the outer wheels of the BCW are not fixed in place, so you can't read from inside to outside and get a consistent response. Instead, the wheels spin around each other. Imagine a pin through the center so that the wheels can rotate independently, and perhaps imagine the outer wheels having little plastic teeth underneath that permit some match-ups and prohibit others, although a lot of match-ups are possible and relatively few are prohibited. Also, it seems like the plastic teeth aren't that strong, so you can pretty much force a match if you want one. 

You might, of course, take the position that none of this detail matters so long as the intervention works, e.g. if people who come to my group actually do stop smoking. And that's fine from a practical standpoint. But it's not fine from a theory standpoint, because a study testing my smoking-cessation group doesn't tell me anything about the general principle of behavior change that I might need to use in my next study, when I try to design an intervention to help people improve their diet. In that scenario, using the incentivization strategy "feed them really good pizza" is likely to be a drawback rather than contributing to the new group's success. 

The policy categories (white outer circle)

I have the same basic critique of the policy categories in the outer circle: It's a great list of possibilities, but the alignment between these and the intervention functions is once again many-to-many. As shown by the number of shaded cells in the following grid, there is even more overlap between the red circle and the white one in the BCW diagram:

Because the intervention functions were themselves in a many-to-many relationship with the three components of the COM-B model, we now essentially have 3 wheels that can spin around one another in almost any order. 

A final challenge is that the authors of the BCW manual encourage intervention designers to consider more than one intervention function or policy category when designing a behavior-change intervention. In one example (p. 126, box 2.1) this actually leads to every single one of the 9 intervention functions being considered as part of a possible intervention. The functions then go through another step of narrowing using the APEASE tool, to identify the ones that will be easiest to implement and/or most likely to succeed -- again, APEASE seems like a great tool for that purpose, but it's outside the BCW framework itself. My question, then, is why not just start with the full list of 9 possible intervention functions (or the full list of 7 policy levers) for every single program you are trying to design? A lengthy process of "behavioral diagnosis" using the COM-B component of the model doesn't add a lot because of the many-to-many relationships between COM-B and the other model elements. And using the rule "always consider all potential strategies" gives you more possibilities to play around with, increasing your chance of a successful solution. Essentially the BCW could come down to a list of possible behavior-change strategies, which can be evaluated for practical utility using the APEASE tool.

The Theoretical Domains Framework (yellow circle) and the Behavior Change Taxonomy

Let's return to the yellow circle on the diagram above (again, not included in the main diagram that's published in the BCW manual), which is Michie's Theoretical Domains Framework (TDF). As I have written previously, the TDF is an integrative list that classifies constructs from various health behavior theories into just 14 domains. Each of the TDF domains -- which are things like "emotion," "knowledge," and "intentions" -- represents a possible influence on people's behavior. It fits reasonably well with the categories that people use in everyday conversations about psychology or reasons for action, and it was developed in 2005 through a systematic process of consensus development that involved experts in both health psychology and systems research. The TDF was therefore the first element of the entire BCW framework that was developed. It also has an advantage in that the TDF domains do link one-to-one with the COM-B elements, not many-to-many, as shown in the exact alignment of the yellow wheel with the green one in the diagram at the top of the page. The TDF domains are essentially a different or expanded way of thinking about the various things that can influence people's behavior -- underlying causes, not interventions.

Michie's team used a similar systematic approach to develop a second list, called the Behavior Change Taxonomy (BCT), in 2013. This is a much larger list -- 113 different items, at one count -- of methods to change people's behavior, instead of the shorter list of behavior's underlying causes in the TDF. Here's the whole list, as published in the BCW manual. (If the list seems overwhelming, don't worry: there's an app for that).

Because of the number of BCT techniques, these aren't shown in the BCW diagram, but they are incorporated in the manual so I think it's still fair to consider the BCT list as part of the total BCW model. The corresponding part of the diagram is the red wheel, where the >100 BCT techniques are replaced by just 9 intervention functions -- both elements describe "ways to help people change their behavior," just as both the COM-B model and the TDF describe "reasons why people behave the way they do." Unfortunately, the BCT does not map nicely onto the 9 intervention functions -- a table in the BCW manual (pp. 151-155) suggests that some BCTs are "more frequently used" with certain intervention functions, but also says that essentially any BCT might be used to address any function. So the BCT is essentially a replacement for the intervention functions -- a much more complicated one to be sure, although also a much better-validated one. 

Drawbacks of the BCW

I think you can already see my view on the major drawbacks of the BCW model: It's not a health behavior theory in the sense of one-to-one linkages and predictive relationships. Its constructs don't line up clearly at the various levels in a way that leads to clear recommendations -- essentially you can identify a problem using the COM-B elements or the TDF, and then you can select a solution using the BCW's 9 intervention functions or the BCT's list of over 100 techniques. But the problem and the solution don't necessarily have a lot to do with one another. If you are empirically focused, you could just start with solutions, reminiscent of Lester Luborsky's dodo bird verdict in psychotherapy: "regardless of your problem, seek any available form of treatment," because they are all equally likely to help you. The BCW guide suggests that this model can be used to answer questions such as "what will it take to bring about the desired behavior change?" and "what types of intervention are most likely to bring about the desired change?" (p. 28), but the BCW alone can't answer those questions due to its many-to-many relationships. That's the role of health behavior theory. The BCW is a tool that can actually be integrated with many different theories, for instance with Social Cognitive Theory in Box 3.3 of the manual (pp. 166-167), but it can't take the place of theory on its own.

Benefits of the BCW

The BCW manual suggests three major advantages of using this tool (p. 20): "First, it includes automatic processes such as habit. Secondly, it explicitly includes factors at a systems level. Thirdly, the specificity of components within the COM-B model [or the BCW], and hypothesized relationships between them, allows a precise description of the relationship between individual determinants and adherence, making it easier to identify appropriate interventions." I take issue with point #3, because I think the precision isn't sufficient to actually accomplish that task. But the other two points are valid -- the BCW's lists of items could lead an intervention designer to consider systems factors that influence behavior, and to consider strategies that address habits instead of just thoughts and beliefs. In fact, the focus on habit is one of the things that I like about the BCW model: Here's a sample graphic that shows behavior transitioning over time from "reflective motivation" (i.e., TMT's Narrative Mind) to "automatic motivation" (TMT's Intuitive Mind), which is clearly a beneficial goal. The figure caption also mentions the intention-behavior gap, which is a major focus of TMT:

So, linkage to multiple theories is another potential advantage of the BCW. Its large list of potential causes of behavior (the TDF) and its even larger list of potential intervention strategies (the BCT) can be grafted on to whatever theoretical model a designer most prefers, and might generate additional insights. I think this flexibility is one reason the BCW has caught on so strongly, but it also exposes the fact that its components are not tightly linked on their own. 

A final potential advantage of the BCW is that it might lead people simply to consider psychological influences on behavior in the first place. In an inpatient hospital example, the BCW manual says that "specifying [their intervention] using theory and BCT methodology provided a more comprehensive understanding of its components, aims, and functions" (p. 175) -- in other words, the BCW adds depth because it led intervention designers to consider some things they otherwise might not have. And in an example about preventing infectious disease, the manual's authors note that "this was originally conceived of as an educational intervention. By taking the steps described [by the BCW], we can see that in addition to education, there are many other types of intervention that can be used to bring about the desired change" (p. 195). This benefit alone makes the BCW worthwhile: It's shocking how many proposed behavior-change interventions start and end with education, despite the scientific field's knowledge for more than 50 years that knowledge alone is insufficient for health behavior change. But again, just using the lists of possible intervention strategies without the wheel diagram might give you exactly the same benefit.