COM-B Model: Diagnose Before You Design

The single most common way a COM-B analysis goes wrong is that the team already knows what it wants to build. Someone has budget for a teaching session, so the diagnosis discovers a knowledge deficit. COM-B is designed to prevent exactly that: it is a diagnostic step that must be capable of embarrassing you. If your analysis could never have concluded “the problem is not capability, so training will not help,” you have not run a diagnosis — you have written a justification.

A concrete illustration of the model doing its job: in a 2023 cross-sectional survey of 114 inpatient nurses at Duke University Hospital, COM-B analysis of unnecessary urine-culture ordering found that nurses’ confidence in ordering decisions had essentially no relationship with their knowledge of the indications (gamma = 0.02). Mean knowledge score was 9.93 out of 16. The binding constraints were opportunity and motivation — nurses described receiving pushback from clinicians when they questioned an order, and described ingrained practice to culture cloudy or foul-smelling urine despite knowing the guidance. The authors’ conclusion is the general lesson: interventions that focus solely on improving knowledge do not influence motivation.

What COM-B actually claims

COM-B was published by Susan Michie, Maartje van Stralen and Robert West in Implementation Science in 2011, as the hub of the Behaviour Change Wheel. It holds that three conditions must be present for any behaviour to occur, each with two subcomponents:

Capability — “the individual’s psychological and physical capacity to engage in the activity concerned.” Split into physical capability (skill, strength, dexterity) and psychological capability (the capacity for the necessary thought processes — comprehension, reasoning).
Opportunity — “all the factors that lie outside the individual that make the behaviour possible or prompt it.” Split into physical opportunity (time, resources, environmental affordances) and social opportunity (culture, norms, the language and concepts available to think with).
Motivation — “all those brain processes that energize and direct behaviour, not just goals and conscious decision-making.” Split into reflective motivation (evaluations, intentions, plans) and automatic motivation (emotions, impulses, habits arising from associative learning).

Two features are routinely dropped in secondary summaries and both matter in practice. First, the components interact: the original paper states that opportunity can influence motivation as can capability. A physical-opportunity fix can therefore move reflective motivation without any motivational content in the intervention at all. Second, the arrow runs back from behaviour: enacting a behaviour alters capability, motivation and opportunity. That is why COM-B is described as a system with feedback loops rather than a static checklist, and why an intervention that gets a behaviour started once can sometimes be withdrawn.

The practical consequence of the necessary-conditions logic is that COM-B is conjunctive, not additive. If physical opportunity is genuinely absent — the form does not exist, the slot is not in the rota — then raising motivation to its ceiling produces no behaviour. Look for the binding constraint before you look for the biggest effect size.

You cannot diagnose a behaviour you have not specified

“Improve hand hygiene” is not a behaviour and cannot be diagnosed. This is where most COM-B work in service-improvement settings fails, long before anyone reaches the wheel.

The Behaviour Change Wheel process begins with three steps that exist purely to force specificity: define the problem in behavioural terms, select the target behaviour, then specify the target behaviour. For the specification step, use the AACTT framework (Presseau et al., Implementation Science, 2019), which names five elements:

Action — the discrete, observable thing done
Actor — who does it (added to the older TACT framework precisely because healthcare goals depend on multiple people doing different things)
Context — the physical, emotional and social setting
Target — who or what the action is directed at
Time — when, how often, for how long

Run AACTT first. If two people on your team write different AACTT specifications for the same project, your COM-B interview data will not be interpretable, because respondents will have been answering about different behaviours. A well-formed spec looks like: the ward nurse (actor) sends a urine specimen for culture (action) for an inpatient with an indwelling catheter (target) on the neuroscience stepdown unit (context) only when documented symptoms meet the local criteria, at the point of order entry (time).

Where COM-B sits in the eight-step process

COM-B is step 4 of eight, grouped into three phases in the Behaviour Change Wheel guide:

Understand the behaviour — (1) define the problem in behavioural terms, (2) select the target behaviour, (3) specify the target behaviour, (4) identify what needs to change.
Identify intervention options — (5) identify intervention functions, (6) identify policy categories.
Identify content and implementation options — (7) identify behaviour change techniques, (8) identify mode of delivery.

Step 4 is the behavioural diagnosis. Everything downstream inherits its errors. This staged structure is what distinguishes COM-B work from a general planning model such as PRECEDE-PROCEED, which organises a whole programme lifecycle including evaluation, rather than drilling from a specified behaviour to replicable intervention content.

Use the TDF as the fine-grained layer

Six COM-B subcomponents are too coarse to build an interview topic guide from. The Theoretical Domains Framework is the standard finer layer, and it maps cleanly onto COM-B. Cane, O’Connor and Michie validated a refined 14-domain version in 2012 using card-sorting with 37 experts, revising the original 12-domain Michie et al. (2005) framework — splitting “motivation and goals” to yield separate Optimism and Reinforcement domains, and removing “nature of the behaviours.” The 14 domains are:

Knowledge
Skills
Social/professional role and identity
Beliefs about capabilities
Optimism
Beliefs about consequences
Reinforcement
Intentions
Goals
Memory, attention and decision processes
Environmental context and resources
Social influences
Emotions
Behavioural regulation

That paper reports three behaviour change experts reaching 100% agreement on allocating the 14 domains across the six COM-B subcomponents. The workable method is therefore: interview or survey against TDF domains, code to TDF, then roll the coded barriers up to COM-B. Going straight to COM-B tends to produce interviews that only ever surface knowledge and time.

From diagnosis to intervention functions

The middle ring of the wheel holds nine intervention functions, defined in the 2011 paper as:

Education — increasing knowledge or understanding
Persuasion — using communication to induce positive or negative feelings or stimulate action
Incentivisation — creating expectation of reward
Coercion — creating expectation of punishment or cost
Training — imparting skills
Restriction — using rules to reduce the opportunity to engage in the target behaviour
Environmental restructuring — changing the physical or social context
Modelling — providing an example for people to aspire to or imitate
Enablement — increasing means or reducing barriers to increase capability or opportunity

The outer ring holds seven policy categories through which those functions can be delivered: communication/marketing, guidelines, fiscal measures, regulation, legislation, environmental/social planning, and service provision. The wheel itself was built from a systematic review of 19 frameworks — an electronic database search returned 1,267 articles of which 8 met the inclusion criteria, and expert consultation added 17 articles of which 11 met the criteria. None of the 19 contained all the relevant intervention functions, which is the argument for the synthesis.

Table 2 of the 2011 paper gives a matrix linking each COM-B subcomponent to the subset of functions judged capable of changing it, and this is the step people skip. Read the matrix from the published source, not from a summary. Reproductions of that tick grid on the open web disagree with each other in individual cells, and a mis-transcribed cell is exactly how a project ends up justifying an education intervention against a physical-opportunity deficit. The structural logic is stable and worth internalising even so: education and training act on capability; restriction and environmental restructuring act on opportunity; persuasion, incentivisation and coercion act on motivation; and enablement and modelling are the broad-spectrum functions that reach across more than one component. No single function serves all six subcomponents — which is the entire reason the diagnosis has to come first.

Filtering with APEASE

The matrix will usually return more candidate functions than you can deliver. The BCW filter is APEASE: Acceptability, Practicability, Effectiveness and cost-effectiveness, Affordability, Side-effects/safety, and Equity. Apply it twice — once to narrow intervention functions, and again to narrow behaviour change techniques. In the CHARMS intervention for cardiac rehabilitation sexual counselling, for example, the team used APEASE to reduce nine functions to five (education, enablement, modelling, persuasion, training), then applied it again to cut a long list of candidate techniques down to twelve. Document the APEASE reasoning; it is the audit trail reviewers ask for.

Specifying content: BCTs and mechanisms of action

An intervention function is still not something you can deliver. Content is specified using the Behaviour Change Technique Taxonomy v1 (Michie et al., Annals of Behavioral Medicine, 2013): 93 defined techniques in 16 higher-order groupings, developed with 18 experts sorting techniques by similarity of active ingredients. Reliability was assessed by six researchers coding 85 intervention descriptions; of the 26 BCTs occurring at least five times, 23 had adjusted kappas of 0.60 or above.

To choose which BCTs plausibly move a given psychological mechanism, the Theory and Techniques Tool from the Human Behaviour Change Project offers an evidence grid built by triangulating a literature synthesis (2,636 author-asserted links across 277 intervention reports) with an expert consensus study. The published triangulation reports 92 confirmed links and 465 non-links. The non-links are arguably the more useful half: they tell you which technique–mechanism pairings the evidence base actively fails to support. Note the authors’ own caveat carefully — the grid identifies plausible connections and has not shown that they are effective. Treat entries as best bets to test, not as guarantees.

Measuring COM-B rather than asserting it

A recurring weakness in published COM-B work is that the diagnosis is qualitative and the evaluation never measures whether capability, opportunity or motivation actually moved. Keyworth and colleagues (British Journal of Health Psychology, 2020) developed a generic six-item self-evaluation COM questionnaire to fill that gap. Reported properties: intraclass correlations of .554–.833; a three-factor confirmatory model fitting well (chi-square[6] = 7.34, p = .29, RMSEA = .02, CFI = .99, TLI = .99); and regression models accounting for 21–47% of variance in behaviour.

Two cautions before you adopt it. Missing data rose sharply from 5.9–7.7% per item at baseline to 18.1–32.5% at follow-up, so plan for attrition on the repeat measure. And ceiling effects reached 22.9% on some items — a six-item generic measure will struggle to detect improvement in a group already scoring near the top. If you need sensitivity for a specific behaviour, expect to supplement it with behaviour-specific items rather than rely on the generic instrument alone. Pair COM measures with distinct implementation outcomes such as adoption, fidelity and penetration; COM-B tells you about determinants, not about whether the intervention was delivered as designed.

Known weaknesses, stated honestly

It is a model of behaviour, not a full theory. West and Michie’s own 2020 introduction positions COM-B as needing to be used alongside more specific models relevant to the behaviour in question — COM-B tells you which class of determinant is binding, not the mechanism by which it operates or how strongly.

The judgement points are undocumented. The EQUITy programme team (2021) put this plainly: the BCW is a method, not a magic bullet, and there is a lack of worked examples showing how to identify and negotiate the numerous points at which judgement is needed. Their specific complaints are worth pre-empting in your own protocol — how much of each evidence type counts, how contradictory evidence is reconciled, and how competing stakeholder views are weighted. Their solution was to run consensus separately by stakeholder group (patients, practitioners, key informants) rather than forcing one group, and to adopt an explicit decision rule requiring endorsement by two or more groups. They also note the BCW has been criticised for a lack of patient involvement, which they treat as a weakness in how it is implemented rather than in the framework itself.

Systematisation has a cost. Jane Ogden’s 2016 critique in Health Psychology Review argues that reducing interventions to catalogued techniques neglects genuine and valuable variability in persons, interactions and theories, and risks turning practitioners into technicians. You do not have to accept the conclusion to take the operational point: a fully specified BCT list does not capture the relational and delivery variance that often drives whether an intervention lands.

How COM-B relates to neighbouring frameworks

COM-B answers what needs to change in the person and their context. It does not replace the other pieces of an implementation science design. Implementation Mapping covers a broader five-task process including identifying adopters and implementers and producing implementation protocols, and can absorb a COM-B diagnosis as its determinants step. PRECEDE-PROCEED spans a whole programme including epidemiological and administrative assessment. The TDF is not an alternative to COM-B but its higher-resolution companion. Choosing among them is mostly a question of scope: use COM-B when you have one specified behaviour and one group of actors, and reach for a programme-level model when you have several behaviours performed by several actors across a service.

Frequently asked questions

Is COM-B a theory or a framework?

It is best described as a model of behaviour rather than a predictive theory. It specifies three necessary conditions and the fact that they interact, but it does not specify functional forms, weights, or how the components combine to produce a given behaviour. West and Michie explicitly say it needs to be used alongside more specific models for the behaviour in question. In practice that means COM-B is excellent for structuring a diagnosis and poor as a standalone basis for prediction.

What is the difference between COM-B and the Behaviour Change Wheel?

COM-B is the hub; the Behaviour Change Wheel is the whole thing. The wheel adds a middle ring of nine intervention functions and an outer ring of seven policy categories, plus the matrix linking COM-B subcomponents to functions. You can run a COM-B diagnosis on its own, but the wheel is what turns that diagnosis into intervention options.

Do I need the TDF if I am already using COM-B?

Not formally, but you will usually want it. Six subcomponents give you too little structure for an interview guide or a coding frame, and teams that code directly to COM-B tend to over-detect knowledge and time barriers. The 14 TDF domains give interviewers something specific to probe, and the mapping back to COM-B is well established. Published applications go both ways — the CHARMS team, for instance, used COM-B directly without the TDF, and reported it as a deliberate simplification.

How do I know which COM-B component is the real barrier?

Collect data rather than reasoning from the armchair, and design the collection so it can falsify your assumption. The urine-culture study is the model: it asked separately about knowledge and about confidence, and the near-zero association between them was what ruled out a pure capability explanation. Cross-tabulating a capability measure against a motivation measure, and testing whether they track each other, is a cheap way to find out whether the deficit you assumed is the deficit you have.

Can COM-B be used outside health?

Yes. The wheel was developed in a health context and most published applications are in health, but it has been applied to workplace, environmental, education and organisational behaviour. Nothing in the definitions of capability, opportunity or motivation is health-specific. What does transfer poorly is the accumulated health-sector evidence about which intervention functions tend to work, so treat effectiveness estimates borrowed from health applications with caution.

How many behaviour change techniques should an intervention include?

There is no evidence-based optimum, and the published examples vary widely; the CHARMS intervention settled on twelve after two rounds of APEASE filtering. More techniques means more to deliver with fidelity and more difficulty attributing effects. The defensible answer is to select the smallest set that covers every COM-B component your diagnosis found to be deficient, document the APEASE reasoning for each inclusion and exclusion, and report the BCTTv1 labels so others can replicate what you did.