A town wants more residents to use a free evening walking route. It can put clear signs at the entrance, improve lighting and let people choose a distance. It can also send repeated messages implying that anyone who stays home is irresponsible. Both approaches seek a change in behavior. Only the first begins by making the desired action genuinely possible and easier to choose.
Behavior change architecture is the arrangement of opportunities, information, cues and feedback around an action. It is broader than a reminder on a screen. The opening hours of a service, the design of a form, the safety of a path and the expectations of a group all affect what people can do. Architecture that preserves agency helps a person pursue a goal they understand, while allowing them to question, adapt or decline the proposed route.
Begin with the behavior and its conditions
“Become healthier” is too broad to design around. “Walk along the lit route twice this week if it suits your health and schedule” identifies an action, place and time. Even that description needs an account of who can access the route and whether walking is the right goal for them. A program should define the action it seeks without pretending that the same action is appropriate for everyone.
Michie, van Stralen and West’s behaviour change wheel begins with a simple diagnostic model: behavior depends on capability, opportunity and motivation, which also interact with behavior over time. Capability includes relevant knowledge and skills; opportunity includes features of the physical and social environment; motivation includes reflective intentions and more automatic processes. The model helps teams ask where a barrier may lie. It does not tell them that a lack of action is a defect in the person.
Imagine that residents understand walking’s possible benefits and want a route, but the crossing is dangerous. A motivational campaign cannot replace a safe crossing. If the route is safe but poorly signposted, information may help. If someone needs an accessible surface or seating, a map of steps is not enough. The intervention should follow the diagnosed constraint, and the diagnosis should include the people who would use the service.
The behaviour change wheel lists several intervention functions, such as education, training, environmental restructuring and enablement. These are categories for thinking, not evidence that every instance of a category works. A sign can be clear or confusing; a training session can build skill or waste time. A responsible team specifies the mechanism it expects and evaluates the actual intervention, including any burden shifted onto participants.
Habits are cue-linked, not proof of weakness
Repeated actions in a stable context can become more automatic. Wood and Neal’s review describes habits as learned links between contexts and responses that can operate alongside goals. A person may intend to make one choice yet repeat a familiar action when the usual cue appears. That is not evidence that they lack values or intelligence. It is a clue that changing an environment or routine may matter more than repeating an instruction.
Lally and colleagues followed people trying to form an everyday health-related behavior in a chosen context. Their modeling showed that automaticity tended to rise with repetition and level off, with considerable variation across participants and behaviors. The frequently repeated “it takes 21 days” slogan is not a conclusion of this study. Its design was observational, based on self-reports in a selected sample, and its estimate does not become a universal deadline for a new habit.
A practical design might help someone attach a chosen action to a reliable cue: after arriving home, place walking shoes by the door; after lunch, take a brief accessible loop. But the cue should serve the person’s own goal. An app that manufactures anxiety whenever a streak is at risk may create repeated use without helping the person form an action they value. Automaticity is useful when it frees attention for a chosen purpose; it is troubling when it keeps a person acting against a revised intention.
Context changes can disrupt both helpful and unhelpful routines. A new shift pattern may make a former plan impossible. A person may miss days because of illness, care duties or weather. The architecture should allow the plan to be adjusted without treating every interruption as failure. A good intervention learns whether the original goal still fits before trying to restore a metric.
Make a plan for a goal the person owns
An intention to act often leaves the moment of action unspecified. Gollwitzer and Sheeran’s meta-analysis examined implementation intentions, often framed as an “if this situation occurs, then I will do that action” plan, across a range of goal pursuits. The evidence supports the value of specific plans for many studied outcomes. It does not mean that a plan can supply missing resources, make a dangerous route safe or substitute for a goal the person does not endorse.
For the walking example, a resident might decide: “If it is still light after dinner on Tuesday, I will take the short route; if it is dark, I will choose another activity.” The second clause matters. It acknowledges context and preserves flexibility. A rigid plan that ignores changing conditions can become an instrument of pressure rather than support.
A service can invite people to make their own plans and show a few concrete examples. It should not infer from a person’s behavior that they consented to a plan they never saw. The person should be able to edit or abandon the plan. The intervention’s job is to help turn a chosen intention into a workable action, not to create an invisible contract with an algorithm.
Arrange choices without hiding them
Choice architecture concerns how alternatives are presented: defaults, ordering, prompts, information and the effort needed to select or reverse an option. Hollands and colleagues argued for careful definitions and an evidence base for interventions that alter small-scale physical environments, such as the placement or availability of options. Their framework reminds us that a change to a setting may influence many people without each person receiving a lecture.
A water fountain that is easy to find, a form that states its alternatives plainly or a walking map that marks accessible routes can reduce unnecessary friction. But architecture can also conceal costs or make dissent hard. A preselected subscription, an obscure cancellation path or a false claim about what “most people” do can convert a convenience feature into a pressure device. The ethical question cannot be answered by the word “nudge” alone.
Mertens and colleagues synthesized choice-architecture studies and reported an overall behavioral effect in their original analysis, alongside substantial variation and evidence of publication bias. A subsequent correction to the paper removed a retracted study’s observations and addressed coding and data errors. That history is one reason not to treat a pooled number as a guarantee for a new setting. In any case, a behavioral effect says little by itself about whether people understood or approved the change.
A useful design review asks whether someone can identify the desired action, the alternatives and the reason the arrangement exists. It asks how to refuse, how to reverse a selection, and whose interests are served by friction on each path. These questions are an Institute synthesis from the evidence and ethics discussion, not an established numerical test of agency. They should be answered with users, especially those for whom the interface or physical environment poses additional barriers.
Agency requires capability and opportunity
Offering a choice is weak if only one option is feasible. An employee may technically be free to take a walking break but have no paid time to do so. A resident may be invited to use a service that requires a phone they do not own. A patient may receive a detailed plan in a language they cannot read comfortably. In each case, measuring only motivation risks misdiagnosing the constraint.
The capability-opportunity-motivation lens encourages a team to inspect each layer. Does the person know what action is proposed and how to do it? Are the time, tools and accessible conditions available? Does the action make sense in relation to their own aim? These are questions, not a formula that automatically discovers the answer. Qualitative accounts, observation and direct feedback can expose constraints a dashboard misses.
These questions can change who is invited into design work. If a route is planned only by people who can walk it easily at midday, the team may miss the experience of someone using a mobility aid at dusk. If an app is tested only with fluent readers and new phones, it may appear effortless while excluding people who would benefit from the service most. Participation does not mean one participant can represent every group. It means the design process actively looks for different constraints and makes room to alter the proposal when they are found.
Agency also includes the ability to pursue a different reasonable goal. A person who chooses rest over a walk after a demanding shift has not necessarily failed to understand the message. The program’s target may be sensible at a population level and still be wrong for that individual on that day. Treating every deviation as an error trains the system to ignore context. Recording why people decline, with their consent and without penalty, can improve the service while preserving the legitimacy of refusal.
Environmental redesign can sometimes be more respectful than repeated persuasion. For the walking route, lighting, benches, clear maps and winter maintenance may change what is possible. They also benefit people who never open an app. Yet such changes have costs and tradeoffs. The town should discuss them with residents and measure use, safety and accessibility, rather than assuming that every increase in foot traffic is an improvement for everyone.
Evaluate the person’s outcome as well as the program’s
A program might report that 30 percent more people entered the route after new signs appeared. That is a useful observation, but it does not isolate the signs from season, weather or a local event. A better evaluation records the intervention date, comparison locations or periods where feasible, who uses the route and whether people feel safe. It should also notice people who avoid the route and ask why.
For a digital intervention, record more than opens and streaks. Did users understand the prompt, change settings, stop notifications, regret an action or continue after the prompts ended? Did the system make it easier to reach a personally meaningful outcome? Repeated engagement may be a side effect of a useful service, but it is not the same as benefit. A design that optimizes its own usage can drift away from the user’s reason for arriving.
Agency can be examined through several imperfect measures: understanding, perceived choice, ability to decline, ease of reversal and observed distribution of burdens. None is complete. Self-reports may be influenced by expectations; behavior alone may conceal pressure. A mixed evaluation gives people room to describe why they acted and tests whether the intervention’s benefit persists without coercive prompts.
Every evaluation also needs a boundary. A study with students in a university app may not transfer to older adults using municipal services. A short experiment may miss later dependence on reminders. When findings are uncertain, a small reversible pilot can be more responsible than an irreversible rollout, especially for high-stakes choices. The pilot should include a route for complaints and changes, not just a plan to maximize the favored action.
Before testing, the team can state what result would prompt a change or stop. A rise in route use alongside more reported near misses at the crossing would demand attention even if the campaign met its traffic target. In a digital service, a higher completion rate accompanied by confusion about fees should not be celebrated as uncomplicated success. Such rules are local governance choices rather than findings from the cited habit experiments; naming them makes the evaluation accountable to the people affected.
What the evidence does not show
The behaviour change wheel is a useful organizing model, but it does not prove that each intervention function is effective or ethically acceptable in every context. Wood and Neal’s habit account and Lally and colleagues’ study do not establish a fixed number of repetitions after which behavior becomes automatic. Habit strength, context and individual circumstances differ.
The implementation-intention meta-analysis does not mean an if-then statement will overcome a lack of opportunity or an unwanted goal. Choice-architecture studies demonstrate effects on measured behaviors in studied settings; they do not demonstrate that every changed choice preserves autonomy. Publication bias, corrections and heterogeneous interventions make broad effect claims especially fragile.
Nor does this entry establish one universal score for “agency.” The proposed questions about explanation, alternatives, refusal and reversal are a practical synthesis. Their answers need to be checked with the people affected by a real design, alongside the behavioral outcome and any unequal costs. The strongest claim supported here is modest: matching an intervention to a real constraint, allowing revision and measuring more than compliance makes the architecture more accountable.
Return to the walking route
The town can start by asking residents why the route is not used. Perhaps the issue is lighting, winter ice, limited time or no desire to walk there. Its response should fit the answer. Signs and maps may help some people; safe crossings and accessible surfaces may matter more for others. No resident needs to be shamed into demonstrating that a campaign worked.
Behavior change architecture preserves agency when it improves the conditions for a chosen action and remains open to correction. It makes a route visible, a plan workable and a decision reversible. The measure of success is not simply that more people move along the intended path, but that the path serves purposes they can recognize as their own.
Research references
- Michie, S., van Stralen, M. M., & West, R. (2011). The behaviour change wheel: A new method for characterising and designing behaviour change interventions. Implementation Science, 6, 42. View source
- Wood, W., & Neal, D. T. (2007). A new look at habits and the habit-goal interface. Psychological Review, 114(4), 843–863. View source
- Lally, P., van Jaarsveld, C. H. M., Potts, H. W. W., & Wardle, J. (2010). How are habits formed: Modelling habit formation in the real world. European Journal of Social Psychology, 40(6), 998–1009. View source
- Gollwitzer, P. M., & Sheeran, P. (2006). Implementation intentions and goal achievement: A meta-analysis of effects and processes. Advances in Experimental Social Psychology, 38, 69–119. View source
- Hollands, G. J., Shemilt, I., Marteau, T. M., et al. (2013). Altering micro-environments to change population health behaviour: Towards an evidence base for choice architecture interventions. BMC Public Health, 13, 1218. View source
- Mertens, S., Herberz, M., Hahnel, U. J. J., & Brosch, T. (2022). The effectiveness of nudging: A meta-analysis of choice architecture interventions across behavioral domains. Proceedings of the National Academy of Sciences, 119(1), e2107346118. View source
- Correction for Mertens et al. (2022). The effectiveness of nudging: A meta-analysis of choice architecture interventions across behavioral domains. Proceedings of the National Academy of Sciences, 119(19), e2204059119. View correction