The Button That Was Supposed to Change Behavior

This series examines social-norm messaging—the use of information about what a reference group does or approves—to ask whether it is a reliable way to change behavior. The focus is not whether a clever sign, email, or comparison can ever work. It is what has to happen between exposure to a norm message and a person’s action, and whether that chain holds across people, settings, cultures, and time. By the end, the useful judgment is conditional: when is a norm appeal worth testing, and when is it being asked to do work that the evidence cannot support?

Why social-norm messaging became a behavior-change favorite

The appeal is easy to understand. A public-health team, school, employer, or business can tell people that others like them are already doing something, or that those people approve of it. The intervention may require no subsidy, new facility, or prohibition. It changes the information around a choice while leaving the formal options open. In that sense, social-norm messaging is often treated as a form of “nudge”: a light intervention intended to redirect behavior without removing freedom of choice.

The message can take several familiar forms. “Six million people have attended an NHS health check” uses social proof: it presents participation as widespread. A clinician may receive feedback that most comparable practices reduced or stabilized antibiotic prescribing, while their own practice belongs to a minority that increased it. That is social comparison, using a reference group to make a person’s position visible. An appeal can also describe what the group believes people ought to do, such as saying that most students consider preventing food waste important. These messages differ in wording, but they share an assumption: people care about how their behavior fits with a social world, and information about that world can change what they do.

That assumption matters because many important behaviors are collective in their consequences. A single person’s hand hygiene, food waste, vaccination, prescription, or use of disposable materials may seem small, but the aggregate outcome is not. A low-cost message that shifts even a small fraction of choices can look attractive to institutions responsible for population-level outcomes. It also promises an efficient route around a difficult problem: instead of changing every person’s private motivation, perhaps the intervention only needs to correct a mistaken picture of what everyone else is doing.

But efficiency is not the same as reliability. A message may be inexpensive to deliver and still fail to be noticed, believed, or treated as relevant. A person may agree with the group and still lack the time, opportunity, money, safety, or practical reason to act. A message may produce an immediate choice without changing a habit. And a published literature can make an intervention look more successful than it is if positive findings are easier to report and circulate than null findings. The important question is therefore not whether social information can matter. It is whether a norm message can be trusted as a general behavior-change button.

The message is not the behavior

To answer that question, the subject needs two distinctions. A descriptive norm is information about what people in a relevant group commonly do. An injunctive norm is information about what that group considers appropriate or approves of. If a message says that most students reduce food waste, it supplies prevalence information. If it says that most students consider wasting food unacceptable, it supplies approval information. The first can act as a shortcut when someone is unsure what is normal or effective. The second can engage the desire for social acceptance and the wish to avoid disapproval. Both may point toward the same action, but they do not operate through exactly the same motivation.

The relevant group is just as important as the message type. “What people do” is not a universal fact in the abstract; it is a claim about a reference group that the recipient recognizes as socially meaningful. A norm appeal can be clear yet irrelevant if the people used as the comparison do not feel like peers. It can also be accurate and still fail if the recipient’s immediate circumstances make conformity less important than avoiding a spill, catching a train, protecting personal health, or completing a task.

The proposed causal path is consequently longer than the advertisement or sign suggests. First, the recipient has to encounter the message and attend to it. Next, they have to interpret it as credible information about a group, forming or revising a perceived descriptive or injunctive norm. That perception may influence an intention or an immediate decision. Only then can the behavior occur, and even an observed action says little by itself about whether the change will persist after the message disappears. A norm appeal can succeed at one link and stop at the next.

flowchart LR M["Norm message"] --> P["Perceived descriptive or injunctive norm"] P --> I["Intention or immediate choice"] I --> B["Observed behavior"] B --> S["Maintained behavior"] A["Attention"] -.-> P O["Opportunity"] -.-> I G["Motivation"] -.-> I C["Competing constraints"] -.-> I

This chain also explains why the same message can have opposing or disappointing consequences. A descriptive norm that sets a mediocre average may encourage people below it to improve, but it can also give people already doing better a reason to relax: matching the average is presented as sufficient. Researchers call this a boomerang effect. Injunctive information can help signal approval, yet a strongly controlling instruction can provoke reactance—the impulse to resist a perceived restriction. Humor or a memorable visual may make a message more noticeable, but that extra salience does not prove that the norm itself caused the final choice.

The distinction between endpoints is more than a technicality. Perceived norms are psychological measures. Intentions are plans or stated willingness. Observed behavior is what someone actually does in a particular situation. Maintained behavior requires evidence over time. The broad research literature often measures the last two without measuring the first link. In the large review examined later in this series, only a small minority of the included studies measured participants’ perceived norms before the intervention, and even fewer used manipulation checks to verify that people understood the intended norm. Without those checks, a null result may mean that the appeal was ineffective—or simply that the proposed mechanism never engaged. A positive result, by the same logic, does not automatically prove that norm perception produced it.

The real question is reliability

The supplied evidence already shows why a binary answer would be misleading. The broadest study is a preregistered systematic review and meta-analysis of randomized trials involving people aged 16 and older in developed countries, focused on health behaviors. It brought together 89 individual studies from 60 papers and more than 85,000 participants. In a conventional random-effects analysis, the average effect of social-norm messaging was statistically positive but small. The individual studies were much less uniform than that single average: some reported moderate or large benefits, many found trivial effects, and some found small negative effects.

Then the analysis asked a harder question about the literature itself. Positive effects were overrepresented among less precise studies, a pattern consistent with publication bias. A Bayesian model without the adjustment reproduced the small positive signal. After robust adjustment for publication bias, the estimated overall effect was close to zero and the evidence favored no reliable average effect. The review’s conclusion is not that every norm appeal is inert. It is that the apparent small benefit is not a stable basis for claiming that social-norm messaging improves health behavior in general. Its tested categories—message type, delivery mode, health domain, and population—also did not reveal a dependable recipe after the bias-aware analysis.

At the same time, a set of five field experiments found a more conditional pattern. A humorous “Be like Toni” sign reduced paper-towel use in two restroom settings, both prescriptive and proscriptive versions worked against a control in one of those settings, and a related sign increased voluntary mask wearing in a bakery. The same message family did not reduce plastic-lid use at a train-station coffee shop. The contrast is not a refutation of the broad review. It shows what an aggregate result conceals: a calm setting in which the action is easy to perform is different from a rushed setting in which a lid may prevent a spill and a traveler’s attention is elsewhere. These are observed contextual differences; the proposed roles of stress, inattentional blindness, and urgency remain explanations to test, not settled causes.

Culture introduces another layer. In an experiment with students in China and the United States, messages varied both the reported prevalence of food-waste prevention and the level of social approval. The messages changed perceived norms, and those perceptions were linked to intentions more strongly in the Chinese sample. Yet the study measured intentions rather than actual behavior, used different survey modes in the two countries, and cannot rank entire cultures by receptivity. It demonstrates that the same normative information can be interpreted within different cultural and institutional environments—not that a national label predicts what every person will do.

Taken together, these findings support a precise meaning of “not reliable.” Social-norm messaging is not a useless idea, and a local behavioral effect can be real. But the intervention is not a general-purpose button whose success can be inferred from a persuasive example, a statistically significant average, or a message format that worked somewhere else. Confidence depends on whether the recipient recognizes the reference group, notices and believes the appeal, has a practical opportunity to comply, and is measured at the endpoint that matters. It also depends on whether the effect survives an honest comparison and follow-up.

That is the standard this series will apply: treat the norm message as a hypothesis about a particular behavior in a particular social and physical setting. The reliability question begins only after the message has been separated from the perception it is meant to change—and from the behavior it hopes will follow.

References