What Good Evidence Can and Cannot Tell You

A social-norm intervention should be evaluated at the link it claims to change. The practical task is to move from a visible message to a verified perceived norm, an observed behavior, and—when durability matters—a behavior that persists. This framework uses the supplied field, cross-cultural, and review evidence to show why each step exists, what it can establish, and what confidence it still cannot provide.

Measure the link you want to claim

The first mistake in evaluating norm messaging is to treat exposure as though it were a mechanism. A poster can be installed, an email can be delivered, or a comparison letter can be mailed without changing what the recipient believes about the reference group. Even when the message is read, the recipient may not understand whether it describes prevalence, approval, or both. A credible test therefore begins by specifying the target behavior and the social proposition that is supposed to influence it.

flowchart LR A["Define behavior and practical constraint"] --> B["Specify reference group and existing norm"] B --> C["Test notice, believability, and perceived norm"] C --> D["Measure observed target behavior"] D --> E["Follow up for maintenance"] E --> F["Compare with an active alternative"] F --> G["Synthesize bias and heterogeneity"] G --> H["Scale, adapt, or stop"] C -.-> X["Revise message or context if the norm was not engaged"] D -.-> Y["Do not upgrade intention to behavior"]

The first gate is behavioral specificity. “Promote sustainability” is too broad to tell an evaluator what should change. Using fewer paper towels, declining a plastic lid, putting on a mask, reducing food waste, or changing a prescription are different actions with different frictions and time horizons. A norm appeal can be relevant to one and irrelevant to another. Defining the action also forces the evaluator to name the practical constraint. If a lid prevents a hot drink from spilling, declining it is not merely a question of social approval. If a person must reach a clinic, purchase a product, or reorganize a meal, information about peers cannot remove that requirement by itself.

The second gate is the reference group. A message should specify whose behavior or approval is meant to matter and why that group is recognizable to the recipient. The relevant comparison may be fellow students, colleagues in a practice, visitors in the same facility, or people entering the same place. The claim is not that “society” behaves in one way. It is that a particular group commonly does something or regards it as appropriate. That distinction matters for both interpretation and measurement: an evaluator should be able to ask whether participants recognized the group and considered its behavior or judgment relevant.

The third gate is message engagement. A manipulation check can test whether people noticed and understood the normative information. Perceived descriptive norms can be measured by asking how prevalent participants believe the behavior is among the reference group. Perceived injunctive norms can be measured by asking how strongly they believe the group approves of or expects the behavior. Believability checks add a related safeguard: a message that changes responses because it appears absurd is not functioning as intended. These measures do not prove behavior change, but without them a null outcome cannot be located in the causal chain.

The broad review shows the cost of skipping this gate.