Descriptive versus Injunctive Norms
What people do and what people approve of are different messages, and mixing them up backfires.
A descriptive norm states what people actually do. An injunctive norm states what people approve of. Telling someone that a bad behaviour is common gives them permission for it, and telling a better-than-average person what the average is pulls them down toward it. Adding an approval signal removes that downward pull.
The norm boomerang
How it shows up in software
Any comparison feature carries both norms whether the team intends it or not. Usage digests, team leaderboards, security nudges, and spending comparisons all tell a user where they sit relative to others. Teams reliably ship the descriptive half and forget the injunctive half, then wonder why their best-behaved users regressed.
Using it well
- Attach an approval signal to every comparison, so a user who is already ahead is told they are doing the right thing rather than only told they are unusual.
- Never advertise low compliance; if only a fraction of users have enabled two-factor authentication, say it matters instead of saying how few do it.
- Choose the comparison group so the desired behaviour is the majority within it.
- Segment the message: users above the norm need reinforcement, users below need the comparison.
Where it turns manipulative
- Publishing a low-adoption number to shame users, such as telling someone they are behind most of their team, punishes the group you most need to keep and often produces resentment rather than change.
- Using comparison to drive spending, by showing a user that similar customers pay for more seats, dresses a sales quota as a norm.
- Reporting a descriptive norm about a harmful behaviour, in the hope that naming it discourages it, spreads the behaviour; the Petrified Forest result is the direct warning.
Where you have seen it
Opower utility reports
Pairs a neighbour comparison bar with a smiley rating, which is the injunctive cue that keeps efficient households from drifting upward.
Google Workspace admin console
Reports the share of accounts with two-step verification enabled alongside a recommendation to enforce it, so the number arrives with a stated standard.
Apple Screen Time
Reports the week-over-week change in the user's own usage rather than a comparison to other people, which sidesteps the boomerang entirely.
What the research says
- Cialdini, Demaine, Sagarin, Barrett, Rhoads and Winter, 2006Well evidenced
Signs at Petrified Forest National Park saying that many past visitors had removed wood increased theft of wood, while a sign asking visitors not to remove it reduced theft relative to the descriptive sign.
Field experiment with real theft as the measure, run in the actual setting. One of the cleanest demonstrations that a well-meant descriptive message can cause harm.
- Schultz, Nolan, Cialdini, Goldstein and Griskevicius, 2007Well evidenced
Households told how their energy use compared to the neighbourhood average moved toward it from both directions; low users increased consumption. Adding a smiley or frowning face, an injunctive cue, removed the increase among low users.
Randomised field experiment with metered consumption. The boomerang and its fix are both well supported here.
- Allcott, 2011Well evidenced
Large-scale evaluation of Opower home energy reports, which carry both comparison and approval cues, found small persistent reductions in household consumption across hundreds of thousands of homes.
Effects are real but small per household; the programme's value comes from scale, not from a large individual change.
Grades are a judgement about the evidence, not about how useful the idea is. Plenty of contested effects are still worth knowing, as long as you do not cite them as settled.