Ask a man in Saudi Arabia whether he supports his wife working outside the home, and he likely says yes. Ask him to estimate how many of his neighbors agree, and he will likely guess that most do not. In one widely-cited study, 87% of young married men privately approved of women working outside the home but the average respondent estimated that only 63% of his peers did. When researchers told a random subset of these men what the others actually believed, the wives of those men became more likely to have applied and interviewed for jobs outside the home.
This suggests that some restrictive norms are not really about preferences at all; they are about coordination failures. People conform to what they believe others expect, even when they themselves would prefer otherwise, and even when their belief about what others expect is wrong. Correct the belief and you may shift the behavior.
This is a powerful insight. But it is one of several mechanisms that the field has identified by which gender norms can be moved. Here’s what we know.
What we mean by a norm
What we casually call “a norm” is at least three things. There are personal beliefs (what I myself think is right); there are descriptive norms (what I believe others actually do); and there are injunctive norms (what I believe others expect me to do). These three don’t always line up. A woman may privately wish to work, believe that her neighbors disapprove of women working, and believe that her mother-in-law would sanction her if she tried — and the last two beliefs may be wrong. This conceptual vocabulary comes mainly from social psychology, especially work by Bicchieri (2006, 2016) and Cialdini et al (1990). Recent World Bank policy notes (here and here) bring this vocabulary into development policy discussion.
So how does change happen? The empirical literature points to several distinct mechanisms. They work in different ways, on different timescales, and under different conditions.
Mechanism 1: Correct beliefs about others
The first mechanism is information correction. If many people privately disagree with a norm but each believes the others endorse it, simply telling them the truth can shift behavior—the underlying preferences are already in place; what is missing is common knowledge.
Evidence from other settings supports the basic insight. A cross-country study documents that misperceptions about other men’s support for women’s right to work are widespread — the majority of male respondents privately endorse women’s freedom to work but believe this view is held only by a minority of their countrymen. A complementary experiment finds the analogous pattern on the female side: Saudi female university students systematically underestimate the expected labor force attachment of their female peers, and providing information about peers’ aspirations raises students’ own expected labor force participation. In Ethiopia, providing men with accurate information about their peers’ acceptance of domestic chores increased men’s participation in tasks such as firewood collection.
Information correction is attractive to policymakers because it is cheap, scalable, and politically neutral. However, it requires that private preferences have already shifted but social expectations have not caught up — a society somewhere mid-transition. Where preferences themselves are restrictive, information about others is the wrong lever entirely. Even where the diagnosis is right, information-style interventions are not guaranteed to work, as this study found.
Mechanism 2: Change minds before they harden
A second strand of evidence suggests that gender attitudes are most malleable in adolescence and that targeted classroom interventions can shift them. The leading recent piece of evidence comes from a school-based RCT in Haryana, India. A multi-year program that engaged adolescent boys and girls in classroom discussions about gender equality shifted attitudes in the direction of equality. Self-reported behavior also became more gender-equal, particularly among boys, though evidence for revealed-preference behavior change was weaker. When participants were surveyed two years after the program ended, the attitude effects had persisted.
This kind of effect operates through a different mechanism than information correction. The Haryana program did not tell students what their peers privately believed; it created space for sustained discussion and reflection. Whether the change will translate into different behavior in marriage and the labor market is the next question, and the answer is several years away. But the underlying evidence is some of the cleanest that we have on what classroom-based programs can do when they catch people early enough.
Mechanism 3: Provide role models
A third mechanism is exposure to counter-stereotypical examples. The most influential evidence here is Beaman et al (2012). India’s 1993 constitutional amendment randomly reserved one-third of village council leadership positions for women in randomly selected councils, producing a natural experiment in exposure to female political leaders. After two election cycles with a female leader, the gender gap in parents’ aspirations for their adolescent children closed by 20%, the gap in adolescents’ own aspirations closed by 32%, the gender gap in adolescent educational attainment was erased entirely, and girls spent less time on household chores. The authors find no evidence of changes in local labor market opportunities, which suggests that the effects operate through a role-model channel rather than through changes in actual returns.
Other work in this vein has expanded the toolkit. Riley (2024) randomized Ugandan secondary school students to watch either Queen of Katwe, a film about a Ugandan girl from a poor background who becomes a chess champion, or a placebo film, immediately before their national exams. Lower-secondary students who watched the role-model film passed their math exam at higher rates; upper-secondary students scored higher overall; and female students were more likely to remain in education in subsequent years, closing the gender gap with their male peers. An older but still important paper, Jensen and Oster (2009), exploits the staggered introduction of cable television in rural India and finds that exposure to outside media reduced the reported acceptability of domestic violence and son preference, increased women’s reported autonomy, and reduced fertility. The estimated effects were equivalent to several years of women’s education.
Role-model interventions can be delivered through mass media at low marginal cost. Their limitation is that they often shift aspirations and attitudes without shifting the structural conditions that determine whether those aspirations can be acted on.
Mechanism 4: Change what people do, and norms follow
A fourth mechanism inverts the standard policy chain. Rather than changing beliefs to change behavior, it changes the structural environment so that new actions become possible and waits for norms to adjust.
The cleanest recent evidence comes from a study conducted in the Indian state of Madhya Pradesh. The authors worked with the state government to randomize whether rural women’s public workfare wages were deposited into their own bank accounts (with training) or into the household head’s account (the status quo). In the short run, women whose wages were deposited into their own accounts worked more in both the public workfare program and, notably, in the private sector. The latter is the telling result: if norms had not been binding, the policy treatment should not have moved private employment. Three years later, women in the treatment group held more liberal work-related norms themselves and perceived their communities as holding more liberal norms. Behavior shifted first, attributable to a structural change in financial control; norms followed.
Whose perceptions actually matter?
Across all four mechanisms, the person whose perceptions need to change is not always the person whose behavior is at stake. Bernhardt et al (2018) found in Madhya Pradesh that both husbands and wives overestimated community opposition to women working for pay but only husbands’ perceptions, not wives’, were correlated with whether women actually worked. In extended households — which describe much of South Asia, parts of sub-Saharan Africa, and parts of East Asia — the relevant power-holder may be further removed still. As my work has shown, mothers-in-law often play a gatekeeping role in their daughters-in-law’s mobility, fertility, and labor force participation, and their preferences and beliefs about community norms can govern outcomes that ostensibly belong to a younger woman.
This has direct design implications. An information experiment that corrects a daughter-in-law’s beliefs about her community is unlikely to help if it is her mother-in-law’s belief that governs her time. A school-based curriculum aimed at adolescents may shift attitudes but fail to translate into behavior if the adolescents return to households where the older generation enforces a different set of norms.
Two things we still need to know
How long do effects last? Most evaluations of norm-change interventions measure outcomes within one to three years. We know strikingly little about whether changed attitudes carry into marriage, into the next generation, or into long-run labor force outcomes. Nor do we know whether the effects survive at-scale delivery through routine bureaucratic implementation rather than research-led pilots.
When does change provoke backlash? A small but important literature documents that norm-change efforts can produce backlash: increased sanctions, intensified gossip, or political reversal. A recent paper finds unintended adverse effects of an intimate partner violence prevention program in Rwanda. The conditions under which interventions produce uptake versus backlash are not well understood. The risk appears higher when change is pushed faster than the supporting institutional and economic environment can accommodate but this is more a hypothesis than a finding.
What this means for policy
Norms can change. The challenge for policy is choosing the right mechanism for the right constraint. What is actually binding here? Is it a misperception of others’ beliefs, an underlying preference, the lack of a viable role model, the structure of a household, or the absence of jobs and infrastructure? Different answers point to different instruments. Information correction will not move norms when underlying preferences remain restrictive. Role models close aspiration gaps but leave structural barriers untouched. The temptation to treat any one mechanism as the lever for gender norms generally is one the recent literature should make us resist.
Then there is a measurement implication. Most of what we record on surveys are personal attitudes. The literature increasingly suggests that what we should also be measuring is what people believe others believe, and what people believe powerholders in their own households believe. That is a more demanding survey design than the field has historically used, but essential if we want to understand norms better.
Join the Conversation