Economists have long argued that contact between social groups can ease prejudice and correct stereotypes. A new field experiment in India suggests that how people meet may be as important as whether they meet—because short interactions with many outgroup members can shape beliefs differently than sustained exposure to just one.
In a study led by Anujit Chakraborty, Arkadev Ghosh, Matt Lowe, and Gareth Nellis, researchers tested two forms of intergroup contact: “deep” contact, defined as repeated, longer engagement with the same outgroup member, and “broad” contact, defined as brief interactions spread across multiple outgroup members. The team hypothesized that these formats would influence learning and relationships in distinct ways.
To examine this, the researchers set up a factory setting in India and recruited Hindu and Muslim men to participate. Workers were paired to complete joint production tasks, creating repeated opportunities for collaboration. Participants were randomly assigned to one of several conditions: a deep-contact arrangement in which a person worked every day with the same partner (either from their own group or the other group), a broad-contact arrangement in which a person worked each day with a different outgroup partner, or a control group.
The results show a clear split between relationship-building and belief-updating. When workers stayed with the same outgroup partner over time, they formed stronger social and economic connections with that individual—evidence that repeated cooperation can deepen cross-group ties on a person-to-person basis.
But when it came to correcting misperceptions about outgroup strangers, the broad-contact approach did better—especially for members of the majority group. Only those who cycled through multiple outgroup partners reduced their mistaken beliefs about unfamiliar outgroup members. The authors interpret this as evidence that interacting with a wider sample of people provides more representative information, making it easier to update beliefs about the group as a whole.
A key part of the analysis uses differences in partners’ productivity measured before the intervention. That within-treatment variation allows the researchers to test whether participants actually learned from who they were paired with. They find that workers in both the deep and broad conditions learned about their own partners. The crucial difference is what happened next: participants in the broad-contact condition were more likely to extend what they learned—generalizing from their experiences with specific partners to beliefs about strangers from the outgroup.
The findings fit a simple model of learning: independent draws from a population (meeting many different people) generate better information about the population than repeated interaction with a single, potentially unrepresentative, individual. Prolonged engagement can build trust with one person without necessarily changing how someone perceives the broader group.
Notably, the study finds limits to both approaches. Despite measurable learning and, in the deep-contact case, stronger cross-group ties with a specific partner, neither type of contact altered behavior toward the wider outgroup.
The project was pre-registered in the AEA RCT Registry (AEARCTR-0013126) and received ethics approval from the University of British Columbia (H23-03476), Duke University (2024-0213), UC San Diego (809306), and UC Davis (2136225-1). Funding came from multiple sources, including the Weiss Fund for Research in Development Economics, UCSD Senate General Research, Duke’s Social Science Research Institute, CEGA’s Development Economics Challenge, and the Institute for Humane Studies.
Why it matters
Efforts to reduce social division often rely on bringing groups into contact, but this evidence suggests program design is pivotal: meeting many outgroup members may do more to correct broad misperceptions than repeated contact with one. At the same time, the lack of behavioral change toward the wider outgroup underscores how hard it is to translate belief-updating into action. Future interventions may need to combine broad exposure with mechanisms that shift incentives and norms beyond the immediate interaction.