In-group bias

The wristband teams

At a company retreat, everyone draws a red or blue wristband to form two teams for the afternoon’s activities. At the end, each person votes for the best presentation of the day and splits a pot of prize points between two people they don’t know, one from each team. Most people’s votes go to their own team, and most give the larger share to the person wearing their color.

The teams were formed by chance an hour earlier, so a wristband color carries no information about whose presentation was better or who deserves more points. This is the clear-cut form, and the one built into the research’s classic design: favoritism appears when nothing but a trivial label separates the groups (see Evidence). Nothing here requires disliking the other team either; people simply give more to their own.

Tough play or a dirty foul

Fans of the Harbor City Gulls and the Ridgeview Foxes watch the same late-game collision. A Gulls fan says, “That’s just playing hard, he went for the ball.” When a Foxes player makes a nearly identical challenge ten minutes later, the same fan says, “That’s reckless. He should be suspended.”

The two plays are the same kind of behavior, judged differently depending only on which team’s player made them. The bias here isn’t about handing out rewards; it has moved into how the evidence is read. The fan isn’t lying, and each judgment probably feels like a straightforward call.

A good fit

Two candidates for an engineering job have similar experience and similar interview scores. One went to the same university as the hiring manager. In the debrief, the manager says, “I just think she’d fit in better with how we work. People from that program are really well trained.”

This one sounds respectable. “Fit” and “training” are real considerations, but the manager has no evidence about this candidate’s training beyond the shared school, and the interview didn’t show a difference. A group the manager belongs to has been turned into a general judgment about an individual in it, much as a halo lets one favorable impression spread to unrelated traits.

• In-group favoritism versus out-group derogation. Favoring your own group (“in-group love”) is different from harming the other one (“out-group hate”). Most of the laboratory evidence shows the first; the second appears under narrower conditions (see Evidence).
• The positive–negative asymmetry. In minimal-group studies, people favored their own group when handing out something good but generally didn’t when handing out something unpleasant. Groups made to feel inferior or outnumbered were the exception.
• Favoritism in judgment versus favoritism in cooperation. Rating an in-group member’s work more highly is a biased judgment. Cooperating more with in-group members can instead reflect an expectation, sometimes correct, that they will cooperate back.

Treating members of your own group differently is often reasonable:

• When membership carries real information. A professional license, a club that requires a skills test or a team that has worked together for years tells you something about its members. Using that is inference, as long as it’s weighted no more than it deserves and gives way to evidence about the person.
• When you have duties to the group. A team captain who spends team funds on the team, or a parent who looks after their own children first, is keeping an obligation, not misjudging anyone.
• When cooperation is actually reciprocal. Helping people who can and will help you back, and whom you’ll deal with again, is sensible. The error is assuming outsiders wouldn’t reciprocate without evidence.
• When you’ve observed the individual. Favoring someone you’ve watched do good work isn’t favoritism because they share your group.

The test: would I judge this work, this behavior or this person the same way if they belonged to the other group?

A reference from the trail

A hiking club needs someone to lead a difficult trail crew. One member recommends another: “I’ve been on ten hikes she led, including two where the weather turned. She kept everyone together and made the call to turn back both times.”

The recommendation goes to a fellow member, but it rests on direct observation of the individual doing the very thing the role needs. Someone outside the club with the same record would deserve the same recommendation; the club is simply where the evidence was gathered.

Dues spent on members

A community choir has $600 left from members’ dues at the end of the year. The treasurer proposes spending it on a dinner for members rather than donating it to a different choir’s trip.

This favors the in-group over an outside group, but the money was paid by members for the choir’s purposes. Spending it on them is a duty to the group, not a judgment that members are more deserving people. It would become in-group bias if the treasurer argued that the other choir wouldn’t appreciate a donation, with nothing to go on but the fact that they’re a different choir.

Researchers have proposed several explanations, and the evidence suggests more than one is at work.

• Social identity. Groups become part of how people see themselves. Seeing your own group as better than others raises your own standing, so people are motivated to favor it. This was the explanation developed from the original minimal-group experiments.
• Expected reciprocity. On a second view, people carry a working assumption that in-group members will look out for each other, so helping them is a good bet. Consistent with this, favoritism in cooperation is stronger when people depend on each other and when both sides know who belongs to which group.
• Projection. People expect others to be like them, and they expect this more of in-group members than of outsiders. A 2005 meta-analysis found projection stronger toward in-groups, and its authors argued that this can contribute to in-group favoritism and to cooperation with in-group members. The same tendency to read others off yourself underlies the False consensus effect.
• Sheer difference. A 2025 set of experiments found that people also give less to a single individual who merely differs from them on something trivial, with no groups at all. Part of the minimal-group effect may come from that simpler reaction to difference.

• Hide the label when judging the work. Removing names, team colors or school names before evaluating takes away what the bias works on. It’s a common practice rather than a remedy tested against this bias specifically. In a large study described below, participants who said they preferred not to know a partner’s group didn’t significantly discriminate between groups, even when they were told it anyway (an exploratory analysis). They chose that preference themselves, though, so the study doesn’t show that hiding the information would reduce favoritism in others.
• Swap the teams in your head. Ask how you’d judge the same play, proposal or behavior from the other side.
• Separate loyalty from assessment. It can be fine to help your own group more; it isn’t fine to believe its members did better work than they did.

Status: replicates robustly. In-group favoritism has been found in meta-analyses of well over a hundred studies and, most recently, in a preregistered study in 20 countries that found it in every one. Its size varies widely with the setting and the culture, and why it happens is still debated.

• Tajfel, Billig, Bundy and Flament (1971) ran the first minimal-group experiments. People were divided into groups on trivial grounds (overestimating or underestimating the number of dots they were shown, or preferring paintings by Klee or by Kandinsky), then used specially designed matrices to divide real rewards and penalties between other participants. They favored their own group, even though nothing but a fairly irrelevant classification separated the groups. In the second series of experiments, the most total reward for everyone didn’t noticeably affect how people divided money, the most reward for their own group did, and the clearest effect was a tendency to maximize the difference between the groups, even at the price of sacrificing other advantages.
• Mullen, Brown and Smith (1992) meta-analyzed 137 tests of in-group bias and found a highly significant effect of moderate size. It was stronger when the in-group was made more salient, through its relative size or by being a real rather than an artificial group, and it interacted with the group’s status.
• Otten, Mummendey and Blanz (1996) found in three studies that people discriminated in the in-group’s favor when allocating positive things but not negative ones, except when being in a lower-status or minority group threatened their social identity.
• Brewer (1999) reviewed cross-cultural and laboratory evidence and argued that identifying with an in-group is independent of negative attitudes toward out-groups, and that much intergroup discrimination is preferential treatment of insiders rather than direct hostility toward outsiders.
• Balliet, Wu and De Dreu (2014) meta-analyzed studies of cooperation in economic games. People cooperated more with in-group than out-group members, a small to medium effect (d = 0.32). It was weak where group membership was the only factor and nobody depended on anybody (dictator games, d = 0.19) and stronger where people depended on each other (social dilemmas, d = 0.42), and stronger when both sides knew each other’s group. The analysis supported in-group favoritism rather than out-group derogation as the source of the difference.
• Rahal and Schulze Spüntrup (2025) ran a preregistered study with about 1,800 participants (after exclusions) in 20 countries. People were assigned to minimal groups based on a color perception task and then divided points between themselves and other players. Facing an in-group rather than an out-group member multiplied the odds of a prosocial choice (giving up a little to benefit the other player) by about 4.6, and favoritism appeared in every country studied, though its size varied substantially between them.

What remains uncertain.

• Why. Social identity and expected reciprocity both have support, and they predict different moderators. Deschrijver and Ramsey (2025), in seven preregistered experiments with more than 1,400 participants, found that people divided resources unequally against a single other person who merely gave a different dot estimate, preferred a different painting or got a different coin flip, with no groups involved. The group versions of their tasks still produced favoritism, somewhat more than the individual versions in an exploratory comparison, so the finding challenges the explanation rather than the effect.
• When outsiders are actually harmed. Out-group derogation is uncommon in standard cooperation games, but it isn’t absent everywhere. Imada and colleagues (2025), in two preregistered studies of choices that would affect future generations, found both favoritism and derogation: compared with people who didn’t know the future generation’s group, people cooperated more when it would be their in-group and less when it would be an out-group, in minimal groups and in real ones.
• Size and culture. The effect ranges from small in some games to large in others, and the 20-country study found large country-level differences. They were linked to country measures of societal uncertainty, but two of the preregistered cultural predictors (individualism and religiosity) related to favoritism in the opposite direction to the one predicted.

Popular version. In-group bias is often equated with hostility toward outsiders. The research claim is narrower and, in a way, more unsettling: people favor their own group even when it was formed a few minutes ago by chance, and they mostly do it by being kinder to insiders, not crueler to outsiders.