People who assume the strangers around them are less empathetic than they really are take fewer social risks and become more isolated — and one large field study found that simply correcting the assumption expanded social networks months later

A crowded room can become socially empty before anyone has said a word.

Not because the people in it are necessarily unfriendly. The problem can begin one step earlier, when each person is privately willing to listen, help or talk but publicly assumes everyone else is less open than they are.

A large Stanford study gave that mismatch a name: the empathy perception gap. Across two years of data, undergraduates tended to estimate that their peers were less caring than those peers described themselves as being. Students who held the colder view were less willing to start conversations, ask for support or reveal something vulnerable. They also grew more isolated over time.

The researchers then tried to change the forecast. They showed students what peers had actually said about their empathy and paired that information with prompts to take small social risks. Four months after the larger experiment, the intervention group reported 0.44 more close friends on average than the control group.

That is a meaningful field result. It is also more qualified than the neatest version of the story. The friendships were self-reported, the setting was one unusually selective university and the strongest intervention included behavioural nudges as well as corrected information. Understanding those details makes the finding more useful, not less.

The first rejection can happen entirely in the forecast

The study’s authors used “social risk” for actions with uncertain interpersonal consequences. Saying hello to someone unfamiliar, asking a classmate to meet for coffee, requesting help and telling a friend that something is wrong can all create warmth. They can also produce awkwardness, indifference or rejection.

If someone expects the average person to respond without much care, withdrawal can feel rational. Why disclose a struggle to a person who probably will not understand? Why begin a conversation that the other person probably does not want?

The difficulty is that avoidance protects the prediction from being tested. It removes the encounters that might reveal a warmer reality. A belief about other people can therefore help create the isolation it appears merely to describe.

The researchers described a loop: underestimate the community’s empathy, avoid social risks, miss opportunities for connection, feel more isolated, and then interpret that isolation as further evidence that the community is uncaring. Nothing in that loop requires anyone to be hostile. Shared caution is enough.

How 5,192 students made the hidden gap measurable

The study published in Nature Human Behaviour was led by Rui Pei, Samantha Grayson, Ruth Appel and colleagues from Stanford, Columbia, Princeton and the Santa Fe Institute. Its longitudinal portion drew on a Stanford undergraduate community of 5,192 people across two years.

Students estimated how empathic other Stanford students were and reported their own tendencies. They also answered questions about wellbeing, loneliness, social connection and everyday behaviour at repeated points. The main adjusted analyses included 4,970 participants and 14,437 observations.

One item makes the mismatch unusually concrete. Students estimated that 87 per cent of their peers would act kindly and help someone who was feeling bad. When students answered the same question about themselves, 96 per cent responded positively. The perceived community was nine percentage points colder than the aggregated self-description of its members.

That comparison has a limit. Reporting a willingness to help is not the same as being observed helping when the moment arrives. Social desirability may make people describe themselves generously. The study established a gap between perceptions of others and others’ self-reports, not a nine-point audit of behaviour.

Even so, the pattern was not confined to one item. Students who perceived peers as more empathic reported better current wellbeing and social connectedness. Empathy perception also predicted slightly higher wellbeing and connection at the next survey point after the earlier level of each outcome was taken into account.

The effects were modest. In the extended analyses, the standardized cross-sectional coefficients were 0.02 for wellbeing and 0.08 for social connectedness. The longitudinal design strengthens the timing of the relationship, but it cannot prove that pessimistic empathy beliefs alone caused isolation. Personality, mood or prior treatment by others may shape both.

The first field experiment changed the story students heard

Observation identified the loop. The first field experiment tested whether it could be interrupted.

It involved 346 students. Posters placed around residential halls presented real community data, including a message that 95 per cent of Stanford students were likely to help someone who was feeling down. A class workshop lasting roughly an hour asked students to estimate their peers’ empathy, compare those expectations with peers’ answers, discuss common social miscalculations and plan a small action.

The examples were ordinary by design: greet someone working in the dining hall, invite a classmate to study, ask a dormmate to eat, tell a friend about a difficult moment, or ask someone to introduce a person they did not yet know.

Students in the control condition underestimated peer empathy by 0.4 points on a seven-point scale. Those exposed to the intervention still underestimated it, but by 0.1 points. Expressed as a proportion of the original gap, that was a 75 per cent reduction.

The intervention group also reported taking social risks 11 per cent more often. The result is encouraging, although it did not isolate a bare fact on a poster. The session included discussion, reflection, examples and action planning. It changed both the available information and what students were invited to do with it.

The second experiment moved from posters to phones

The larger follow-up field experiment included 419 students. It separated participants into a control condition and three active conditions: information about campus norms, behavioural nudges delivered to smartphones, or both.

The messages suggested manageable acts, such as complimenting a stranger or catching up with someone the student had not spoken with recently. The nudges ran for two weeks. Under the preregistered primary approach, the researchers pooled the three active conditions to gain statistical power and compared them with the control group.

During the week after the intervention, students in the active conditions were 89 per cent more likely than controls to report a social-risk action. The campaign also narrowed the empathy perception gap, though less dramatically than the first experiment: controls underestimated peer empathy by 1.0 point on the seven-point scale, compared with 0.9 points in the pooled intervention group.

Two months later, intervention participants were about twice as likely to register for a research event built around an extended conversation with strangers. Registration is a behavioural outcome rather than a survey attitude, but it still measures willingness to attend, not whether a lasting relationship followed.

The longer result came four months after the intervention. Students in the active conditions reported an average of 0.44 more close friends than students in the control condition. The figures and intervention details are also laid out in the researchers’ chapter of the World Happiness Report 2025, while the study’s data and analysis code are available through its public OSF repository.

What “0.44 more close friends” does and does not mean

An average increase of 0.44 does not mean almost half of a friend was added to every network. It means the distribution of reported close-friend counts shifted upward in the intervention group. Some students may have added none, some one and some more.

The measure was also self-reported. The research did not require every named friend to confirm the relationship, establish that each tie was equally close or show which friendship began because of which message. This was a real-world strength and a measurement limitation at the same time.

The component results add a more important qualification. Compared with controls, the combined norms-and-nudges group reported 0.64 more close friends in the winter follow-up, a statistically significant difference. The nudge-only estimate was 0.47, narrowly missing the conventional significance threshold. The norms-only condition did not differ significantly from control.

In other words, the pooled experiment supports the intervention package. It does not cleanly establish that correcting the empathy estimate by itself created lasting friendships. The action prompts may have been essential, or the separate groups may simply have been too small to detect subtler effects. The study cannot decide between those explanations.

Nor did an extra fraction of a reported friend amount to a demonstrated cure for loneliness or mental illness. The disaggregated analyses found no significant wellbeing difference in the spring. This paper tested one possible barrier to connection, not a complete theory of why young adults feel lonely.

Why a campus can look colder than its students are

People have privileged access to their own intentions. Someone knows that they would probably listen if approached, enjoy an unexpected conversation or welcome an invitation. From everyone else, they mostly see surfaces: headphones, a hurried walk, a quiet table, a delayed reply.

Visible reserve can then be mistaken for private indifference. If many people make the same error, each person’s caution becomes evidence for everyone else’s. Social psychologists sometimes describe related patterns as pluralistic ignorance: individuals privately reject a norm but infer from public behaviour that others accept it.

The Stanford findings also sit beside a longer record of pessimistic social forecasting. In the classic “mistakenly seeking solitude” experiments, commuters expected talking with strangers to be less pleasant than keeping to themselves. Those instructed to connect reported a more positive journey, while the person approached did not pay an experiential cost.

Earlier Silicon Canals coverage of deep-conversation experiments found a similar forecasting error: people expected more awkwardness and less interest than strangers actually reported. And research on giving compliments found that givers overestimated how uncomfortable recipients would feel.

None of those studies says every approach will be welcomed. Together, they suggest that the average forecast often carries a pessimistic bias precisely where a small first move could produce better information.

Calibration is not blind trust

A corrected average is not a promise about the next stranger.

The Stanford results do not instruct people to disclose indiscriminately, overlook power differences, ignore personal safety or keep approaching someone who has signalled disinterest. The participants were undergraduates in one well-resourced American university. The accurate norm for that campus may not be the accurate norm for another community.

There is also a difference between empathy and compatibility. A person can respond kindly without becoming a friend. An invitation can be declined for reasons that have nothing to do with the inviter, and larger networks are not automatically better networks. Solitude chosen freely is not the same as isolation produced by fear.

The practical shift is smaller and more defensible. When sound local evidence shows that people are more receptive than the community assumes, making that evidence visible can reduce an imaginary part of the cost. Pairing it with one manageable action gives reality a chance to answer.

That might be why the phone nudges matter so much. Information can revise a belief while leaving behaviour untouched. A prompt arriving in daily life can turn the revision into a greeting, invitation or honest sentence, each of which generates evidence that another poster cannot.

The study’s most interesting claim is not that a campaign manufactures friendship. It is that some social distance begins as a shared measurement error. Correct the forecast, make the first move small, and a few relationships that seemed unavailable may finally have room to start.