Does similarity shape who becomes friends?

Similarity matters in friendship. But it enters the process before two people choose each other.

A school, workplace, neighborhood or activity first shapes who is likely to meet. Within that field, people often continue contact with someone who resembles them on something important. Some friendships then become more alike, while others end when differences become costly.

The newest neuroscience adds one more layer. In 2025, researchers showed that some similarities in how 41 incoming students’ brains responded to the same films existed before most of them met and predicted specific differences in their friendship network eight months later.

It did not show that a scan could tell who would become friends.

Similarity starts with who is available to meet

Researchers call the clustering of similar people homophily.

The word can sound like a preference: people seek others like themselves. That is only half the mechanism.

Families, schools, organizations, neighborhoods and occupations create opportunity sets. A person at a running club meets more runners. Students in one program share educational filters and a calendar. Colleagues encounter people selected by the same employer.

The classic sociology review by Miller McPherson, Lynn Smith-Lovin and James Cook separates this structured opportunity from the additional similarity created when people choose ties inside it. Friendship cannot select across people who never encounter one another.

That distinction prevents a common mistake. Similar friends do not prove that two people detected a deep match. Their environment may have made both the similarity and the meeting more likely.

Across 1,523 pairs, similarity appeared early

Angela Bahns and colleagues studied 1,523 interacting pairs across 11 independent field samples. They compared personality traits, attitudes, values, recreation and alcohol or drug use.

Within-pair similarity was statistically significant in 178 of 206 variable-by-sample hypothesis tests across 81 unique variables: 86% of those tests.

That number is easy to misuse. It does not mean that friends were 86% alike. It means that, for 86% of those tests, members of the observed pairs were more similar than random pairings from the same samples.

The timing was more revealing than the total.

Similarity did not rise with relationship length, closeness or how much the pair discussed a topic. Intimacy produced only modest effects. What mattered more was whether both people considered an attitude important.

Two follow-up studies asked students in large lecture classes to choose a classmate they had never spoken to for a first meeting. Those newly acquainted dyads were generally no more similar than chance. Only 23% interacted again. The pairs who did continue contact showed significant similarity later.

The researchers called this niche construction: where people have choices, they build a social environment compatible with their traits and values. The evidence is consistent with early selection. It does not prove that similarity was the only reason a second interaction happened.

Friends do not converge on everything

Selection is not the whole story. Friends can influence one another after a tie forms.

The problem is that a snapshot of two similar friends cannot tell which process occurred. They may have started alike, become alike or experienced the same environment.

Kevin Lewis, Marco Gonzalez and Jason Kaufman followed one college cohort’s Facebook ties and stated tastes over four years. Students sharing some music and movie tastes—but not book tastes—were more likely to become friends.

Evidence of later taste convergence was much narrower. It appeared for one of 15 taste clusters: classical and jazz music.

Other longitudinal research finds influence for particular behaviors and peer norms. The useful conclusion is not that selection always wins. It is that similarity is not one exposure. Political attitudes, exercise, alcohol use, humor and music can move through different social mechanisms.

A new study measured neural similarity before friendship settled

Yixuan Lisa Shen and colleagues followed an incoming MBA cohort of 288 students.

Forty-one participated in the final fMRI analysis. Most were scanned soon after arriving: the median delay was three days and the most common delay was one. Only 18 of the 820 possible pairs reported any earlier interaction, and the researchers repeated key analyses without them.

Inside the scanner, every participant watched the same 14 film clips. The researchers compared how activity changed moment by moment within 214 brain regions for every pair.

They then mapped reciprocal friendships across the full cohort after two months and again after eight months. A tie counted only when both students named each other.

The design improves on studies that scan people after they are already friends. Similarity preceded most of the social history being measured. It still remained an observational study in one unusually dense graduate program.

The headline comparison found no difference

At month eight, the 820 scanned-participant pairs included 93 direct-friend pairs, 544 pairs separated by two friendship steps and 183 separated by three.

Future friends were not more neurally similar than both non-friend groups combined.

They also did not differ from friends-of-friends alone.

The significant result appeared in a narrower contrast. Direct friends had shown greater pre-existing similarity than pairs three steps apart in part of the left orbitofrontal cortex, a region involved in subjective value. The effect remained when pairs with previous interaction were excluded.

Then the researchers accounted for similarity in age, gender, nationality, hometown, undergraduate background and prior industry. The orbitofrontal result no longer survived correction for testing 214 regions. Gender similarity made a measurable contribution to that reduction.

The scan may have captured alignment in taste, attention or interpretation. It did not reveal a demographic-free compatibility signal.

Growing closer was different from not growing closer

The researchers also tracked whether each pair moved closer or farther apart in the network between months two and eight.

Of the 820 pairs, 279 moved closer, 445 did not change and 96 moved farther apart.

Again, the broad comparison was null. Pairs who moved closer were not more neurally similar than unchanged and farther-apart pairs combined. They also did not differ from unchanged pairs alone.

But pairs who moved closer did differ from pairs who moved farther apart. Their pre-existing responses were more similar across 40 cortical regions, the bilateral thalamus and left amygdala. The pattern implicated systems involved in attention, interpretation and emotional response.

This was the study’s strongest result—and its most specific.

Network distance can change because friendships elsewhere form or dissolve. The comparison does not say that neural alignment caused a direct tie. It says that response similarity measured near the start of one community preceded opposite social trajectories inside it.

The genetic-similarity claim remains disputed

One genetic study shows why opportunity must stay in the explanation.

Benjamin Domingue and colleagues analyzed an ancestry-restricted Add Health sample of about 5,500 adolescents. They reported that friends were modestly more genetically similar than randomly paired unrelated students. Comparing friends with random pairs from the same school reduced the estimate by about half, which the authors interpreted as evidence for school-level social structure.

That was an observational comparison, not an experiment on school assignment. A 2020 reappraisal argued that residual population stratification could account for the apparent friend similarity and that the original claim was not justified by the data.

This is not evidence for a friendship gene.

Institutions sort people by place, family resources, prior achievement and other correlated conditions. Population structure can also remain inside an ancestry-restricted analysis. Both can produce apparent correlation before one student chooses another.

The same logic applies far beyond genetics. If two friends share an interest, ask whether the interest created the meeting, the meeting strengthened the interest or both occurred inside the same setting.

What the evidence says at each stage

Original sources

• McPherson, Smith-Lovin and Cook, 2001
• Bahns et al., 2017
• Lewis, Gonzalez and Kaufman, 2012
• Laursen, 2017
• Domingue et al., 2018
• Yengo et al., 2020 reappraisal
• Shen et al., 2025

Independent editorial summary. The authors, journals, universities and platforms are not affiliated with LifeScore.