Most friendship surveys stop at one question: who are your friends? A team at MIT and Tel Aviv University added a second one, and the second question is what makes the study land.
They gave eighty-four people a roster of everyone else in their course and asked them to score each name, from “I do not know this person” up to “one of my best friends.” Then they asked each person to go back through the same roster and predict the score they would get in return.
Ask people to guess how they are scored back
The scale ran from zero to five. Zero meant a stranger, three meant “friend,” five meant one of your best. Anything above a two counted as naming someone a friend, which is a detail with weight in it: friendship here is a threshold on a number line, not a word anybody volunteered.
The prediction column came back tracking the first one almost perfectly. The correlation between what a person gave and what they expected to receive was 0.95, which in this kind of research is about as close to a straight line as human answers get. Almost every nomination arrived with the assumption that it was being returned — 94 percent of them, in the paper’s count.
That reads less like vanity than like the ordinary arithmetic of a friendship. Nobody runs a check on whether the person they would drive to the airport would drive them. The assumption holds up the structure without ever being tested.
All of it comes from one cohort: eighty-four students taking a four-month applied-management course at a Middle Eastern university, aged twenty-three to thirty-eight, ninety-six percent of them in their third year. The two headline percentages also sit on different bases, one counting nominations and the other counting ties in the network, and the paper never reconciles the totals, so each is quoted here to its own denominator rather than folded into a single fraction.
The other side of every tie
From those nominations the researchers built a directed network: eighty-four people, 775 friendship ties. A tie counted as reciprocal when both people had named each other, unilateral when only one had.
Fifty-three percent were reciprocal. Across those ties, the correlation between the two sides’ actual scores was 0.36 — measured only on the friendships, where the 0.95 was measured across every name on the roster.
The obvious objection is that a handful of over-nominators could drag the average down. The authors checked. Calculated per person, the share of someone’s friendships that were mutual formed a normal distribution centred around a half. This was distributed across the room rather than concentrated in a few unusually eager people.
They counted reciprocity the same way in five other friendship networks, from American, European and Middle Eastern universities, and found rates between 34 and 53 percent. An earlier study of adolescents had put the figure at 64 percent, noticeably higher, and the authors attribute part of that to design: those teenagers could nominate at most ten friends, five male and five female, while this cohort was asked about every single person in the room, and complete nomination networks turn up more one-way ties. That paper, Vaquera and Kao in Social Science Research, is the reason the number is better read as a range than a fact.
Something else is missing from all of it, and it matters more than the range. The paper reports nothing about loneliness, distress, or how any of these people felt, and the instrument it publishes is a scoring form: four questions per name about closeness and who knows whom. A tie in this data is a score above two on a survey form, not a friendship as anyone lives one, and no part of this research can tell you that an unreturned friendship hurts.
The paper was really about persuasion
None of this was designed as an essay about who likes whom. It was designed to work out what determines whether a buddy system does anything — the arrangement used for smoking cessation, weight loss and diabetes management, where you are paired with someone whose job is to keep you at it.
To test it the team went to a different population entirely: the MIT Friends and Family cohort, not the eighty-four students, and a fitness experiment run inside it in late 2010. Of the hundred-odd active subjects, most signed up and were split into three conditions; in two of them, participants were shown the progress of a pair of “buddies.”
Across those two groups, seventy-six people in all, the direction of a friendship predicted how much someone’s activity changed. Reciprocal ties moved people. So did ties where the buddy considered the participant a friend and the participant did not return it. Ties running the other way, where the participant had named the buddy and the buddy had not named them, produced no statistically significant effect in that regression. Tie strength mattered less than direction did.
In the Friends and Family network — again, not the course survey — one-way ties tended to run from the person with fewer connections toward the person with more, a pattern that held after controlling for demographics. The authors do offer an explanation, and offer it as a suggestion: that many unreciprocated friendships are aspirational, people reaching upward through a social hierarchy. What the paper measured was the direction of the ties. It never measured the motive, and the two belong apart.
No clinician has been anywhere near this article, and the survey was a network-science instrument rather than a clinical one; what the researchers measured was scoring behaviour, and what this piece can do is describe it. Persistent loneliness, or real distress about friendships, is something a therapist or counsellor can work on.
So a paper about exercise apps ends up holding an ungainly fact. Of the three kinds of tie the study could see, the buddy least likely to move a participant was the one the participant had appointed a friend without being appointed back. The study had no way to ask how a person would know the difference. Most days, how would any of us?
