Nobody teaches you how adult friendship operates. School and university hand it to you free, bundled with timetables and shared corridors, and then one day the infrastructure is removed and you are expected to keep producing the same result by will alone. It goes badly for a lot of people, and the reasons turn out to be structural rather than personal.
The famous number, and the argument surrounding it
Any conversation about this eventually reaches 150. The figure comes from the anthropologist Robin Dunbar, whose social brain hypothesis proposed that primate group size tracks the relative size of the neocortex, and who extrapolated from the primate data to a human figure of roughly that magnitude — the number of people with whom one might maintain genuine, reciprocal, who-owes-whom-a-drink relationships.
It became one of the most quoted numbers in popular social science, and it is under sustained attack. A 2021 paper in Biology Letters, 'Dunbar's number' deconstructed, reran the analysis with updated datasets and modern statistical approaches. Different defensible methods produced averages ranging from roughly 16 to 42 in one case and 69 to 109 in another, and the confidence intervals were enormous, spanning from single figures to several hundred. The authors' conclusion was blunt: you cannot derive a cognitive ceiling on human group size this way, and naming any specific number is not supportable from that evidence.
Dunbar and colleagues have contested that reading, and the argument is genuinely unresolved. What survives the fight is not the precise integer but the qualitative claim that everybody can check against their own life: attention is finite, maintenance costs time, and the number of people you can properly keep up with is smaller than the number of people you know.
Circles rather than a single list
The more useful part of that research tradition was never the headline figure anyway. It was the observation that personal networks appear to nest in layers of roughly increasing size and decreasing intensity — a handful of people in the innermost ring, a slightly larger group beyond that, then wider bands of friends and then acquaintances. Each layer costs less per person and returns less.
A 2020 paper by Dunbar in Proceedings of the Royal Society A added a detail worth sitting with: family and friends occupy the layers differently. The innermost circles tend to be split fairly evenly between the two, friends predominate in the middle band, and relatives make up more of the outer reaches. Kin persist without maintenance because the relationship exists whether or not anyone rings. Friendship has no such guarantee, which is precisely why it decays when the ringing stops.
Friendship runs on accumulated hours
The most quotable finding in modern friendship research came from Jeffrey Hall, whose study in the Journal of Social and Personal Relationships tried to put a figure on the cost. Working with people who had recently relocated and with first-year students, he estimated that moving someone from acquaintance to casual friend took something in the region of forty to sixty hours of time together in the early weeks, with the transition to close friendship requiring considerably more and best-friend status running past two hundred hours.
Treat those as orders of magnitude rather than a schedule. The estimates rest on participants recalling how much time they had spent, which is not a precise instrument. But two secondary findings are more robust and more useful. First, the type of time matters enormously: hours spent on leisure and everyday, unstructured talk did the work, while hours spent side by side on obligatory tasks did comparatively little. Second, the total required is large in a way that most adults have not consciously priced in. A monthly coffee, sustained faithfully, buys you about twelve hours a year.
Proximity does more of the lifting than character
People tend to explain their friendships in terms of compatibility, which flatters everyone involved and gets the causation the wrong way round. The classic evidence here is a study of a postwar student housing complex, where researchers found that who became friends with whom was predicted with embarrassing accuracy by architectural accident — distance between front doors, which way a staircase faced, whose flat sat beside the letterboxes. People did not become friends because they matched. They became friends because the building kept putting them in front of one another, and matching was constructed afterwards.
Researchers who study adult friendship formation tend to describe a similar recipe: physical proximity, repeated and unplanned interaction, and a setting relaxed enough that people eventually lower their guard. Note that the middle ingredient is the one adult life destroys. Childhood and university supply unplanned repetition by default. Adult scheduling replaces it with planned meetings, which are rarer, require organisation, get cancelled, and never produce the incidental hours that friendship actually consumes.
How friendships end, which is usually by not happening
Dramatic ruptures are memorable and rare. The ordinary ending is attrition: someone changes job, moves across a city, has a child, starts a shift pattern, and the shared setting that had been silently generating contact disappears. Neither party decides anything. The intervals lengthen, the effort required to close the gap grows in proportion to the gap, and at some point the relationship converts from friendship into fondness at a distance.
It helps to distinguish that from the acquaintance network, which is a different asset with a different function. The wide outer layer of people you know slightly is where information, opportunity and novelty travel, and it can be maintained at very low cost. Confiding, reciprocity and reliable turning-up belong to the small inner circles. Confusing the two produces a familiar modern complaint: a person with hundreds of contacts, several thousand followers and nobody they could ring on a bad Tuesday night.
Reading the loneliness evidence carefully
Which brings us to the health claims, where accuracy matters most because the numbers get quoted loosely. The foundational work is a meta-analytic review led by Julianne Holt-Lunstad, which pooled data across many studies and reported increased likelihood of mortality associated with social isolation, loneliness and living alone, each in the region of a quarter to a third above baseline. In 2023 the US Surgeon General issued an advisory on social connection that drew heavily on this body of work and framed it as a public health priority.
Three cautions. The widely repeated comparison to smoking a certain number of cigarettes a day is a rhetorical translation of statistical effect sizes, not a measured equivalence, and several researchers close to the underlying data have asked people to stop deploying it that way. The findings are also observational, so causation runs in both directions: isolation may damage health, and poor health reliably isolates people. And the framing of a sharp modern epidemic is contested, since some long-run survey analyses find loneliness in the general population rather more stable across the decades than the headlines imply.
None of that undoes the sensible conclusion. Connection appears to matter for how long and how well people live, and it is built from hours, proximity and repetition rather than from sentiment. The practical implication is unromantic: if you want friends in adulthood, the reliable move is to manufacture the conditions — a regular fixture, a shared activity, a place you turn up to whether or not you feel like it — and let the friendships assemble themselves in the background, which is the only way they have ever been made.
