How Many Friends You Can Actually Keep
Dunbar's 150 cannot be derived the way it was derived. The layered structure underneath it was then confirmed on 185,000 people, and the outermost layer is about 150. Both of those are true and almost nobody quoting the number knows either.
Key takeaways
- The 150 comes from extrapolating a regression between neocortex size and group size in primates, and Dunbar's fuller argument is that language replaced grooming as the way to service a group that large.
- Lindenfors, Wartel and Lind reran that extrapolation. Different accepted methods gave averages of 69 to 109 and 16 to 42, with intervals running from 4 to 520 and 2 to 336.
- Their conclusion is that specifying any one number is futile and that a cognitive limit on human group size cannot be derived this way.
- The layered structure was then confirmed on more than 185,000 people across Facebook and Twitter, with layer sizes and contact frequencies closely matching offline networks.
- Which means the number turns up again by a completely different route, and has better evidence behind it than the argument that made it famous.
You have a number in your head and it is almost certainly 150. Our position is that the number is not defensible and the thing it was invented to describe very much is, which makes this one of the more interesting arguments in social science and not one of the more embarrassing ones.
The number comes from Robin Dunbar, and the reasoning behind it is better than its reputation. In the early nineties he observed that across non-human primates, group size covaries with the relative size of the neocortex. Bigger social brain, bigger troop. Primates hold those troops together by grooming each other, and grooming time rises in a straight line with group size, which sets a ceiling: past a certain troop size an animal would have to spend an intolerable proportion of its day picking through other animals' fur.
Extend the regression line to humans and it predicts a group size close to what anthropologists find in hunter-gatherer and traditional horticultural societies. Dunbar's further argument is the part people forget. Humans cannot groom a group that size either, so language, he proposed, is the substitute: a way of servicing many relationships at once. The supporting observation is that roughly sixty per cent of human conversation time goes on talking about relationships and personal experiences, an unflattering description of your week and probably an accurate one.
That is a real theory. It makes predictions, it connects two things nobody had connected, and it explains why the small talk you consider a waste of time is the actual work.
The trouble is the number. In 2021, Patrik Lindenfors, Andreas Wartel and Johan Lind asked a narrow question in Biology Letters: is there statistical support for extrapolating that regression to a human figure at all. They reran the analysis on complementary datasets using different accepted methods, and the answers did not converge. Bayesian methods put average group size somewhere between sixty-nine and 109. Phylogenetic generalised least squares put it between sixteen and forty-two.