Dunbar's number, an extrapolation turned law
One hundred and fifty stable relationships: the figure has circulated everywhere since the 1990s. Its origin is a regression line drawn on primates.
Where the figure comes from
Robin Dunbar published in the late 1980s and early 1990s work relating, in primates, neocortex size to average social group size. He obtained a statistical relationship between the two, which he interpreted as the trace of a cognitive constraint.
The idea, later known as the social brain hypothesis, is that maintaining individualised relationships costs mental resources, and that those resources limit the number of partners an individual can keep track of. The hypothesis concerns the evolution of species, not individuals.
By placing the human value on that line, he obtained a prediction of about one hundred and fifty. The figure is therefore an extrapolation from a between-species regression, not a measurement made on humans, and it inherits all the uncertainties of that regression.
Historical checks
Dunbar then looked for confirmation in field data: village sizes in agrarian societies, military units, greeting card lists, address books. Several of those observations fall within the announced order of magnitude, which did much to spread the figure.
Those confirmations are selective and heterogeneous: they concern very different objects, assembled after the hypothesis was formulated, as critics have noted. A military company and a list of greeting cards do not measure the same thing, and nothing indicates how many counterexamples were set aside.
The 2021 reanalysis
Patrik Lindenfors, Andreas Wartel and Johan Lind published in 2021 a reanalysis of the relationship between neocortex and group size, using statistical methods that account for relatedness between species and so avoid treating closely related species as independent.
They conclude that the confidence interval around the human value is extremely wide, from a few dozen to several hundred, which leaves the single figure statistically meaningless. Their criticism does not deny that any limit exists; it disputes that it can be quantified in this way.
The reply and the state of the debate
Dunbar replied that the criticism bears on only one source of evidence and neglects behavioural data. The debate is unsettled, and it turns less on the existence of a cognitive limit than on the possibility of fixing its value.
Work on personal networks does find orders of magnitude, with considerable dispersion between individuals, which is compatible with a loose constraint and not with a fixed number. A network of forty people and a network of three hundred are both ordinary configurations.
What remains useful
The layered structure proposed by Dunbar and his colleagues, where the number of relationships grows in steps as intensity falls, is better supported empirically than the central value. The innermost layers contain few people, and it is their composition that varies most.
Contact data, notably telephone records, find a hierarchy of that shape in most people studied, without the exact thresholds reproducing from one study to another. What recurs is a shape, not a list of numbers.
Key points
The figure comes from an extrapolation off a primate regression, not from a human measurement.
The 2021 reanalysis concludes the confidence interval is too wide to retain a single value.
The layered structure of decreasing intensity is better supported than the central number itself.
This article describes documented mechanisms and disagreements. It recommends no position, no party and no vote, and commits to no reading other than that of the work it cites.
And you, where do you stand?
The Civix test situates your convictions on six dimensions and matches you with one of the twelve profiles, without ever telling you what to think.
Start my test