How many people do we really know

The question about people with whom one discusses important matters produces very low answers, often a median between two and three people in American and European surveys.

That figure surprises because it is read as a count of friends. It actually measures a very narrow circle, defined by a demanding criterion, and is not meant to describe the whole set of relationships. Nor does it say anything about what such a circle means to those in it: two confidants is an ordinary number.

Measures based on actual contact, by diary or call record, produce orders of magnitude of a few dozen people over a year. Diaries require every exchange to be noted for several weeks, demanding work but less dependent on memory.

That intermediate level is the one that best matches ordinary use of the word relationship, and it is also the one the layers proposed by Dunbar describe. The study of Christmas cards conducted by Russell Hill and Robin Dunbar in 2003 provided an approximation of it.

Scale-up methods estimate the number of people one could recognise and name, from the frequency of rare first names in the declared network. Developed notably by Peter Killworth and Russell Bernard, they are also used to estimate the size of populations that are hard to count.

The estimates obtained run to hundreds or even thousands. They measure a capacity for recognition, not a relationship, which explains the gap with other methods.

An article announcing a collapse in the number of friends and another announcing vast networks may both be accurate, if they use different instruments.

The useful question to ask of a figure is therefore which generator was used, information almost always available in the original article and almost never in its coverage. Next come the reference period, the population surveyed and the mode of administration, each of which weighs on the result.

Digital platforms have added a fourth measure, the number of contacts displayed, which corresponds to none of the previous three. It lumps together confidants, colleagues long lost from view and strangers, and comparing it with surveys makes little sense.

Every method finds considerable dispersion between individuals, greater than the average gaps between countries or periods.

That finding limits the reach of national averages: a single figure describes poorly a distribution whose extremes are far apart, and it leads to commenting on gaps smaller than ordinary variation. A narrow circle and a wide one both fall within that variation, and neither constitutes the norm.