Network composition varies with social position

French work, notably from surveys on contacts and personal networks, finds broader and more diversified networks in the categories with the most education. As early as 1988, François Héran drew from this the idea that sociability is distributed like a cultural practice.

It also finds a different composition: the share of relationships from work and voluntary activities rises with social position, that of neighbourhood and family relationships falls in proportion. The gap therefore lies in the origin of ties as much as in their number.

The notion of social capital, which Pierre Bourdieu formalised in 1980 as the set of resources linked to possession of a durable network of relationships, gave this question a framework, without the means to measure it on a large scale.

Raj Chetty and a large team published in Nature two articles using the connection data of a very large social network in the United States, covering several tens of millions of people.

They build a measure of connection between different income groups and associate it, at territory level, with indicators of social mobility measured separately. The mobility used is that of children from low-income families, compared with the income they reach in adulthood.

The degree of connection between people of different incomes is, among the variables tested, the one most strongly associated with social mobility observed in a territory.

The authors decompose that connection into two terms: exposure, that is the composition of the places frequented, and friending bias, that is the propensity to connect within a mixed place. Both contribute, and two institutions of comparable composition can produce very different ties.

The analyses are ecological: they concern territories and not individuals followed over time, which risks attributing to an individual mechanism a regularity at aggregate level. The direction of the association also remains open, since more mobile territories may produce more ties across backgrounds as much as the reverse.

The ties measured are those declared on a platform, and social mobility is measured from another source, for different cohorts. The authors discuss those limits, which have not been lifted.

It brings attention back to the composition of places, schools, neighbourhoods, workplaces, associations, which determines possible encounters before any preference.

It is the same logic as work on proximity and induced homophily: networks record a structure of opportunities, and altering it means acting on places rather than on dispositions. The result says nothing, on the other hand, about the value that more homogeneous networks have for those who make them up, which this work does not measure.