Homophily, and what it does not prove
Our relationships resemble us, one of the most consistent findings in network sociology. Inferring a preference for the similar is a frequent error.
The finding
The term was proposed in 1954 by Paul Lazarsfeld and Robert Merton, who already distinguished similarity of status from similarity of values. Miller McPherson, Lynn Smith-Lovin and James Cook published in 2001 a synthesis surveying the regularity of the phenomenon across many dimensions: origin, education, age, occupation, convictions.
The phrase they used, that like attracts like, met with considerable success and has often been read as a psychological property, whereas their article insists on structures.
Two very different mechanisms
Choice homophily denotes a preference: at equal opportunity, people connect more with those who resemble them. Induced homophily denotes a supply effect: one does not meet a representative population, because schools, neighbourhoods and occupations are themselves socially composed.
Both produce exactly the same observation. Distinguishing them requires knowing the composition of the places frequented, information rarely collected in surveys, which ask about relationships and not about the opportunities they arose from.
A third mechanism complicates the reading further: influence. Friends may become alike because they spend time together, rather than spend time together because they were alike. Separating selection from influence with observational data is a methodological problem reputed to be very difficult.
What constrained settings show
Studies in situations where group composition is known and encounters are not chosen, notably in boarding schools or allocated student housing, allow the two mechanisms to be separated.
They generally find real choice homophily, but markedly weaker than the homophily observed in the general population, indicating that most of the phenomenon stems from the structure of opportunities. That result says nothing about what is desirable; it locates the phenomenon within social organisation.
The consequence for information
Homogeneous networks expose their members to homogeneous information. That mechanism is invoked in work on polarisation, where it long predates digital systems and already featured in the electoral surveys of the mid-twentieth century.
Analyses of platform data find strong homophily, often stronger than in offline networks, without the gap being attributable to the algorithm rather than to the composition of those who sign up.
Measured variation
The degree of homophily varies by dimension. It is very high for origin and education, weaker for political convictions in several European surveys, where some friends do not even know how the other votes.
It also varies with social position: several French studies find more diversified networks in intermediate categories than in the most advantaged or the most precarious ones. Similarity is therefore not a constant but a social variable in its own right.
Key points
Choice homophily and homophily induced by the supply of encounters produce the same observation.
Settings with known composition show a real preference, but far weaker than the raw observation.
The degree of homophily varies by dimension measured and by social position.
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