The perceived assortativity of social networks: Methodological problems and solutions

Networks describe a range of social, biological and technical phenomena. An important property of a network is its degree correlation or assortativity, describing how nodes in the network associate based on their number of connections. Social networks are typically thought to be distinct from other networks in being assortative (possessing positive degree correlations); well-connected individuals associate with other well-connected individuals, and poorly-connected individuals associate with each other. We review the evidence for this in the literature and find that, while social networks are more assortative than non-social networks, only when they are built using group-based methods do they tend to be positively assortative. Non-social networks tend to be disassortative. We go on to show that connecting individuals due to shared membership of a group, a commonly used method, biases towards assortativity unless a large enough number of censuses of the network are taken. We present a number of solutions to overcoming this bias by drawing on advances in sociological and biological fields. Adoption of these methods across all fields can greatly enhance our understanding of social networks and networks in general.

💡 Research Summary

The paper revisits the widely held belief that social networks are inherently assortative—that is, high‑degree nodes tend to connect with other high‑degree nodes—while most other types of networks are disassortative. The authors begin by surveying the literature and find that the majority of studies reporting positive assortativity in social systems rely on “group‑based” data collection: individuals are linked whenever they share membership in a common group (e.g., a class, a workplace, a club). In practice this method treats each group as a complete subgraph (clique), which artificially inflates the number of edges among high‑degree nodes and therefore biases the assortativity estimate upward.

To quantify this bias, the authors construct synthetic social networks and simulate the process of taking repeated “censuses” of the network under the group‑based paradigm. When the number of censuses is small, the observed assortativity is strongly positive, reflecting the over‑representation of intra‑group ties. As the number of censuses increases, inter‑group connections begin to appear, and the measured assortativity converges toward zero or even becomes negative, indicating that the apparent positivity was a methodological artifact rather than an intrinsic property of the network.

The paper contrasts these findings with a range of non‑social networks (technological, biological, ecological) that are typically built from direct measurements of physical or functional links. Because such data are not generated by imposing cliques, these networks consistently display neutral or negative assortativity, reinforcing the notion that the methodological differences, not the underlying system type, drive the observed disparity.

Recognizing the problem, the authors propose several concrete solutions. First, they advocate for “individual‑based” data collection, such as sociometric surveys, wearable sensors, or digital trace data, which capture actual dyadic interactions rather than inferred group co‑membership. Second, they recommend longitudinal sampling—repeated observations over time—to capture the dynamic formation and dissolution of ties, thereby reducing reliance on a single static snapshot. Third, they suggest employing rigorous statistical controls, including null‑model randomizations and bootstrapping, to test whether the observed assortativity exceeds what would be expected by chance given the group structure. Fourth, when group‑based data must be used, they advise applying weighting schemes that adjust for group size and the degree distribution of members, effectively correcting the over‑connection bias introduced by treating groups as cliques.

In the discussion, the authors argue that these methodological refinements have implications far beyond social network analysis. Accurate assortativity estimates are crucial for models of information diffusion, epidemic spread, and innovation adoption, all of which depend sensitively on the pattern of high‑degree node connections. Moreover, the paper calls for interdisciplinary awareness: researchers in biology, physics, and computer science should also scrutinize how their data‑collection protocols might inadvertently impose structural biases.

The conclusion reiterates that social networks are not fundamentally more assortative than other networks; rather, the perceived difference stems from the prevalent use of group‑based sampling. By adopting individual‑based measurements, increasing the number of censuses, and applying appropriate statistical corrections, scholars can obtain unbiased assortativity values. Implementing these practices will enhance the reliability of network science across domains and lead to a more accurate understanding of how social structures influence collective phenomena.