Introduction
In network science, the principle of "assortative mixing" describes the tendency of similar nodes to connect. In social contexts, this often manifests as degree assortativity ($r$): popular individuals (high-degree nodes) tend to associate with other popular individuals. While first-order assortativity is a well-documented feature of human social structures, this research identifies a deeper structural property: second-order assortative mixing.
Defining Second-Order Mixing
The authors propose that the similarity between connected individuals extends into their respective neighborhoods. Specifically, if two people are friends, is there a correlation between the "most prominent" or "average" status of their respective acquaintances?
To quantify this, the study introduces two new coefficients:
- $R_{max}$: The Pearson correlation between the maximum degree of the neighbors of two connected nodes.
- $R_{avg}$: The Pearson correlation between the average degree of the neighbors of two connected nodes.
These metrics use "excess degree" (degree minus one) and exclude the mutual link between the nodes to ensure the measurement captures neighborhood properties rather than the direct link itself.
Empirical Results across Diverse Networks
The researchers analyzed eleven networks, ranging from film actor collaborations to biological protein interactions and synthetic models. The results revealed a striking pattern: second-order mixing is a unique signature of social networks.
As shown in the table above, social networks (a-e) consistently exhibit high positive values for $R_{avg}$ and $R_{max}$, often significantly exceeding the standard first-order coefficient $r$. For instance, in the Scientist collaboration network, while $r$ is a modest 0.161, $R_{avg}$ reaches 0.680. Conversely, technological and biological networks (f-i) show near-zero or even negative correlations, suggesting that second-order mixing is not a universal property of all complex networks but a specific characteristic of human social organization.
Key Insights
- Hidden Parity: Some networks, like Jazz Musicians, show negligible first-order assortativity ($r \approx 0.02$) but high second-order mixing ($R_{avg} \approx 0.54$). This suggests that even when individuals of different status connect, they often share a similar "social reach" through their best contacts.
- Robustness: The effect persists even after removing the most connected "hubs," proving it is a pervasive structural feature rather than one driven by a few celebrities.
- Independence: Through null hypothesis testing and random rewiring, the authors demonstrated that second-order mixing cannot be explained by standard properties like degree distribution or clustering coefficients. It is an independent topological property.
Conclusion
Second-order assortative mixing provides a new lens through which to distinguish social networks from other types of complex systems. While the exact behavioral mechanisms—such as social preference based on connection quality—remain to be fully explored, this property serves as a powerful discriminator for identifying human-driven social structures. Future modeling must account for this "two-hop" correlation to accurately simulate realistic social dynamics.
Summary updated