Social Structure of Facebook Networks

We study the social structure of Facebook “friendship” networks at one hundred American colleges and universities at a single point in time, and we examine the roles of user attributes - gender, class year, major, high school, and residence - at these institutions. We investigate the influence of common attributes at the dyad level in terms of assortativity coefficients and regression models. We then examine larger-scale groupings by detecting communities algorithmically and comparing them to network partitions based on the user characteristics. We thereby compare the relative importances of different characteristics at different institutions, finding for example that common high school is more important to the social organization of large institutions and that the importance of common major varies significantly between institutions. Our calculations illustrate how microscopic and macroscopic perspectives give complementary insights on the social organization at universities and suggest future studies to investigate such phenomena further.

💡 Research Summary

This paper presents a comprehensive analysis of Facebook friendship networks from a single snapshot taken in September 2005, covering 100 American colleges and universities. For each institution the authors obtained the complete set of users (identified by .edu email addresses) and all undirected friendship ties among them, together with five self‑reported categorical attributes: gender, class year, major, high‑school, and residence (e.g., dormitory). The study treats each university as an independent network and further extracts four sub‑networks per institution: the largest connected component of the full network, the largest connected component of the student‑only network, and the largest connected components of the female‑only and male‑only networks.

The analysis proceeds on two complementary scales. At the dyad (pairwise) level the authors compute assortativity coefficients for each attribute using Newman’s formulation. Across the 100 schools, class year and gender exhibit the strongest positive assortativity (r values typically between 0.3 and 0.5 for class year, 0.2–0.4 for gender). High‑school assortativity is modest but systematically larger in larger institutions, indicating that alumni from the same high school tend to remain connected even when attending a big university. Major and residence show more heterogeneous patterns, with major assortativity being especially pronounced in smaller colleges.

To move beyond descriptive statistics, the authors fit two statistical models to the dyad data. First, a logistic regression assuming independent dyads includes an intercept (overall edge density) and four nodematch terms (same gender, class year, major, high‑school, residence). The regression reveals that sharing a class year yields the largest increase in log‑odds of a friendship (β≈1.2), while the effect of sharing a major or high‑school varies with institution size. Second, they estimate exponential random graph models (ERGMs) that augment the same nodematch terms with a triangle statistic to capture transitivity. The triangle coefficient is positive in virtually all cases, confirming the presence of clustering beyond attribute homophily. Due to computational limits, ERGMs are fitted only for the 16 smallest schools, but they illustrate how structural dependence can be incorporated.

At the macroscopic level the paper applies a suite of community‑detection algorithms (modularity maximization, spectral partitioning, hierarchical clustering, etc.) to each network. The resulting partitions are compared to the ground‑truth partitions defined by each attribute using normalized mutual information (NMI) and adjusted Rand index (ARI). Consistently, partitions based on class year achieve the highest similarity to algorithmic communities, while major‑based partitions align poorly, especially in large universities where major assortativity is weak. Gender partitions are most evident in the gender‑specific subnetworks, suggesting that gender homophily operates more strongly within single‑gender groups. Residence shows moderate alignment, reflecting the role of dormitory or house affiliation in shaping local clusters.

The authors synthesize these findings to argue that microscopic measures (assortativity, regression, ERGM) and macroscopic community structure provide complementary perspectives on social organization. A key insight is that high‑school commonality becomes a dominant organizing factor in large institutions, whereas shared major is a salient factor only in smaller colleges. This suggests practical interventions: leveraging high‑school alumni networks during orientation at large universities could accelerate social integration, while emphasizing major‑based study groups may be more effective at smaller schools.

Limitations are acknowledged. The data represent a single time point from 2005, predating many later Facebook features and before the platform opened to non‑students. Missing attribute values are treated as “missing” rather than imputed, and the assumption that online friendships faithfully mirror offline ties is discussed but not tested. Nevertheless, the breadth of the sample (100 institutions) and the combination of dyadic and community‑level analyses provide robust evidence that online social networks can reflect underlying offline social structures.

The paper concludes by outlining future directions: longitudinal studies to capture network evolution, multilayer analyses incorporating inter‑university ties, inclusion of additional demographic variables (race, socioeconomic status), and comparative work with newer social media platforms. Such extensions would deepen our understanding of how digital interaction patterns map onto, reinforce, or reshape real‑world social organization.