Social Network Analysis
- Introduced by
- Moreno, J.L.; formalized by Wasserman & Faust
- Year
- 1934 (sociometry); 1994 (modern formalization)
- Type
- Structural/relational analysis framework
- Family
- Network science
What does it do?
Social Network Analysis (SNA) is a structural method that maps and measures relationships and flows between people, groups, organizations, or other entities modeled as nodes connected by ties (edges).
Rather than focusing on individual attributes, SNA reveals how the pattern of connections shapes behavior, influence, information flow, and outcomes within a system.
The intuition behind the method
Traditional research treats units (people, organizations) as independent cases. SNA treats the connections between them as the primary unit of analysis. Who you are linked to — and how centrally you sit in the web of relationships — shapes what information you receive, how much influence you have, and whether groups form around you. By drawing a map of a network, researchers can see invisible structures: brokers bridging disconnected clusters, influential hubs, isolated peripheries, and tight-knit communities that act as echo chambers.
History
Social network analysis traces to Jacob Moreno's sociometry in the 1930s and Georg Simmel's relational sociology. Graph-theoretic foundations were developed in the 1950s–1970s by Bavelas, Harary, and colleagues. The field was consolidated by Wasserman and Faust's landmark 1994 textbook and the founding of the journal Social Networks. The rise of online platforms and digital trace data after 2000 triggered explosive growth, making SNA a core method in computational social science.
When to use it
Use SNA when the research question is fundamentally relational — about how connections, flows, or structural positions shape outcomes — and when relational data (ties between actors) can be collected or reconstructed. It is well-suited to studying information diffusion, collaboration patterns, influence hierarchies, organizational structures, citation networks, and social cohesion. Do not use SNA when you have only attribute data (no tie data), when the network boundary cannot be reasonably defined, when the sample is too sparse to yield meaningful metrics, or when the research question is about individual-level attributes rather than relational structure.
Limitations and considerations
- Network boundary specification is consequential and often theoretically ambiguous.
- Relational data collection is burdensome; survey-based name generators suffer recall bias and non-response.
- Many SNA metrics assume complete network data; missing ties distort centrality estimates.
- Standard SNA metrics treat nodes as independent, violating the independence assumption required by classical statistical tests.
- Large networks require computational resources and specialized software (e.g., igraph, NetworkX, Gephi).
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Captures structural phenomena — brokerage, cohesion, centrality — invisible to attribute-based methods.
- Applicable to diverse domains: social, organizational, bibliometric, biological, and digital networks.
- Rich suite of metrics and models allows both descriptive and inferential analysis.
- Integrates naturally with visualization tools that communicate complex structure intuitively.
- Applicable to both small (survey-based) and very large (computational) networks.
- Compatible with longitudinal and multilayer extensions for dynamic or multiplex networks.
Applications
- Mapping organizational communication and knowledge-sharing networks.
- Identifying influential spreaders and bottlenecks in information diffusion studies.
- Bibliometric and co-authorship network analysis in scientometrics.
- Community detection in online social platforms and digital trace data.
- Epidemiological contact tracing and disease spread modeling.
- Supply chain and inter-organizational dependency mapping.
How it works
How it works
The researcher specifies the boundary of the network (which actors are included) and the type of tie to study — friendship, collaboration, citation, advice-seeking, information exchange. This conceptual decision is substantive: different tie definitions produce structurally different networks even with the same actors.
Relational data are gathered through surveys (roster-based or free-recall name generators), archival records (co-authorship, email logs, transaction data), or direct observation. The result is a list of ordered or unordered pairs, often with tie weights or timestamps.
Ties and actors are encoded as a graph G = (V, E) where V is the set of nodes and E is the set of edges. The graph may be undirected or directed, binary or weighted. The adjacency matrix or edge list is the input to all subsequent computations.
Formula
G = (V, E) \text{ where } V = \text{nodes}, E = \text{edges}Node-level metrics — degree, betweenness, closeness, eigenvector centrality, clustering coefficient — are computed to characterize each actor's structural position. Graph-level metrics — density, diameter, average path length, transitivity — describe global network properties.
Community detection algorithms (e.g., Louvain, Girvan-Newman) or blockmodeling identify subgroups. ERGMs or QAP regression test hypotheses about tie formation. Core-periphery and bipartite decompositions reveal macro-level structure.
Structural findings are interpreted in light of the substantive question. High betweenness centrality may indicate brokerage or gatekeeping; dense clusters may indicate cohesion or redundancy. Visualization (force-directed layouts, chord diagrams) aids communication of results.
Common pitfalls
- Treating metric values (e.g., betweenness centrality) as meaningful without validating that the boundary and tie definition match the research question.
- Applying ordinary regression to network data without accounting for dyadic dependence (use QAP, ERGM, or multilevel models instead).
- Interpreting visualization aesthetics — node placement in force-directed layouts is arbitrary — as reflecting real distances or importance.
- Conflating density with cohesion or centrality with importance without theoretical grounding.
- Ignoring missing data: even 10–20% missing ties can substantially alter centrality rankings.
Frequently asked questions
What is the minimum network size for meaningful SNA?
There is no strict minimum, but very small networks (fewer than ~15 nodes) yield unstable centrality rankings sensitive to single tie changes. Descriptive SNA can work on small networks; inferential methods like ERGM require larger networks (typically 30+ nodes) for stable estimates.
Can I use SNA with incomplete or sampled network data?
Missing ties are a serious threat to validity. Strategies include sensitivity analysis, link-prediction imputation, or ego-network designs that do not require full network data. Report the extent of missingness and test robustness of key conclusions.
How do I choose between different centrality measures?
The choice should be theoretically motivated. Degree centrality captures direct activity or exposure; betweenness centrality captures brokerage and control over flows; closeness centrality captures speed of reaching others; eigenvector centrality captures influence through high-status connections. Each answers a different structural question.
How do I test hypotheses statistically in SNA?
Standard regression violates the independence assumption. Use permutation-based methods (QAP correlation and regression) for tie-level hypotheses, or Exponential Random Graph Models (ERGMs) for network-level structural hypotheses. For longitudinal data, Stochastic Actor-Based Models (SAOMs) are appropriate.
Is visualization the same as analysis?
No. Network visualization is a communication tool; node placement in force-directed algorithms is determined by aesthetic criteria, not by structural importance. Always report quantitative metrics alongside visualizations.
Related methods
Sources
Wasserman, S. & Faust, K. (1994). Social Network Analysis: Methods and Applications. Cambridge University Press.
ISBN: 978-0-521-38707-1Copied.Scott, J. (2017). Social Network Analysis (4th ed.). SAGE Publications.
ISBN: 978-1-4739-5515-1Copied.