Network Analysis is a methodological approach for studying structures composed of entities and relationships among those entities. In social research, entities can include individuals, groups, organisations, communities, websites, institutions or other social actors, while relationships can represent friendship, communication, collaboration, advice, exchange, conflict or other defined connections. Network Analysis treats relationships as data rather than merely as contextual information and uses formal concepts derived from graph theory and related mathematical approaches to describe network structure.
A network is commonly represented by nodes and edges, with edges potentially directed or undirected and weighted or unweighted. The method can examine individual positions, group structures, paths, clusters, centralisation and patterns of connectivity. Network Analysis is broader than one particular statistical technique because it encompasses data representation, descriptive measures, visualisation, structural modelling and inferential approaches. Its defining feature is the analytical treatment of relational connections as fundamental units of observation.
The mathematical foundations of Network Analysis are closely related to graph theory, which represents objects as vertices and relationships as edges. Social scientists adapted these formal concepts to investigate social structure, group relations, interpersonal connections and organisational systems. Early sociometric research developed ways to represent interpersonal choices and group patterns, while later researchers introduced increasingly sophisticated measures and matrix based techniques.
Computerisation greatly expanded the size of networks that could be analysed, allowing researchers to examine thousands or millions of nodes and ties. Contemporary Network Analysis can use relational survey data, administrative records, communication logs, collaboration records, online interaction data and other sources. The development of computational methods has made it possible to study dynamic networks, multilayer networks and large scale diffusion processes. Nevertheless, computational capacity does not remove the conceptual requirement to define what the nodes and edges represent. A network generated from online follows, messages, transactions or friendship nominations answers different questions even when the same people appear as nodes.
Network representation is central to Network Analysis. An adjacency matrix records whether or how strongly pairs of nodes are connected, while an edge list records relationships as pairs or sets of connected nodes. Directed networks distinguish the source and target of a relationship, while undirected networks treat the connection as reciprocal or non directional. Weighted networks assign numerical values to edges, such as frequency of communication or volume of exchange.
Bipartite or two mode networks represent relationships between two different types of entity, such as people and organisations or authors and publications. Multilayer networks represent multiple relationship types separately or jointly, allowing researchers to distinguish friendship, advice and work ties rather than collapsing them into one network. Temporal networks incorporate changes in nodes or edges over time. These representations matter because structural measures depend on how the network is defined. Combining different types of relationship into one undifferentiated edge can conceal important differences, while excessively narrow definitions can omit relevant connections.
Network Analysis uses measures that describe local and global structure. Degree measures the number of ties connected to a node, with in degree and out degree used in directed networks. Degree centrality can identify actors with many direct connections, but it does not necessarily indicate influence or power. Betweenness centrality measures the extent to which a node lies on shortest paths between other nodes and can identify potential brokerage positions.
Closeness measures a node's distance from other nodes under particular definitions and can indicate structural proximity, although disconnected networks complicate interpretation. Eigenvector based measures evaluate connections in relation to the importance or connectivity of neighbouring nodes. At the network level, density measures the proportion of possible ties that are present, while components identify connected regions. Clustering and transitivity measures describe tendencies for connected actors to have connections among themselves. Centralisation describes the extent to which network structure is concentrated around particular nodes. Every measure has assumptions and should be interpreted according to the network type and substantive meaning of the relationships.
Network Analysis also investigates groups and structural patterns. Community detection methods seek densely connected regions or partitions according to specified algorithms, but detected communities are analytical constructions rather than automatically existing social groups. Cliques are subsets in which every member is directly connected to every other member under a defined network, while k cores identify subsets satisfying specified minimum connectivity conditions. Structural equivalence concerns actors occupying similar patterns of relationships, even if they are not directly connected.
Brokerage refers to positions connecting otherwise separated parts of a network, and structural holes describe gaps between groups that can create opportunities for actors positioned between them. Cohesion can be assessed through connectivity, density and other structural properties. Researchers must avoid treating a visually obvious cluster as proof of a socially meaningful community without additional evidence. Network structure can emerge from geography, organisational rules, homophily, selection, institutional constraints or random processes. Network Analysis can describe these patterns but requires substantive theory to explain why they occur.
Network Analysis can also be used to investigate diffusion and processes unfolding through connections. Information, behaviours, innovations, diseases, norms and other phenomena can spread along social ties, although the presence of a relationship does not guarantee transmission. Researchers may examine whether connected actors become similar over time, but similarity can arise through social influence or through homophily and selection. Distinguishing influence from selection is one of the major challenges in relational research.
Longitudinal network data can help by observing changes in relationships and outcomes across multiple periods. Stochastic actor oriented models, temporal network models and other specialised approaches can examine co evolution of networks and behaviour under explicit assumptions. Network interventions can target highly connected or strategically positioned actors, but the appropriate target depends on the process being studied because high degree, high betweenness and other forms of structural position have different implications. Network Analysis therefore provides a framework for investigating relational mechanisms while requiring careful causal reasoning about how ties and outcomes develop together.
Data quality is particularly important in Network Analysis because missing or incorrectly recorded relationships can alter structural measures substantially. A missing node may fragment a network, while a missing edge can change centrality, density or community structure. Network boundaries can also create artificial patterns. If researchers define the network too narrowly, actors may appear disconnected even though they maintain relationships outside the study population. If they define it too broadly, irrelevant relationships can obscure the structure of interest.
Respondent generated network data may be affected by memory, differing definitions and incomplete reporting. Digital network data may be technically extensive but still omit offline relationships and platform independent interactions. Privacy is a major concern because relational datasets can reveal who communicates with whom even when message content is unavailable. Researchers should therefore minimise unnecessary collection, secure identifiable network data and consider the privacy implications for both participants and their contacts.
Network Analysis is valuable because it provides a formal way to study relational structure that conventional variable based approaches can overlook. It can reveal brokers, clusters, isolated actors, cohesive groups, structural gaps and patterns of connectivity that are difficult to capture through individual level characteristics alone. It is used in social science, organisational studies, epidemiology, communication research, bibliometrics, criminology, political research and many other areas.
Its limitations include sensitivity to network boundaries, missing data, measurement choices, dependence between observations and the difficulty of drawing causal conclusions from static relational structures. Visual network diagrams can be useful for exploration and communication, but visual prominence should not be confused with analytical importance. A large node on a graph may simply have been drawn large by an arbitrary software setting rather than because it has greater social influence. Strong Network Analysis therefore combines mathematically defined measures with substantive knowledge, transparent data construction and appropriate statistical reasoning.
