attributes (Diani, 2004; Huckfeldt & Sprague, 1987; Knoke, 1990). Many actor attributes remain unaltered across the numerous social settings in which they participate (a woman’s age, race, and education remain unchanged whether at home, at work, and at church). In contrast, many structural relations occur only at specific time-and-place locales and either vanish or are suspended when participants are elsewhere (e.g., student-teacher and doctor-patient relations do not exist outside school and clinic settings, respectively). A man holding a menial factory job requiring little initiative may be the dynamic leader of his church and an enthusiastic softball team player. Such behavioral differences are difficult to reconcile with unaltering gender, age, and status attributes but comprehensible on recognizing that people’s structural relations can vary markedly across social contexts within which they are embedded. The structural-relational explanations favored by network analysts depart markedly from substantialist approaches premised on static ‘‘thing-concepts’’ as their primary units of analysis: essences, self-action, norm-based conformity, rational choice, and variable-centric and social identity approaches (Emirbayer, 1997). In assuming that patterned relations influence social entities apart from their attributes, network analysis offers distinctive theoretical and empirical explanations of the origins of social action.
Second, social networks affect actor perceptions, beliefs, and actions through diverse structural mechanisms that are socially constructed by relations among entities. Direct contacts and more-intensive interactions dispose people and organizations to be better informed, more aware, and more susceptible to influencing or being influenced by others. Indirect relations through intermediaries (in popular imagery, agents who broker connections for their clients) also bring exposure to new ideas and potential access to useful resources that may be obtained through exchanges with others. For example, in a classic network study by Mark Granovetter (1973), job seekers typically obtained less useful information from their intimate circles, whose members already shared and circulated the same intelligence, than from their weaker and more distant social contacts. Relational structures provide complex pathways for assisting or hindering flows of knowledge, gossip, and rumor through a population (Fang, McAllister, & Duffy, 2017). A variety of structural-relational factors explains racial differences in the spread of HIV/AIDS infections among young men who have sex with men (Mustanski, Birkett, Kuhns, Latkin, & Muth, 2015) and the propagation of financial distress through the international banking network during the global financial crisis of the aughts (Kojaku, Cimini, Caldarelli, & Masuda, 2018). Physical illness, mental health, and recovery from substance abuse are strongly affected by people’s social support networks (Cullen, Mojtabai, Bordbar, Everett, Nugent, & Eaton, 2017; Stevens, Jason, Ram, & Light, 2015), with social media exerting some unusual impacts (Lu & Hampton, 2017; Pallotti, Tubaro, Casilli, & Valente, 2018). Structural relations are vital to building cohesion and solidarity within a group but may also reinforce prejudices and intensify conflict with out-groups (Bliuc, Faulkner, Jakubowicz, & McGarty, 2018; Roversi, 2017). Competitive and cooperative relations enable innovation in corporate supply chains (Delgado-Márquez, Hurtado-Torres, Pedauga, & Cordón-Pozo, 2018), mobilization for collective action by social movements (Diani, 2016), and the operation of ‘‘dark networks’’ for drug trafficking, immigrant smuggling, and terrorist campaigns (Wu & Knoke, 2017). By channeling information, money, and other types of resources to particular structural locations, networks help to create interests and shared identities and to promote shared norms and values. Network analysts seek to uncover the mechanisms through which social relations affect social entities and to identify the contingent conditions under which particular mechanisms operate in specific empirical contexts.
The third underlying assumption of network analysis is that structural relations should be viewed as dynamic processes. This principle recognizes that networks are not static structures but are continually changing through interactions among people, groups, or organizations. In applying their knowledge about networks to leverage advantages, network entities also transform those structural relations, both intentionally and unintentionally. For instance, in an intervention experiment to reduce conflict and bullying among students in 56 schools, experimenters comprehensively measured every school’s networks, then randomly selected “seed groups” of 20 to 32 students to be encouraged to take public stands against conflict (Paluck, Shepherd, & Aronow, 2016). Disciplinary reports of conflict fell by 30% in the treatment schools compared to control-group schools, but the effect was stronger for seed groups containing more students who attracted greater student attention. Apparently, those popular students changed their network peers’ beliefs and behaviors by publicly stigmatizing conflict and bullying as less socially normative. Such dynamics exemplify the more general ‘‘micro-to-macro problem’’ in the theory of social action (Coleman, 1986). The core issue is how large-scale systemic transformations emerge out of the combined preferences and purposive actions of individuals. Because network analysis simultaneously encompasses both structures and entities, it provides conceptual and methodological tools for linking changes in actors’ microlevel choices to macrolevel structural alterations. The increased availability of longitudinal datasets, especially large online networks, coupled with methodological developments for analyzing multilevel relations, are accelerating research on cross-level dynamic processes (Lazega & Snijders, 2015; Snijders, Steglich, & Schweinberger, 2017). Likewise, developments in temporal exponential random graph models (TERGMs) and stochastic actor-oriented models (SAOMs), such as SIENA, hold great promise to advance our understanding of network dynamics (Leifeld & Cranmer, 2019; Leifeld, Cranmer, & Desmarais, 2018).
2.2 Entities and Relations
The two indispensable elements of any social network are entities and relations. Their combination jointly constitutes a social network, as described in the next subsection. Entities may be individual natural persons or collective actors such as informal groups and formal organizations. Common examples of individual actors include children on a playground, high school students attending a prom, employees in a corporate work team, staff and residents of a nursing home, and terrorists operating in a covert cell. Collective actors might be firms competing in an industry, voluntary associations raising funds for charities, political parties holding seats in a parliament, and nations signing a military alliance. Other types of entities lack human agency, such as bills debated in a legislature, dances attended by students, and books read by library patrons. Sometimes networks are comprised of diverse types of entities, such as a healthcare system consisting of doctors and nurses, patients, clinics, hospitals, laboratories, insurance companies, and governmental regulations.
A relation is generally defined as a specific kind of contact, connection, or tie between a pair of entities, or dyad. Relations may be either directed, where one actor initiates and the second actor receives (e.g., advising, selling), or undirected, where mutuality occurs (e.g., conversing, collaborating). A relation is not an attribute of one entity but is a joint dyadic property that exists only so long as both participants maintain their association. An enormous variety of relations among individual and collective entities may be relevant to representing network structures and explaining their effects. At the interpersonal level, children befriend, play with, fight with, and confide in one another. Employees work together, discuss, advise, trust, undermine, and betray. Among collectivities, corporations exchange goods and services, communicate, compete, sue, lobby, and collaborate. In healthcare systems, physicians refer patients to specialty clinics, pharmacies, laboratories, hospitals, imaging centers, nursing homes, and hospices. Which specific type of relation a network researcher should measure depends on the particular objectives of the research project. For example, an investigation of community networks will likely examine various neighboring activities, whereas a study of banking networks would investigate financial transactions. Of course, some analyses scrutinize multiple types of relations, such as the political, social, and economic ties among corporate boards of directors. We present a general classification of relational contents in the next subsection.
Social science researchers rely heavily on measuring and analyzing the attributes of individual or collective units of analysis, whether through survey, archival, or experimental data collection. Although attributes and relations are conceptually distinct approaches to investigating
