Читать книгу Social Network Analysis - Song Yang - Страница 15

Three elements of network research design shape the measurement and analysis strategies available to researchers: social settings, relational form and content, and level of data analysis. Every network data collection project must involve making explicit choices about these elements before beginning fieldwork. Varying combinations of them generate the wide range of social network investigations published in the research literatures of numerous disciplines.

Social Settings. The first steps in designing a network study are to choose the most relevant social setting and to decide which entities in that setting comprise the network entities. Ordered on a roughly increasing scale of size and complexity, a half-dozen basic units from which samples may be drawn include individual persons, groups (both formal and informal), complex formal organizations, classes and strata, communities, and nation-states. Some two-stage research designs involve a higher-level system within which lower-level entities comprise the actors. Common examples are hierarchical social settings such as corporations with employees, schools with pupils, hospitals with physicians, municipal agencies with civil servants, and universities with colleges with departments with professors.

The earliest and still most common network projects select small-scale social settings—classrooms, offices, factories, gangs, social clubs, schools, villages, artificially created laboratory groups—and treat their individual members as the actors whose relations comprise the networks for investigations. Recent examples include bullying and homophobic teasing among middle school students (Merrin, De La Haye, Espelage, Ewing, Tucker, Hoover, & Green, 2018), helping and gossip networks among employees of a Turkish retail clothing company (Erdogan, Bauer, & Walter, 2015), and the effects of ethnic diversity on the spread of word-of-mouth information in two matched rural Ugandan villages (Larson & Lewis, 2017). Small settings have considerable advantages in sharply delineated membership boundaries, completely identified populations, and usually researcher access by permission from a top authority. However, network analysis concepts and methods are readily applied to larger-scale formations, many of which have porous and fuzzy boundaries, including clandestine networks. Examples include peer network origins of adolescent dating behavior (Kreager, Molloy, Moody, & Feinberg, 2016), criminal organizations in communities of Calabria, Italy (Calderoni, Brunetto, & Piccardi, 2017), and strategic alliances among multinational corporations in the Global Information Sector (Knoke, 2009).

Relational Form and Content. Network researchers must decide on which particular relations to collect data. Relations among pairs of social actors have both form and content, a dichotomy that Georg Simmel (1908) proposed in his classic analyses of association. The two elements are empirically inseparable and only analytically distinguishable. Contents are the interests, purposes, drives, or motives of individuals in an interaction, whereas forms are modes of interaction through which specific contents attain social reality. Simmel argued that the task of sociology is to identify a limited number of forms—sociability, superiority, subordination, competition, conflict, cooperation, solidarity—that occur across a wide range of concrete settings, social institutions, and historical contexts. A particular form can vary greatly in content. For example, the basic forms of superordination and subordination are ever present in government, military, business, religious, athletic, and cultural institutions. Conversely, diverse contents like economic interests and drives for power are manifested through forms of competition and cooperation.

The form-content dichotomy also applies to social network analysis. Relational form is a property of relations that exists independently of any specific contents. Two fundamental relational forms are (a) the intensity, frequency, or strength of interaction between pairs of entities and (b) the direction of relations between both dyad members—null, asymmetric, or mutual choices. Relational content refers to its ‘‘substance as reason for occurring’’ (Burt, 1983, p. 36). Substantive content is an analytic construct designed by a researcher to capture the meanings of a relation from the informants’ subjective viewpoints. When people are asked, ‘‘please identify your close friends, friends, and acquaintances” in some social setting, the intended relational content is “friendship.” The results of this query depend on how each actor first conceptualizes the meanings of the three proffered response categories and then classifies the other actors according to recollections of diverse interpersonal interactions. Obviously, people may vary markedly in their interpretations of both the friendship labels and those activities that they consider to be indicators of greater or lesser intimacy. Friendship dyads are never precisely reciprocated and the level of intimacy may be very unequal; for example, one dyad member considers the second person a “best friend,” but the second member views the first person as a “friend.” The National Study of Adolescent Health (Add Health) found that girls and Asian Americans were most likely to have reciprocated friendships, whereas interracial friendships were much less common than friendships between students of the same race (Vaquera & Kao, 2008).

The choice of relational content, also called type of tie, is largely determined by a project’s theoretical concerns and research objectives. A study of healthcare networks could inquire into people’s interpersonal sources of trusted information and advice about health-related matters, whereas a project on political networks might ask them to identify others with whom they discussed or participated in political affairs. Some substantive problems imply that more than one analytically distinct relational content should be investigated, in which case measuring and simultaneously analyzing two or more types of ties (i.e., multiplex networks) is an appropriate strategy. For example, psychologists asked 132 undergraduates at Midwestern University to list their Facebook friends who fulfilled each of five social functions (i.e., types of ties): sharing social activities, discussing personal matters, providing instrumental support, providing emotional support, and sharing success and happy events (Gillath, Karantzas, & Selcuk, 2017). Students with higher attachment avoidance were likely to ascribe fewer multiplex social roles to their networks’ members, implying a lower degree of social trust.

Inexplicably, network analysts have conducted little research on the connections among diverse domains of relational contents. Ronald Burt (1983) examined survey respondents’ perceptions of relational contents and uncovered substantial confusion, redundancy, and substitutability among the 33 questions posed to a sample of Northern Californians. He concluded that just five key questions would suffice to recover the principal structure of relational contents in the friendship, acquaintance, work, kinship, and intimacy domains. However, we still need much more research on the similarities and differences of meanings that people attach to commonly used relational terms and labels in a wide variety of network settings. A cognitive map of the structural connections among relational content domains would enable researchers efficiently and accurately to select specific contents most relevant to their theoretical and substantive concerns.

Until that desideratum arrives, in the spirit of Simmel we propose a small typology of generic contents:

 Transaction relations: Entities exchange control over physical or symbolic media, for example, in gift giving or economic sales and purchases.

 Communication relations: Linkages between entities are channels through which messages may be transmitted.

 Boundary penetration relations: Ties consist of membership in two or more social formations, for example, voluntary associations or social movement organizations.

 Instrumental relations: Actors contact one another in efforts to obtain valued goods, services, or information, such as a job, an abortion, political favors, or religious salvation.

 Sentiment relations: Perhaps the most frequently investigated networks involve actors expressing their feelings of affection, admiration, deference, loathing, or hostility toward one another.

 Authority/power relations: These types of ties, usually occurring in formal hierarchical organizations, indicate the rights and obligations of position holders to issue and obey commands.

 Kinship and descent relations: These bonds of blood and marriage reflect relations among family roles.

Levels of Analysis. After deciding the social setting and the relational forms and contents, researchers have several alternative levels at which to analyze the structures in data that they collect for social network projects. Details of appropriate measures and methods appear in Chapters 3 through 5, but here we summarize four conceptually distinct levels of analysis that analysts could investigate.

The simplest level is the egocentric network, consisting of one actor (ego) and all other actors (alters) with which ego has direct relations as well as the direct relations among those alters. This set is also called ego’s ‘‘first zone,’’ in contrast to second and higher zones consisting of all the alters of ego’s alters, and so on. If a network’s size is N actors, an egocentric analysis would have N units of analysis. Each ego actor can, in turn, be described by the number, intensity, and other characteristics of its linkages with its set of alters, for example, the proportion of reciprocated relations or the density of ties among its alters. An egocentric analysis of incarcerated California youths indicated that respondents reporting no close friendships within the facility had lower postinterview misconduct than those who nominated peers, suggesting an influence or amplifying effect of friends on misbehavior (Reid, 2017). In some respects, egocentric analysis resembles typical attribute-based survey research, with a respondent’s individual characteristics such as gender, age, and education supplemented by measures derived from that person’s direct network relations. Egocentric network research designs are well suited to surveys of respondents who are unlikely to have any contact with one another. The 1985 General Social Survey of the adult U.S. population (Marsden, 1987) pioneered procedures for identifying and eliciting information about a respondent’s alters, which we describe in some detail in Chapter 3.

A second level of analysis is the dyadic network, consisting of pairs of actors. If the order of a pair is irrelevant—as in marital status where persons are either unmarried, cohabiting, married, separated, or divorced—a sample of N actors has (N² – N)/2 dyadic units of analysis. But, if the direction of a relation matters, as in giving orders and taking advice, then the sample contains (N² – N) ordered dyads. The most basic questions about a dyad are whether a specific type of tie exists between two actors, and, if so, what is the intensity, duration, or strength of that relation? A closely related issue is whether a dyad without a direct tie is nevertheless indirectly connected via ties to intermediaries (e.g., brokers, go-betweens). Typical analyses seek to explain variation in dyadic relations as a function of pair characteristics, for example, the homophily hypothesis that ‘‘birds of a feather flock together’’ or the complementarity hypothesis that ‘‘opposites attract.’’ Dyadic empathy—‘‘a combination of perspective taking and empathic concerns for one’s romantic partner”—is associated with higher sexual satisfaction, relationship adjustment, and sexual desire of first-time parents (Rosen, Mooney, & Muise, 2017, p. 543).

A third level of network analysis is, unsurprisingly, triadic relations. A set of N actors has triples, the number of ways to take N actors, three at a time. All possible combinations of present and absent directed binary relations among the actors in a triple generates a set of 16 distinct triad types. A basic descriptive question for empirical network analysis regards the distribution of observed triads among the 16 types, a summary tabulation called the triad census. Substantive research on triadic structures concentrated on sentiment ties (liking, friendship, antagonism), with particular interest in balanced and transitive triadic relations (e.g., if A chooses B and B chooses C, does A tend to choose C?). Because we lack space to review triad analysis methods, interested readers should consult the research program of James Davis, Paul Holland, and Samuel Leinhardt (Davis, 1979) and a comprehensive treatment by Wasserman and Faust (1994, pp. 556–602) for details.

Beyond the three microlevels, the whole network (also called complete network) is the most important macrolevel of analysis. Researchers use the information about every relation among all N actors to represent and explain an entire network’s structural relations. Typical concerns are the presence of distinct positions or social roles within the system that are jointly occupied by the network actors and the pattern of ties within and among those positions. Although a whole network has N actors and (N² – N) dyads (assuming directed relations and self-relations are generally ignored), these elements add up only to a single system. Examining the causes or consequences of structural variation at the whole network level of analysis typically involves measures of the global structural properties. An example is a Dutch online social network of more than 10 million users living in 438 municipalities (Norbutas & Corten, 2018). Communities with higher network diversity were more economically prosperous than less-diverse communities, whereas greater network density at the community level was negatively associated with prosperity.

The four levels of network analysis imply that emergent phenomena at one level cannot be simply deduced from knowledge of the relations at other levels. For example, transitivity of choice relations is a substantively important variable for theories of friendship formation (‘‘a friend of my friend is my friend’’), which can be observed at the triadic level but not at the egocentric or dyadic level. For another illustration, Mark Newman (2001) found that coauthorship networks in biomedical research, physics, and computer science were each structured as “small worlds,” where only five or six steps were necessary to connect random pairs of scientists. However, biomedical research was dominated by many people with few coauthors, in contrast to other disciplines characterized by a few people with many collaborators (see also, e.g., Ebadi & Schiffauerova, 2016; Maggioni, Breschi, & Panzarasa, 2013). The adaptability of network principles and procedures to investigate structural relations across multiple levels of analysis underlies its bourgeoning popularity for theorizing about social action and guiding empirical research.