Читать книгу Gathering Social Network Data - jimi adams - Страница 16

In principle, anything that could be represented as a graph could be considered a network and analytically could be examined with SNA. This has often been the practice in the physical and biological sciences (Borgatti et al., 2009), and some in the social sciences have even argued that as one of the beauties of social network analysis—that regardless of the type of nodes or ties between them—the analytic principles can be applied to any network in much the same way (Wellman, 1988).¹⁶ However, just because different networks can be analyzed with the same approaches doesn’t mean they necessarily should be. Different network types could lead to different applications of the same descriptive concepts; many core network ideas (e.g., centrality or communities) have multiple alternate strategies for their measurement, and it’s often easiest to select between those based on differences between the types of networks being described.¹⁷ Moreover, the theoretical mechanisms that provide accounts for different explanatory expectations within networks differ substantially depending on the type of network being examined (Erikson, 2013; Fuhse, 2019, Valente & Pitts, 2017). In either case—and as with any solid social science research—the aims of a network study (whether descriptive, explanatory, or otherwise) must carefully consider what the research aims to address in order to determine what sorts of data will allow them to best examine those questions. Here, we must consider what type(s) of ties the questions are about, how readily researchers can actually capture the types of ties required by their research questions, and whether they will be limited to some sort of proxies for the relationships of actual interest.

¹⁶ My former PhD advisor has even been accused at times of literally seeing everything as a network. At a recent workshop we both contributed to, he was the primary person making that accusation.

¹⁷ For a review of various strategies for conceptualizing and operationalizing the differences between centrality measures, see Borgatti and Everett (2006); for a similar treatment of network communities, see Fortunato (2010) and Porter, Onnela, and Mucha (2009).

Borgatti et al. (2009) provide a typology of the types of ties that are frequently the focus of social network research;¹⁸ for a summary, see Table 1.2. They differentiate between three primary types of ties: social relations, interactions, and flows.¹⁹

¹⁸ The terms relationship and tie are often used more or less interchangeably in the social networks literature. I will attempt to avoid this unnecessary confusion, aiming to use tie as the “catch-all” term and relationship in the specific meaning provided here (see also Erikson, 2013; Kitts, 2014). In leaning on examples from others, I may occasionally slip into the literature norm of also using relationship as the generic term.

¹⁹ See Figure 3 in Borgatti et al. (2009). Their typology also includes a fourth type that I will not address in this book: similarities. These are merely dyadic comparisons of some individual attribute (e.g., same gender). While similarities are dyadic measures, they are not conceptually relational by nature. As such, their measurement and modeling are not captured any better by network approaches than by individually oriented research methods and analytic strategies. Similarities are often useful in the analytic modeling of social networks. However, since measuring similarities do not rely on any uniquely network approaches, I leave you to other research methods texts for optimizing their capture.

Table 1.2a

^aAdapted from Borgatti et al. (2009).

Social relations capture the various relationally defined positions a person can occupy with respect to others; these often have a strong social basis and/or foundations in theoretical social science literature. Role theory asserts behavioral expectations upon occupants of certain roles (e.g., parents should behave in particular ways toward their children). Recognizing the relational basis of those roles allows us to identify how these expectations derive from the pattern of relationships that define the role, rather than a more essentialist notion of role expectations determined from the label itself. For example, a parent’s role is determined by the kinship ties they have to their co-parent, their children, and often even their own parents.²⁰ In social relationship terms, a role is defined by the constellation of these others to whom the person is connected. Such kinship relations have been the focus of relational social scientists for decades (Bott, 1957; Stack; 1974, D. R. White & Jorion, 1992; H. C. White, 1963). In addition to kinship ties, Borgatti et al. (2009) describe other social relations that are based on other roles (e.g., friends), affective relationships (e.g., likes/dislikes), or cognitive links (e.g., knows the work of). This variety of social relationships shares a number of common features that make their measurement more readily available—they are generally relatively temporally stable ties, and each member of the relationships can generally readily identify both members’ participation in the relationship. This makes gathering relational information about such roles easily incorporated into a survey-based research design by simply tacking such questions onto individually oriented surveys.

²⁰ The “Category” label in Table 1.2 should not be interpreted as indicating those differences only apply to the row specified (e.g., interactions can be subjective or objective, and relations can be mutual or directed), but these are primary delineations on the types of ties that are often the focus of research in these domains (e.g., the perception vs. reality of diffusion [knowledge vs. information]).

Social interactions capture the joint participation by pairs of nodes in shared activities. The types of interactions that are most commonly studied are things like sent and received messages, engaging in sexual intercourse, the joint use of injecting drug equipment (e.g., needles), or other shared experiences (e.g., meals). Interactions are often more temporally fleeting than social relationships and frequently aim to capture the behavioral nature of shared activities—as opposed to the social nature of roles.

Moreover, the interaction examples provided in Table 1.2 introduce the notion that ties can also be undirected (mutual) or directed. An undirected social relationship looks the same from the perspective of each party involved; each sibling is sibling to the other. Contrastingly, a directed relationship necessarily involves two members of differing, complementary, roles. A parent–child relationship involves two members occupying different roles. Many interactions are directed as well, involving sender and receiver roles (e.g., a speaker and a listener if the interaction is a specific speech unit within a conversation).

Often these roles or interactions can form the basis for potential flows between partners, which are the final type of ties identified by Borgatti et al. (2009). So, the needle sharing mentioned above may lead to disease transmission, or conversations may allow knowledge to pass from one individual to another. Flows may also be the primary tie type of interest, independent of how roles or interactions shape their possibilities (e.g., in studies of financial remittances). Importantly, scholarship has shown that identifying the actual transmission of ideas through a population (e.g., diffusion of knowledge) can provide considerably different estimates than when we ask people to account for who influenced them on a particular idea (i.e., perception of information flows) (J. Young & Rees, 2013). The objective–subjective distinction here is therefore primarily one for researchers to carefully consider in deciding which is the aim(s) for their research.

A project’s aims can often lead researchers to be readily able to identify one (or more) of these tie types as its primary conceptual focus. Furthermore, in many cases, this conceptualization is easily translatable into a measurement strategy. However, in other cases, simply because that identification is conceptually possible does not mean that gathering data on that tie type in the theoretically salient dimension is equally viable. For example, suppose your interest is in mapping the risk-relevant network that promotes a chlamydia epidemic. The relevant network that you would want to map would include all sexual contacts (interactions) that occur between sero-discordant partners.²¹ Additionally, sero-discordance is not a permanent status, so to properly map that risk network, you’d need those interaction data at the level of individual acts, along with each individual’s time-specific sero-status. It is highly implausible that this level of measurement precision would be available to even the most scrutinizing researcher’s data collection efforts.

²¹ That is, one partner who has chlamydia and one who does not. In other words, your interest is in data on the population of potentially transmitting interactions.

While this particular example is extreme, it reflects a common occurrence in social network data collection efforts. There often arises conceptual slippage between the level at which researchers desire to gather data and the level that is accessible to them. That is, research must regularly rely on relational proxies—often that move “up” in the level of generalization (i.e., from flows toward social relations). We may only be able to measure social relationships that include sexual contact, not each sexual act, when studying a chlamydia outbreak. Tie directionality can similarly require measurement proxies. For example, researchers may have access to only one member of a reported relationship, and if that person reports having provided support to their partner, we must take them at their word that the other partner received that support (but see Barrera, 1986). While careful qualifications within analytic interpretations can potentially acknowledge the limitations of such proxies, researchers have increasingly acknowledged that such slippages have implications beyond the measurement level and have argued for thinking about different types of ties as having different theoretical—as well as methodological—implications that researchers must consider (Kitts, 2014).