One of the distinguishing features of network institutionalism is the availability of a range of quantitative techniques designed to analyze the properties of networks. The development of these techniques grew out of the use of graph theory to represent networks, though much recent network analysis also draws on algebraic methods. It is beyond the scope of this article to provide more than a cursory discussion of these methods. However, several book-length introductions are available. Scott (1998) and Degenne and Forse´ (1999) provide useful surveys of social network analysis and Wasserman and Faust (1994) provide a compre- hensive, but more mathematically demanding treatment. Several software pro- grams are also available for social network analysis, of which the most popular is UCINET.
Prominent techniques of social network analysis include centrality and ‘‘sub-group’’ identiWcation. Centrality is a particularly useful measure because it identiWes the relative importance or prominence of individual actors in a network based on information about all the actors in the network. Various measures of centrality have been developed (degree, closeness, betweenness, etc.) that seek to capture diVerent aspects of what it means to be a central actor. For example, betweenness centrality deWnes centrality in such a way as to identify actors likely to serve as important brokers. Another class of network techniques identify ‘‘sub-groups’’ within the network and they are particularly useful for identifying social cleavages or factions. These techniques range from those that identify sub-groups in relatively inclusive terms (e.g. component analysis) to those that are much more restrictive (e.g. clique detection).
Social network analysis also distinguishes between ‘‘cohesion’’ and ‘‘equivalence’’ as the basis for sub-groups. The cohesion approach suggests that sub-groups are based on the density of direct dyadic ties. Hence, the greater the number of ties within a group, the more cohesive it should be. By contrast, the equivalence approach argues that sub-groups will be composed of actors with equivalent ties to third parties. Marx’s analysis of class formation is a classic example: workers are brought together not by their direct solidaristic ties, but by their common opposition to employers.
The distinction between cohesion and equivalence is related to a broader set of discussions in network analysis. Research on what came to be known as the ‘‘small world phenomenon’’ discovered that people were often connected to quite distant others through a surprisingly short number of intervening steps. As Watts (2003) has clariWed, this is most surprising when networks are relatively ‘‘sparse.’’ Watts found that small world networks have particular properties. They exhibit high local clustering combined with a limited number of ‘‘shortcuts’’ between clusters. Granovetter (1973) also built on the small world phenomenon in his inXuential argument about the ‘‘strength of weak ties.’’ He found, for instance, that jobs were often not found directly through friends (strong ties), but through friends of friends (weak ties). The logic is that weak ties often ‘‘bridge’’ across clusters. Burt (1992) has further reWned this logic in his work on ‘‘structural holes.’’ He argues that informa- tion in small tightly knit clusters is redundant (everybody knows everybody’s business). Moreover, clustering creates ‘‘holes’’ in the global network that limit the Xow of information. Thus, ties that bridge across structural holes (‘‘shortcuts’’ in Watt’s terms, ‘‘weak ties’’ in Granovetter’s) are powerful conduits of information.
The cohesion perspective suggests that the critical mechanism in networks oper- ates through direct dyadic ties. An extension of this logic suggests that the stronger the tie (e.g. the more frequent, intimate, and intense the interaction), the more cohesive the relationship. At the global network level, then, a denser network is presumed to be a more cohesive one. The logic extends to multiple networks. Network analysis refers to the situation in which two actors are tied together in diVerent types of ways—for example friendship, advice, co-work, residence—as multiplexity. In the cohesion logic, the more multiplex the network, the stronger it is. By contrast, the equivalence perspective emphasizes the importance of indirect as well as direct ties. Actors are similar not because they have strong ties to one another, but because they have similar ties to others. Actors who are structurally equivalent are therefore interpreted as having a similar position in the network. Multiple networks are important when they reinforce structural equivalence.
The diYculty of collecting network data has been one of the limits on the more widespread usefulness of social network methods. Two basic classes of network data exist. Egocentric networks begin with a focal actor or actors (ego) and then collect network information on relationships of ego to others (alters). A later phase of data collection collects further information on the relationships between ego’s alters. The general problem with egocentric data is that it is highly selective, since by deWnition it reXects only ego’s network. Alternatively, a complete network provides a more comprehensive perspective. Data for a complete network are collected by Wrst identifying a group of actors and then collecting information on relationships between all of them. Such data can be diYcult to collect for two reasons. First, identifying connections between all the actors in a network creates a large volume of data for even a small number of actors. Second, complete networks confront a problem of boundary speciWcation. As the small world phenomenon demonstrates, everyone may be (at several removes) connected to everyone else. So where should the boundary be drawn? Network analysts generally solve this problem in one of two ways—each of which corresponds to a diVerent technique for gathering the data. One approach is to specify the boundary at the outset on the basis of non-network criteria—for example the boundary of the organization or work unit, the policy sector, or geographical units. In such cases, it is often useful to begin with a complete list of the individuals, groups, or organizations contained within this boundary. The researcher then asks each actor on the list about their relationship with every other actor on the list. A second approach is often used when the boundary is diYcult to specify ahead of time. In fact, identiWcation of who is part of the network may be one of the main purposes for gathering data. In this case, snowball sampling is used to collect network data. Much like egocentric data, this approach starts with a few focal actors and then asks them about their relationships. It then builds outward, asking actors speciWed in the Wrst round of interviewing who they are related to. Sampling may continue until the discovery of new actors drops oV.
