Most everyone reading this is very familiar with the kind of "data structure" that is used in many statistical studies. The rectangular array of data that we are used to seeing in SPSS, SAS, Excel, and other programs is a "structure" that is defined by its rows (which represent cases) and columns (which represent variables). An example is shown as Figure 6.1.
Figure 6.1. Rectangular data array.
Earlier, we emphasized that the social network perspective leads us to focus our attention on the relations between actors, more than on the attributes of actors. This approach often results in data that have a different "structure" in which both rows and columns refer to the same actors, and the cells report information on one variable that describes variation (in the case of the example below, simple presence or absence of a tie) in the relations between each pair of actors. An example is given as Figure 6.2.
Figure 6.2. Square data structure for social network data
A "data structure" is simply the way in which information is recorded. These two examples are both two-dimensional (rows and columns). It is possible for a data structure or data object to have more than two dimensions. For example, if we wanted to also record information about the network relations of who is married to whom, we would usually create another table of actors by actors (that is, the row and column indexes would be the same), and record the presence or absence of marital ties. If we "stacked" the two tables together, would would have a 4 by 4 by 2 "data structure". Counts of the rows, columns, and matrices (or "slices") do not include the labeling or indexing information (i.e. it's not 5 x 5 x3).
Social Network Analysis Data Structures
Network analysts work with a variety of data structures. In this chapter, we'll look at tools for creating and manipulating the most common types.
One major "type" of data structure is the actor-by-actor matrix (like the friendship data above). This kind of structure is, by definition a "two-dimensional" and "square" (the number of rows and columns are equal) structure. The information in each cell provides information about the relation between a particular pair of actors.
The two-dimensional actor-by-actor matrix is very often expanded into a "third dimension" by adding "slices" that represent additional kinds of relations among the actors. For example, we might have an actor-by-actor matrix of Bob, Carol, Ted, and Alice that records the degree of "liking" directed from each to each. In addition, we might add a second "slice" that records the presence or absence of a kinship relation between each pair. These kinds of 3-dimensional network data structures are "multiplex". That is, they represent multiple relations among the same set of actors. Some of the special issues in working with multiplex data are discussed in Chapter 15.
The other major "type" of data structure that network analysts use looks a lot like the "rectangular data array" from conventional statistical work. The data structure consists of rows (representing actors) by columns (representing attributes of each actor - what would be called "variables" in statistics). Such an array might record just one attribute, in which case the data structure would be a "column vector". Or, such an array might record a number of attributes of each actor. Network analysts think of this kind of "rectangular" array of actors by attributes simply as a collection of vectors.
The "rectangular" data structure (called an "attribute" data set) is used in a number of ways in network analysis.
- It can record attributes of each actor that we know from other sources (e.g. gender, age, etc.).
- It can record attributes of each actor that arise from their position in the network itself (e.g. the "between-ness centrality" score of each actor).
- It can record what part or sub-part of a network an actor falls in. For example, a column in an "attribute" data structure might consist of the letters "A", "B", and "C" to indicate which of three "factors" each actor was a member of. This is called "partition".
- It can be used to tell UCINET how the actors in a matrix are to be re-arranged, or "permuted"
The "rectangular" data structure can also be used to record information about the relationships between two types of nodes (called bi-partite data). This use is so common and so important that it has a special name - an "incidence" or an "affiliation" matrix. For example, the rows might be indexed by actors (e.g. Bob, Carol...), but the columns might be the organizations that employ the actors (IBM, Sun, Microsoft...). Entries in the cells indicate the presence or strength of the relation between an actor and an employer.
Incidence or affiliation data is particularly important in many social network analyses because it is "multi-level". Actors may be tied together because they are present in the same place, time, or category (that is, they are in the same "incident", or are "affiliated" with the same structure). But such data also show how "incidents" are tied together by the "co-presence" of actors. Incidence data involving two kinds of actors (bi-partite data) are very important in network analysis because they are often our best window into questions of "agency and structure" or "macro-micro linkages".
In this chapter we will describe some of the most common kinds of manipulations that social network analysts use in creating data structures, and changing their structures to meet the needs of particular research questions. Even though this chapter is going to be a bit long, it hardly covers all the possibilities. Different questions require different data structures. The examples and tools here will get you started.