Using the articles that we classified as purchasing-related, we conducted a social network analysis in order to better understand the underlying social network of purchasing knowledge production. Specifically, we sought to identify which
universities or institutions represented the largest sources of education in the purchasing field and to identify which institutions produced the most purchasing knowledge through publication. The adjacency matrix in Table 3 and the
corresponding sociogram in Figure 2 provide an example of how we calculated these relationships.
Table 3. Sample Adjacency Matrix (Carter, Leuschner, et al., 2007)
Sample Adjacency Matrix
University
A University
B University
C University
D University E
University A - 13 2 0 1
University B 12 - 1 0 5
University C 4 6 - 0 9
University D 2 4 5 - 2
University E 1 2 1 0 -
Figure 2. Sample Sociogram (Carter, Leuschner, et al., 2007)
To examine the influence that universities and other institutions have had with respect to educating and publishing in the purchasing field, we calculated the
network centrality of the data. This network centrality measurement used Leavitt’s (1951) work, which stated that the extent of participation by actors within a network can be visualized. The network centrality measurement also relied heavily on Freeman’s (1979) degree and betweenness dimensions of centrality.
We used this method of centrality analysis for this research because it
addresses the goals of this research and because this methodology has been widely accepted and adopted within the field of social network analysis research (Scott, 2000). This type of social network analysis research has been conducted previously.
Carter, Leuschner, et al. (2007) conducted a social network analysis to examine the measure of network centrality among universities, but they only examined the Journal of Supply Chain Management from 1965–2004.
The term degree, as defined by Carter, Leuschner, et al. (2007), is “the number of ties that an actor has to other actors in a network” (p. 18). For the purpose of this research, the different institutions that educate and publish in the purchasing field represent “actors.” Carter, Leuschner, et al. (2007) demonstrated that an actor’s degree is determined by summing the links between that actor and the other actors within a sociogram. An example of this means of calculation is shown in Figure 3 in Chapter IV. Another method for calculating an actor’s degree is by using an adjacency matrix, where the values within a row or column are added to determine the degree, as shown in Figure 2. For the purposes of this research, a degree is represented by each occurrence of an author being educated at one university and publishing at another institution or university. An example of this would be if they were educated at University A and published at University B.
To best represent the directional flow of education to publishing that we sought to analyze in this network, we used a matrix structure to frame the data. For our purposes, the column of university names in Table 3 represents the institution at which the author received his or her terminal degree, and the row of universities represents his or her current affiliation from which they published the journal article.
For the five-university example depicted in Table 3 and Figure 2, University A has an “in-degree” of 19. This in-degree value means that of the 19 occurrences of authors publishing at University A, 12 of those being published were educated at University B, four at University C, two at University D, and one at University E.
Providing the counterpoint to this is University D, which has a degree of 0 because it
did not have any published authors who received their educations from University A, B, C, or E.
Betweenness, from a social networking perspective, can be defined as the total number of paths that pass through a particular actor who is on the shortest path connecting two other actors (Freeman, 1979). When viewed with an academic
perspective, an actor with relatively high betweenness centrality has greater
influence over the network. This influence implies that it can act as a liaison between actors that have a lower betweenness centrality and exist in more isolated areas of the network (Ronchetto, Hutt, & Reingen, 1989). Using Freeman’s (1979)
methodology for calculating betweenness centrality values, it was first necessary to put the relational values shown in the adjacency matrix (see Table 3) into a binary format to calculate the number of paths that pass through any given actor. This methodology indicates that by taking the shortest routes in the sample network shown in Table 3 and Figure 2, it is possible to calculate the betweenness scores for each of the sample universities. For example, University B is between Universities D and C, A and C, and E and C, and as a result, achieves a betweenness score of three. The same exercise can be completed for the other universities and results in University A receiving a score of one, and both Universities C and D receiving scores of zero because no paths cross through either institution.
To examine the measures of network centrality, we used a database of
information derived from the journal articles we reviewed. This database contained a list of the authors, the schools from which they received their terminal degrees, and the school or institution which they were affiliated with at the time the article was published. We used this information to construct a 653 by 653 cell matrix in a spreadsheet that encompassed all of the possible relations between the different institutions in a similar format to the example depicted in Table 3. We then imported this spreadsheet into UCINET 6, a powerful social network analysis tool (Borgatti, 2002). After that, we ran the data through the NetDraw interface of UCINET 6 to conduct the analysis.
The reliability of aggregate social network analysis measures (such as popularity) is higher than the reliability of “choices” made by individual actors (Burt, Marsden, & Rossi, 1985). This means that because our conclusions are drawn from the analysis of the data points as a whole, the recommendations drawn from the social network analysis are more reliable than the data points individually. To ensure that the results obtained from our social network analysis were valid, we utilized large sample sizes. Research conducted by Scott (2000) indicated that “if the sample is large enough, [social network analysis] estimates ought to be reliable” (p.
59). Our specific findings from the social network analysis can be found under the Social Network Analysis heading in Chapter IV.