Columbia’s CCP: A Case Study
Appendix B: Network Analysis Strategy
Distances among the targets were measured using a structural equivalence approach (cf. Lorrain & White, 1971), which overcomes some of the short- comings of the conventional graph theory. Following the lead of Heinz and Manikas (1992), distances among the targets were measured by determining the overlap of acquaintances for any two actors, defined here as “the degree to which the persons who are in contact with each of them are the same peo- ple (p. 840).” The main benefit of this structural equivalence approach is that it circumvents the problem of missing data and allows us to compare patterns of contact for individuals who are not interviewed. This is only possible be- cause our sample includes a sufficient number of respondents who know both individual targets. The alternative approach (i.e. the graphic theoretic ap- proach, which measures similarity by counting the number of links in the communication network to get from person A to X) would require the collec- tion of data from all people in the chain.
Multidimensional scaling was used to analyze our network data. As Scott (1991, p. 151) observes, “The mathematical approach termed ‘multidimen- sional scaling’ embodies all the advantages of the conventional sociogram and its extensions (such as circle diagrams), but results in something much closer to a ‘map’ of the space in which the network is embedded. This is a very im- portant advance.” For the present analysis, we have used the non-metric multi-dimensional scaling technique called “smallest space analysis,” which uses asymmetrical adjacency matrix of similarities and dissimilarities among the targets. (See Kruskal & Wish,1978; Scott, 1991 for a discussion of advan- tages over metric MDS). The data have been recoded to binary form, so that
indicates person X has had no prior contact with person Y and 1 indicates
that X and Y have had some contact, i.e. at least “every few months.” The non-metric MDS program is able to produce a matrix of Euclidean distances (based on rank orders) which is used to create a metric scatter plot. These plots are displayed as the two-dimensional figures below.
The output of MDS is a spatial display of points, where each point represents a target person in the network. The configuration of points should inform us about the pattern of affiliations and contacts in the network. The smaller the distance between two points, the greater the similarity between these two in- dividuals with respect to their social contacts. The location of person X in multidimensional space is determined both by X’s own social connections and by the connections of those who have chosen X as an affiliate. The MDS analyses were performed using SPSS Windows 6.1.
Technically, the data could be analyzed at either the individual or organiza- tional level and each approach has some advantages. At this time, we have
BOTEC Analysis Corporation