SOCNET is a service of INSNA, the professional association for social
network researchers (http://www.insna.org). To unsubscribe, send
an email message to
First, let me once again thank everyone who responded to my earlier messages on the subject of “Network Chemistry.” It has been an education, still underway, to work through the materials recommended in those replies. Still, however, even the most sophisticated analysis of organizational networks fail, it seems to me, to address theoretically or substantively the issue to which one respondent pointed privately — even theories of how network properties affect relationships assume social actors free to choose those relationships. They do not address “forced relationships, “ i.e., the fact that in large organizations individuals may be assigned to project teams regardless of how they feel about the other members in them. The bosses who make these assignments may wisely consider how well or badly people have worked together before or whether personal differences are likely to affect their joint performance. But other considerations, availability and ability to provide necessary skills, even things as elusive as perceived “leadership” or “creativity” are given higher priority in assembling teams.
In the advertising industry from which my data are taken, creative teams require people who can both work effectively together as peers in coming up with compelling ideas and provide specific skills required for their implementation. Historically, the minimal team was a dyad, a designer (who produced the visual) and a copywriter (who produced the words), who together created a one-off print ad. An art director might be added to coordinate their efforts and mediate their inevitable quarrels. As ads moved from idea to execution, production staff might be added to the team, an illustrator or, more commonly since the 1930s or so, a photographer to shoot the photograph. If the photograph included models, stylists to dress them and hair and make-up artists might also be required. If an ad were part of a larger campaign, which might include multiple ads in multiple media, a creative director would be in charge of the project as a whole. The advent of TV required larger teams, including producers, film directors, and cameramen/cinematographers. Staging, lighting, film editing, narration, mixing (combining audio with visual imagery) and special effects all required individuals with specific expertise. In sum, relationships in project teams are always at least two-sided. On the one side there are relationships between individual actors who, at least for a team’s duration, form a clique. On the other there are the specific constellations of roles brought together to form a team to achieve a specific purpose. And the roles are prior to the individuals in the processes of forming teams.
These considerations point to a series of questions about my data, related to how projects and roles interact to affect team composition. So far I have considered three approaches to analyzing the effects in question. The starting point for all three is 2-mode multiplex networks in which Creators (the actors) are related to Ads (the events) by Roles (the relationships that connect Creators to Ads. Partitions define Ad attributes; Ads can be broken down into discrete categories by Agency, Medium, and Industry Category. Creators may have more than one Role relationship to an Ad. The same individual may, for example, be both the copywriter and the creative director. Roles are coded by assigning nominal values to lines. We might ask, then,
- Are there significant differences in the team size and role composition depending on Ad attributes? To find out we can extract subnetworks defined by attributes and examine the distributions of roles within them.
- Are there significant differences in the Attributes of ads linked by specified Role relationships? To find out we can convert line values to relations and then, using Pajek, either
- Observe the effects on network properties of removing one or more specified relationships, or
- Extract the network defined by one or more specified relationships and note differences in the distribution of Ad attributes.
- Another possibility is to create new partitions, generalized cores or Louvain communities, for example. Then extract subnetworks and proceed to 1 and 2 for the subnetworks in question.
- Yet another is to extract the 2-mode networks described in 1-3, convert them to 1-mode networks and see what differences emerge.
My question for my colleagues here is, as always, what am I missing here? Can you suggest measures, procedures, or concepts that would improve the plans sketched here in 1-4? Your help will be deeply appreciated.