Federal Networks After Disasters; More Lessons for the Budding Social Network Analyst

Many posts this summer have focused on issues and ideas related to modeling extreme events. Additionally, several posts have focused on network analysis and some of the most common misuses of these methods. Today, we find an article that fits well into both categories.

Recently, I discussed the idea of modeling emergency response to catastrophic events as a non-cooperative game played among responders attempting to coordinate at disaster sites. Clearly, however, emergency response can also be modeled as a dynamic cooperative system wherein entire governments must coordinate to reconfigure themselves in response to large-scale changes caused by catastrophes. An article appearing in the latest issue of the BEP Journal of Homeland Security and Emergency Management attempts such a model by examining the evolution of federal interagency networks after disaster events. In “Interorganizational Coordination in Complex Environments of Disasters: The Evolution of Intergovernmental Disaster Response Systems,” author Naim Kapucu examines disaster response as a complex adaptive system (CAS) and analyzes how the creation of the Federal Response Plan (FRP), the National Response Plan (NRP), and the National Response Framework (NRF) altered the relationships between federal agencies. As Kapucu states:

Disasters are continuously evolving complex processes that involve many evolving relationships…The concept of CAS captures the processes of change in complex environments in which a set of interdependent units which is capable of reallocating its resources and actions to achieve a stated goal under changing conditions. As units within a system change their relationships to one another, the system as a whole changes its relationship to the environment in which it operates.

The implementation of this idea, however, departs from a CAS model to one that focuses almost exclusively on static network analysis of data derived from documents detailing the interorganizational relationships established by the FRP, NRP, and NRF. In relying on network analysis techniques to examine how these macro-relationships the author makes several serious errors in his application of the methods. First, he relies on a single metric, degree centrality, to determine the most central actors in the data. No one metric should ever be used to make such a determination, and as such, degree centrality is particularly poorly suited for this task given the equal weight it gives to pendant edges.

More egregious, however, is the blending of actor type within the data, and the application of social network techniques to non-soical data. The author states, “In the “formal” network representation, each node, besides the department or agency, represents an emergency support function (ESF). Each node represents a mini-network of agencies in a specific ESF.” That is at once confusing and seemingly problematic. By collapsing differing and hierarchically clustered actor types—departments, agencies and ESF—the analysis mistakenly treats these actors as existing in a single mode, or layer, which can result is misleading or outright incorrect conclusions. In a simpler context, imagine some data representing a network of people attending different meetings. People are connected to meetings that they attend, and we run degree centrality on this network to find meeting X is the most central actors. We are interested, however, in how these meetings affect the potential relationship among attendees; therefore, we must first transform the data to represent these ties rather than the ties to meetings by creating an affiliations matrix.

The author also makes a mistake all too common in the network analysis literature: applying methods designed for social network data to data that does not represent social ties. In this case, the author is clearly dealing with non-social data; connections among federal agencies and departments. When discussing the growth of the network, however, the author states, “In any network there are (k * k-1) unique ordered pairs of actors, where k is the number of actors. The number of
logically possible relationships then grows exponentially as the number of actors increases linearly…The population of actors “have structure – interaction patterns- that determine which parts of actors are likely to interact and which parts are unlikely to do so” (Axelrod and Cohen 1999, 6).” These properties are only applicable to social networks, and there is no theory (to my knowledge) that defines this relationship for institutional networks. In network analysis, as in any complex statistical methods, it is critical that the methods being implemented fit the data.

My interest in modeling extreme events is focused more on micro-dyanmics, however, examinations of macro changes such as this are extremely important—especially given the political and institutional goals of many terrorist organizations. The CAS model may be a useful method for understanding this, and should be further explored. That said, greater care must be taken when attempting to use network analysis within this model to discover how these networks grow and address extreme events.

Photo: Wikimedia

Automatically Generated Related posts:

1. Structurally Induced Random Graph Model of Social Networks
2. Networks and ‘Implication for Network Centric Warfare’
3. Social Media Monitoring During a Domestic ‘Crisis’
4. Network Analysis in the Social Sciences
5. Deluge of New Networks Literature