Responder Communication Networks in the World Trade Center Disaster: Implications for Modeling of Communication Within Emergency Settings

Abstract

Using archival materials obtained from the Port Authority of New York and New Jersey, we analyze networks of communication and interaction among responders to the World Trade Center disaster. Our findings indicate substantial variability in individual radio communication system usage, with both communication volume and number of partners exhibiting distributions with long upper tails. Responder communication patterns are well-described by a fairly simple four-role structure and exhibit substantial similarity across responder groups (both specialist and non-specialist). Occupancy of coordinating roles is influenced by formal institutional status, but the vast majority of hub role occupancy appears to be emergent in character. Examination of both radio transcripts and police reports suggests that much of the communication among WTC responders is centered on problems of spatial reasoning and peer location, possibly providing an explanation for the importance of improvised coordination at the event site. Although these problems appear to have posed substantial challenges for responders at Ground Zero, we find the global communication/interaction network among Port Authority officers to be fairly well-connected, with little evidence of large-scale fragmentation (despite perceptions to the contrary). Implications of these findings for the modeling of communication networks in emergency settings are discussed.

Keywords:

• disaster
• emergency response
• interpersonal communication
• simulation design
• social networks
• World Trade Center

This material is based upon work supported by the National Science Foundation under awards NSF ITR #0331707 and #0331690.

The authors would like to thank Kathleen Tierney, Jeannette Sutton, Katie Faust, and Christine Bevc for their input and encouragement.

Notes

¹Note that individual communication volume (number of messages sent/received) is strongly correlated with number of partners for all networks in this sample (median correlation 0.96, IQR 0.064). Thus, the dichotomized networks are a fairly good proxy for the total communication networks.

²Each channel in the PA data set appears to have been used by the members of a single organization or sub-organization. In addition, there is little evidence of the use of multiple channels by responders within the PA data set. These observations are consistent with similar findings by the 9/11 Commission (National Commission on Terrorist Attacks Upon the United States, 2004).

³The Yule pmf is given by Pr(X = x) = (ρ − 1)Γ(x)Γ(ρ)/Γ(ρ + x)), for x ∊ 1, 2,… For large x, Pr(X = x) ≈ k x ^−ρ; thus, the Yule is a natural model for discrete power law phenomena.

⁴The Bayesian Information Criterion (BIC) for a given model in the present context is equal to − 2In Inn, where is the maximized likelihood, n is the number of data points, and k is the number of model parameters. The minimum-BIC model is (asymptotically) that favored by the Bayes factor under a sufficiently diffuse prior; see, e.g., Wasserman (2000).

⁵Indeed, no moments of the Yule distribution exist for p < 2, implying infinite means and variances for the fitted models. Such extreme extrapolations are unwarranted for physical reasons, but do support the qualitative observation of intense activity by certain individuals.

⁶Monte Carlo χ² (compensating for small cell size) yields p < 0.0005.

⁷Formally, the betweenness of a vertex υ is defined as , where g(i, j) is the number of geodesics from i to j, and g(i, j|k) is the number of i, j geodesics containing k.

⁸The Krackhardt connectedness (Krackhardt, 1994) of a graph is the fraction of dyads which are semi-path connected. Here, we take the local Krackhardt connectedness of an informant's observed network to be equal to the connectedness of his or her reported network, after removing him or her.

⁹lt is possible, of course, that the apparent (weak) effect of institutional status on occupancy of coordinative roles is itself spurious: actors whose formal positions entail coordination may differ from their peers in other respects, and this unmeasured heterogeneity may drive the underlying process. We cannot answer this question with existing data, but note that such possibilities might be explored via simulation.