The network governance of crisis response: case studies of incident command systems.

INTRODUCTION A crisis occurs. How to respond? The crisis could be like the Oklahoma City bombing in 1995. Government and nonprofit responders were on the scene minutes after a massive truck bomb destroyed a federal building. Using a management structure called the Incident Command System (ICS), one person was appointed as incident commander, responsiblefor directing all other responders. An after-action report argued that the "Oklahoma City Bombing should be viewed as ultimate proof thatthe Incident Command System works" (ODCEM n.d., 36).

As a result of its perceived success in situations like Oklahoma, the ICS has been mandated by the Department of Homeland Security (DHS) for all crisis situations. The DHS characterizes the ICS as a command and control style of management, emphasizing the importance of a clear hierarchy of authority. But if we look closely, the portrayal ofthe ICS as a hierarchy is misleading. The incident commander at the Oklahoma City bombing was a local government fire chief. He was directing not only his own employees but also responders from other organizations, other levels of government, and the nonprofit and private sectors. Numerous interdependent organizations were working together toward a common goal--these are the definitional characteristics of a network (Hall and O'Toole 2000; Provan and Kenis 2008, 231). This article examines the ICS and, in doing so, addresses two research questions. First, how do we categorize the ICS as a structural form? Crises generally demand the capacity of multiple organizations working together, and crisis response therefore reflects a network structural form. But the ICS is used to coordinate multiple response organizations under a temporary central authority with a hierarchical structure. Any crisis response using the ICS therefore reflects an intriguing mixture of network and hierarchy. This first research questionis addressed by examining the evolution of the ICS, describing its characteristics, and by reviewing competing perspectives on the ICS. Although previous treatments of the ICS focus on its hierarchical aspects, this article suggests that it is better understood as a highly centralized mode of network governance, designed to coordinate interdependent responders under urgent conditions. A network governance perspective leads to a second research question: How do network characteristics influence ICS operations? Answering this question is the primary theoretical contribution of the article and is done by examining the ICS in practice in multiple settings. Case evidence suggests three specific ways in which the network aspects of crisis response affected ICS operations. First, coordination became more difficult as the number and range of responding organizations increased. Network members brought their organizational views to the ICS, fostering tension between members. The ICS especially struggled to incorporate new members who represented the emergent aspects ofthe network. Second, the ICS assumes a clear line of command, but the cases illustrate the shared nature of authority in crisis response networks. The question of who is in charge can be a contentious one, and the legitimacy of command is negotiated among network members. Third, the cases illustrate how network values in the form of working relationships and trust shaped the operation of ICS. The final section discusses the case evidence and emerging theory.What are the implications for managing crises? What are implicationsfor network theory more broadly? The particular history and nature of the ICS also makes it a source of unusual insight for issues that network theorists have identified as important (Provan, Fish, and Sydow 2007; Provan and Kenis 2008). THE EVOLUTION OF THE ICS Provan and Kenis (2008) note that there has been little research on how network governance forms evolve, but hypothesize that the evolution of such forms will follow a functional logic, with change prompted by the search for greater effectiveness. The evolution of the ICS offers an example of how network governance forms are created and diffused. This history follows three stages: a functional origin, a voluntary adoption, and a mandatory diffusion. The ICS began as a calculated effort to respond to a specific coordination problem. Following a series of California wildfires in 1970,local, state, and federal agencies came together to discover how to better integrate their efforts by developing a common language, management concepts, and communications. Centralization of authority normally occurs within organizations and on a relatively stable basis. Thecritical innovation of the ICS was to temporarily centralize response authority to direct multiple organizations. The incident commander is responsible for directing and coordinating the tactical efforts ofthe organizations involved during the response. In the years that followed its creation, the ICS was perceived by practitioners as successful in reducing coordination problems and improving fire response effectiveness (Bigley and Roberts 2001; Buck, Trainor, and Aguirre 2006; Cole 2000; Moynihan 2008). As its reputationgrew, crisis responders outside California began to use the ICS not only to fight forest fires but also for other tasks, such as hazardous material cleanups, earthquakes, and floods. This second stage of the evolution of the ICS saw it applied outside its original environment, creating the danger of a suboptimal matching between governance structure and task. However, it is important to note that crisis responders were voluntarily adopting the ICS, at least in part because theyperceived it as a tool to solve the problem of interorganizational coordination common to most crises. The aftermath of9/11 led to the third phase of the evolution of the ICS, when it became required of all federal crisis responders and all state and local responders receiving federal funding. In 2004, theDHS released two closely related policy statements intended to shapethe response to a wide range of domestic emergencies. The National Incident Management System and the National Response Plan represented an effort to nationalize crisis management policy in unprecedented ways, with the ICS as a central feature (US DHS 2004; FEMA 2007). A NETWORK GOVERNANCE PERSPECTIVE ON CRISIS RESPONSE The emergency management literature features conflicting perspectives on crisis response (Drabek and McEntire 2003; Trainor 2004). A command and control model, often presented by practitioners, champions a hierarchical approach. By contrast, a coordination and communication model argues that crisis response inevitably depends on collaborative processes to succeed. This section presents both views and then bridges them using a network governance perspective. Crisis policy documents present the ICS as useful because of the benefits that hierarchy or centralization provide in coordination amidcrisis. The management characteristics that the DHS associates with the ICS are that of an organization with clear command structures, centralized planning, communications and accountability procedures, anda limited span of control (US DHS, 2004, 9-12; FEMA 2007, 45-7, see also Moynihan 2005, 8-9). The only concession to the networked natureof crisis response is the acknowledgement that a unified command maybe appropriate if the incident crosses the jurisdiction of multiple organizations. Practitioners also emphasize the hierarchical nature of the ICS. California incident commanders include among the most prominent strengths of the ICS as the hierarchical chain of command, the use of centralized plans, and the limited span of control (Cole 2000). The response community has largely rejected scholarly critiques of the ICS and "has been almost universal in its praise of ICS" (Buck, Trainor, and Aguirre 2006, 3). The continuing attraction of the ICS for policymakers is demonstrated by the fact that in the midst of policy revisions in the aftermath of Hurricane Katrina, the role and nature of the ICSremained essentially unchanged (FEMA 2007). Scholarship on crisis response argues that centralization is unrelated or even destructive toward actual response capacity and driven by a desire for political control (Waugh and Streib 2006). In particular, the ICS has been criticized for ignoring the importance of interorganizational relationships, the spontaneous nature of response, the role of unorganized volunteers, and the potential for conflict between organizations (Drabek and McEntire 2003; Waugh and Streib 2006). The coordination and communication perspective argues that any large-scale crisis inevitability requires an intersectoral and cross-jurisdictional capacities (Boin and 't Hart 2003), collaborative skills (Dynes 1970), and a decentralized flexibility to deal with the ambiguity and turbulence of crisis situations (Tierney and Trainor 2004, 164). Both the command and control view, and the coordination and communication approach, suffer limitations. By focusing on the hierarchicalproperties of the ICS, policymakers neglect how network factors shape its operations. The intergovernmental nature of crisis tends to fragment authority (Tierney and Trainor 2004, 163-4). Even though organizations fall under a command during the operation of a crisis, this authority does not have the qualities of a strong hierarchy, enjoying only temporary and partial control over members that retain significant autonomy (Leonard and Howitt 2006, 11). The continuing practitioner preference for the ICS suggests a functional value not acknowledged by its critics (Moynihan 2008). The centralization imperative for crisis response is urgency. When combined with the need for a network of interdependent responders, urgency creates a coordination problem. Gradual processes of interorganizationalconsensus building and mutual adjustment take too long. Responders need a central coordinating mechanism to direct resources and resolve conflict in a timely fashion. In short, the problem with the command and control view of response is that it fails to incorporate relevant criticisms about the need to foster collaboration among relatively autonomous actors, but the coordination and communication literature fails to acknowledge the imperative for some form of centralized direction. Both literatures characterize the ICS as a hierarchical means to attempt to control crisisresponse, although disagreeing on the utility of such an approach. Anetwork governance approach challenges these competing perspectives by recasting the ICS as a formal mechanism to foster network coordination. This approach recognizes that network can be governed via centralized mechanisms (McGuire 2006), while also acknowledging the complexities created by the network setting. Of existing models of network governance, the ICS most closely resembles the network administrative organization (NAO). (1) Consistent with the description of the NAO, the incident command does not exist independent of the network, and its exclusive purpose is network governance. NAOs are best suited for networks that demand a wide variety of competencies to be managed (Provan and Kenis 2008). Crisis response requires an array of interdependent competencies, and it is the need to rapidly integrate these competencies that gave rise to and continues to provide the compelling logic for the ICS. The contrast between a network governance and hierarchical view ofthe ICS is illustrated in figure 1. The left-hand side of the figurerepresents the dominant view of the ICS (US DHS 2004, 13). In this figure, a hierarchy allows the incident commander to direct the crisisfunctions of logistics, operations, planning, and finance/administration. But if we consider the ICS in terms of its members, we see it as a network, albeit a highly centralized one. The incident commander is at the center of the network, surrounded by organizations that have ongoing intercrisis dyadic relationships, as illustrated by the right-hand side of figure 1. [FIGURE 1 OMITTED] It is now possible to answer the first research question: How do we categorize the ICS as a structural form? The ICS is a mode of network governance, akin to an NAO, and designed to centralize authority in crisis response networks. The ability of the ICS to foster coordination depends upon the characteristics of the network it manages. The following sections take up this issue. First, the data and methods used to analyze the ICS are discussed. DATA AND METHODS Case Selection For qualitative research, cases are generally not selected at random but to serve a theoretical purpose (Brower, Abolfia, and Carr 2000, 368; Eisenhardt and Graebner 2007, 27). This article does not seek to explain the ultimate success or failure of crisis response outcomes but to understand the operation of the ICS in different settings, informing an emerging theory about how the network aspects of crisis response affect the ability of centralized forms of network governanceto foster coordination. Interorganizational coordination is treated as a success benchmark because the ICS was designed to meet this goal, on the assumption that a better coordinated crisis response will bemore responsive and effective. This approach is consistent with Winter's (2003) recommendation that when policy outcome causality is difficult to determine, implementation should be judged by intermediate operational outputs represented by delivery behavior. The primary criterion in case selection was diversity of task setting. This is particularly relevant for the conceptual unit of analysis, the ICS, since there remains a significant lack of research about how it operates outside its original setting of forest fires, even asthe DHS mandates its use for all types of crises (Bigley and Roberts2001, 1295; Cole 2000, 225). Where there is limited research on a topic, case studies and an inductive approach are especially useful in identifying how casual connections actually occur (Brower, Abolafia, and Carr 2000, 367; George and Bennett 2004). A multicase approach helps to ground the evidence in a variety of empirical settings, guarding against a danger that findings are tied to a unique aspect of a single case, or for studies of the ICS, the particular function of fighting forest fires. One shortcoming of focusing on multiple cases is that the explication of theory comes at the expense of descriptively rich case narratives. To mediate this problem, case evidence is presented in distinct sections that inform different theoretical propositions, and a table is employed to summarize how findings are relevant totheory (Eisenhardt and Graebner 2007; Miles and Huberman 1994). Another selection criterion was adequate documentation to pursue the content analysis method. Cases were only selected if multiple in-depth sources existed, and it was possible to reassemble a detailed descriptive account of the event. (2) Triangulating sources and establishing a detailed descriptive base provides some reassurance that interpretations are not shaped by an idiosyncratic account of the event. Content Analysis The nature of crises makes it difficult to develop and apply ex-ante research designs. So how can a researcher capture a comparable amount and range of information across multiple cases to examine concepts of interest? There is no perfect solution, but the research approach pursued here represents a systematic and transparent effort to dealwith this problem. The research method employed is to code textual information, primarily from secondary sources, on each of the crises. This is a feasible strategy because crises generate significant public documentation. This documentation is descriptively rich and draws on resources--including hundreds of interviews and access to otherwise unavailable documents--that few research teams could match. A total of 61,995 text units, the equivalent of 3,855 pages of reports, testimony, and interview transcripts, were coded (or about 16 text units per page). How was the coding instrument developed and applied? The author coded text documents using a qualitative software package called QSR N6. The software enables the analyst to allocate specific chunks of text to one or more of dozens of possible thematic codes. This allows for systematic coding of relevant variables and comparison of these variables across cases. The coding process represents a mixture of inductive and deductive analysis--new codes can be added as suggested by the data, and the interpretation of the code can be modified in accompanying memos (Miles and Huberman 1994, 58-62). The process of coding shapes the relevance, meaning, and interconnection of concepts in a way that simply reading the text does not. The theory emerges via "recursive cycling among the case data, emerging theory, and later, extant literature" (Eisenhardt and Graebner 2007, 25). In qualitative research, the researcher usually has a good deal ofdiscretion in deciding which factors to focus on, and so it is worthexplaining the selection of the three primary concepts examined in this article--network diversity, shared authority and trust--as opposed to a myriad of alternatives. Each of the concepts examined satisfied three conditions. The first was that the concept frequently reoccurred in the data and appeared to be important in understanding the case outcomes. Second, the concepts were largely overlooked in previous descriptions of the ICS. Identifying and elaborating the concepts therefore filled a knowledge gap. Third, the empirical phenomena appeared to "fit" with a series of codes relevant to understanding network operations. After the initial examination of the first case, it becameclear that network theory could provide a useful theoretical framework, and the coding instrument was modified accordingly. The interpretation and importance of network concepts would evolve significantly as they were shaped by the case data. Content analysis using computer software offers some advantages over a less structured approach (Miles and Huberman 1994). First, it isthe only manageable way to organize large amounts of qualitative data. Second, it challenges the researcher to be systematic in identifying the presence, frequency, and nature of a concept. Third, it provides some check on the reliability of the data, with transparency, rather than replication, as the goal. (3) CASE SUMMARIES Wildland-Urban Fires, 1993 Laguna Fire, and 2003 Cedar Fire: Wildland-urban fires are larger than traditional forest fires and an increasing threat. The urban setting means that more lives are at risk, tasks such as evacuation become more complex, and a wider network of responders are involved. The Laguna Fire burned from October 26 to November 4, 1993, affecting the cities of Laguna Beach, Irvine, and Newport Beach, the community of Emerald Bay, and the surrounding unincorporated area. In total, 441 homes were destroyed, 14,337 acres burned, and $528 million in damage was caused. Although more than 26,000 people were evacuated and many were injured, no deaths resulted from the fire. A unified command between the Orange County Fire Department andthe Laguna Beach Fire Department was established to contain the fire. A decade later, the Cedar Fire also burned a significant portion of Southern California, damaging 335 structures, burning 193,646 acres, and causing $204 million in damage. Shortly before 6 p.m. on October 25, 2003, the fire originated from the Cleveland National Forest, near Julian, California. The United States Fire Service established the initial command, but as the fire moved beyond its federal jurisdiction, the California Department of Forestry and Fire Protection activated an incident management team. The fire entered the City of San Diego by the morning of October 26, thereby involving the San Diego FireDepartment. The 1995 Oklahoma City Bombing: At 9.02 a.m., April 19, 1995, a rented Ryder truck containing 4,800 pounds of explosives detonated beside the Alfred P. Murrah Federal Building at Oklahoma City. The massive explosion destroyed about one-third of a building containing about 600 workers and 250 visitors. The attack killed 168 people, and 426 were treated for injuries in local hospitals. The response began immediately. The Red Cross was on site within 7 minutes of the blast and had more volunteers than it could handle within a half hour. By this point, a Federal Bureau of Investigation (FBI) agency representative was also on site, and the State Emergency Operation Center was fully operational. The Chief of the Oklahoma City Fire Department, Gary Marrs, quickly established the incident command. The 2001 Attack on the Pentagon: At 9.38 a.m. on September 11, 2001, American Airlines Flight 77 crashed into the Pentagon, killing thecrew of 6, 58 passengers, and 125 occupants of the Pentagon. Responders quickly arrived on the site, contained the fire, rescued surviving occupants, and provided immediate medical treatment. James Schwartz, the Assistant Chief for Operations of the Arlington County Fire Department, was the incident commander. He gradually expanded the command into a unified command by including other agencies. The response tothis event has been characterized as a success by th e 9/11 Commission (2004, 314), which recommended the widespread use of the ICS. An after-action report also emphasized the positive role of the ICS: "The primary response participants understood the ICS, implemented it effectively, and complied with its provisions" (Titan Systems Corporation2002, Introduction-11). Exotic Newcastle disease 2002-2003: Exotic Newcastle disease (END)is a highly contagious bird disease, with similar symptoms, modes oftransmission, and rates of fatality as avian flu. An outbreak was confirmed in California on October 1, 2002, and was later found in Arizona, Nevada, and Texas. Quarantines were also placed in Colorado and New Mexico. A task force was created to eradicate the disease, directed by an incident command. As END spread, additional incident commands were created, under the direction of a single area command, jointlydirected by state and federal government veterinarians. Task force teams visited private residences and commercial bird premises. In 932 premises, an infection was suspected, the value of the birds were appraised, the birds were euthanized (4.5 million in total), and premises were cleaned and disinfected. The task force eliminated END and limited its impact on the poultry industry, and by September 16, 2003, final quarantine restrictions were removed. Hurricane Katrina 2005: By almost any measure, the response to Hurricane Katrina was a failure. Over 1,800 people died, and tens of thousands were left without basic supplies. Responders were warned aboutthe potential effects days before Katrina made landfall at 6.10 a.m.on Monday, August 29. Federal Emergency Management Agency (FEMA) andstate emergency responders began 24-hour operations on Saturday. TheMayor of New Orleans ordered a mandatory evacuation by 9.30 a.m. on Sunday and opened the Superdome as a refuge of last resort. Late Monday morning levees began to overtop and breach, causing massive flooding. Search and rescue operations began that afternoon, but communications also started to fail. On Thursday, buses arrived to begin evacuate shelters of last resort. Evacuations were not completed until Saturday, and some remained stranded on highways until a week after landfall. Three major operational commands were established during Katrina, but there was no unified command that took charge of the entire response operation (House Report 2006, 189). First, the Federal Coordinating Officer, the official federal response commander, formed a Joint Field Office with the state coordinating officer. Second, the Principal Federal Official was treated by many DHS officials as if it had command authority, even though it was created by the National Response Plan to provide ground information to the DHS rather than offer operational capacity. FEMA head Michael Brown initially filled this role. Third, Joint Task Force Katrina, led by General Russel L. Honorr, directed Department of Defense forces. The next sections explore the role of network factors in the cases, summarized in table 1. In each section, the concept is introduced, and selected case examples are used to illustrate and elaborate the concept. More case detail can be found in Moynihan (2005, 2007, 2008). Network Diversity Perhaps the most straightforward complication that a network setting brings for the ICS is the number and diversity of organizations involved. As crises increase in size and complexity, they require greater capacities, which imply a larger and more diverse network of responders. This increases the heterogeneity of backgrounds, beliefs, and interests of network members, which, in turn, creates a greater coordination burden than faced by small homogenous networks (Provan and Milward 2001, 418; Scharpf 1993, 201). As a network manager, incident commanders devote significant energy to meeting the concerns of members and face a more challenging task as the number and variety of theseconcerns increase (Agranoff and McGuire 2001). The wildland-urban fire responses featured a relatively small and homogenous group of actors with similar backgrounds. This facilitatedease of coordination, but the cases do suggest that coordination difficulties occurred when responders incorporated unfamiliar agencies, especially those with limited or no background in the ICS. Agencies that lack ICS training are perceived as being less effective partners,and higher levels of government often fail to incorporate local responders into the ICS because of perceived capacity problems (MCS 2003). In the Oklahoma and Pentagon cases, responders could focus on a limited set of tasks in a specific area. Even so, thousands of responders were involved, complicating efforts to secure the incident perimeter. Both incident commanders especially struggled to incorporate and direct the extended network that developed as volunteers arrived, andlarge amounts of unsolicited services and resources were offered. There were no standard procedures to store, track, and manage materialssuch as donated rescue materials, food, supplies, clothing, and financial donations. The enormous scope of the Katrina disaster led to a response network so diverse that there was a failure to fully comprehend which actors were actually part of the network (partly because of a large voluntary component), the skills they offered, and how to use these capacities (House Report 2006, 302). Over 500 organizations have been identified in the Katrina network (L. Comfort, unpublished data). The sheer number of tasks led to the creation of many task-specific networks within the broader response network, dealing with goals such as evacuation, delivering materials, recovering bodies and providing mortuaryservices, medical services, public safety, restoring communications and power, search and rescue, and temporary shelter. There were basicproblems in coordination both within and across these networks, disagreements about what to do and who was to do it, and many examples ofindividual organizations operating as solo actors rather than in coordination with others (Moynihan 2007, 33). Although only in the Katrina case did the size of the network outstrip the control of the officials trying to direct it, network diversity posed two other problems across the cases. First, the inclusion of multiple agencies with distinct backgrounds and cultures created uncertainty about how members would behave and interact with one another (Koppenjan and Klijn 2004). The intent of the ICS was to help to overcome such integration problems by offering a standard framework andcommon language for all participating agencies. Bigley and Roberts (2001) argue that a key component of ICS effectiveness is the ability to foster shared mental models among responders to encourage consistency in behavior and integration of actions under the trying conditions of crisis. But building common cognitive frameworks is challenging in networks where participants bring the perspective of their home organization, profession, or training, which may clash with the perspectives of other network members. The cases illustrate not just differences in perspectives among network members but varying levels of understanding of the ICS. This posed a significant barrier to coordinatedaction, as a network will struggle when some members do not understand the primary form of network governance. The second impact of network diversity arises from the emergent nature of crisis response, and the difficulty of incorporating new members once a crisis begins. Most governmental actors involved in a crisis have formal responsibilities, specified during crisis preparation.However, emergent members are typically nonprofit and private actorswho are largely unknown to planners ahead of time or not considered important enough to include in plans (Stallings and Quarantelli 1985). Emergent actors hover on the edge of the network and can form theirown ad hoc network (Tierney and Trainor 2004). Even within well-established networks, boundaries are difficult to define, as is determining who is "in" and who is "out" (Laumann, Marsden, and Prensky, 1983;Raab 2002). This problem is exacerbated in crisis response situations, which encourage a flood of potential members and where many first responders are private citizens. The integration of emergent members was a consistent difficulty across all the cases. It is impossible to fully foretell the range of capacities offered or needed ahead of time, and so the ICS inevitably incorporates new members as the crisis occurs. The ICS may seek out such actors to gain needed resources, or emergent actors may succeed in joining the formal network through a connection with the incident commander or core member. But centralized authority may also operate in opposition to spontaneous emergence, dampening, or ignoring the capacities of nonpublic actors (Drabek and McEntire 2003). Overwhelmed during a crisis and unfamiliar with emergent actors, incident commanders lack the time to learn what emergent capacities are on offer. Emergent actors may not understand how to direct their efforts. Whether formally part of the ICS or not, emergent actors are relevant to the crisis network if they are devoting resources to achieve network goalsand acting in ways that complement, distract, or conflict with network actions. The evidence on network diversity suggests the first of three propositions made in this article: PROPOSITION 1 : Even with centralized network governance, network diversity makes crisis response coordination more difficult. SHARED AUTHORITY Previous analyses of the ICS have emphasized the importance of central command, and indeed, it is a definitional prerequisite of an ICS. The DHS (2004) describes the importance of clear lines of authority, whereas Bigley and Roberts (2001, 1296) point to the "compelling authority system" of the ICS and the role of the incident commander as the final arbiter of disputes. However, by failing to recognize the network elements of the crisis response, previous discussions underestimate the difficulties in establishing and operating a central command. Network research treats authority as a shared commodity (Brass et al. 2004). This is sometimes assumed to mean that there is equality among members. But it can simply mean that authority is dispersed enough to that multiple actors have enough autonomy to disrupt coordination if they wish to. In public networks, it is often the case that onegovernmental actor is more powerful than others. Even so, research has noted the political and contested nature of such authority (Agranoff and McGuire 2001, 315; Raab 2002, 619). The dynamics of coordination change as authority becomes dispersed. Even powerful network members cannot simply assert their authority,but must to some degree negotiate its terms, and establish why theirrole is legitimate. In utilizing this authority, they cannot issue orders clearly contrary to the interests of network members, but rely more on facilitation. Case evidence supports the view that shared authority is subject to ambiguity and disagreement. In the majority of cases, the crucial questions of who was in charge and how authority was transferred were a source of contention and negotiation between members. In the Oklahoma City and Pentagon cases, locally based incident commanders quickly established and maintained a command presence to avoid a federal usurpation of local control. At Oklahoma, FEMA initiallywanted to take over, but local responders refused to cede control. In the Pentagon case, incident commander Schwartz felt it necessary toconvene the principal organizations for a meeting on the evening of 9/11 to explain the basis for his authority. He recalled: "I do fullybelieve that had there been a gap in that command presence, FEMA, and perhaps other federal agencies, would have driven a truck through it" (Varley 2003, 6). Both cases illustrate the ambiguity of jurisdictional claims. The traditional response system is bottom--up, providing local governments with jurisdictional authority until they become overwhelmed. But because Oklahoma and the Pentagon involved a crime scene as well as a disaster, the incident commanders had to share authority with the FBI.This balance of power worked because the FBI proved willing to deferto the incident commanders on issues of search and rescue, whereas the incident commanders were sensitive to the needs of crime scene investigation. Another complication was that the Stafford Act of 1988 provided a legitimate legal argument for complete federal control sinceboth attacks occurred on federal property. Both cases also illustrate the contested and ultimately shared nature of authority. Marrs succeeded in maintaining sole control of the Oklahoma incident, although even with the authority of an incident commander he found that he still had to respond to the needs of the various network members. Although in a very similar situation to Marrs, Schwartz moved to a unified command where multiple organizations had an input into decisions and even invited FEMA to participate, reasoning: "I knew I wanted to know where FEMA was all the time, and I figured the best way to do that, as well as get their expertise, was to have them up there with me in the command post. I was just looking for practical solutions" (Varley 2003, 19). The fluid and contested nature of authority never gave way to a clear system of network governance in Hurricane Katrina, leading to duplicative and uncoordinated efforts (House Report 2006, 194-5). Efforts to foster unified command with state and local officials faltered partly because much of the local government emergency infrastructure was destroyed. It did not help that there were three separate federal commands (described in the case summary). The lack of clarity about who was in charge gave rise to responders "'freelancing,' or just showing up without coordinating with the appropriate authorities at FEMA or the state. They would bypass the command structure" (House Report 2006, 189). The failure to establish unified command in Katrina was also partly due to confusion with new policies outlined in the National Response Plan and the National Incident Management System. These policies introduced new roles and laid out the rules for how responders were supposed to coordinate. Lack of knowledge about these roles and rules led to coordination failures. For example, Louisiana officials brought in consultants after Katrina made landfall to provide training on theICS. Deputy Federal Coordinating Officer Scott Wells expressed his frustration: "There was no unified command under the National ResponsePlan. They didn't understand it. They had no idea ... What does it tell you when two days into a catastrophic disaster a state gets somebody in to explain ICS to them?" (Senate Report 2006, 27-15). But confusion about new policies also extended to the federal level. A large-scale exercise before Katrina revealed "a fundamental lack of understanding for the principles and protocols set forth in the National Response Plan and National Incident Management System" (Senate Report 2006, 12-10) and a particular confusion about the respective roles of the Principal Federal Official and the Federal Coordinating Officer. This confusion that would reoccur during Katrina and was resolved onlywhen the DHS appointed Admiral Thad Allen to both positions. The concept of shared authority suggests the limits of trying to exert hierarchical control over a network of responders. Simply havingthe structure of an ICS does not make clear who is in charge. The legitimacy of centralized authority is weakened if there are competing claims for control. Even when an incident commander is in place, the primary institutional loyalty of members is to their home organization rather than the temporary ICS. Members retain a significant degree of autonomy, meaning that the incident commander is not truly a commander. He can issue an order, but whether and how the order is obeyed depends upon the willingness of the network members to accept the legitimacy of his position and the specific task. The incident commanderneeds many of the skills of a network manager: to create shared norms and commitment to the network, to foster and implement agreements on roles that make the most of differential member skills, and to exploit working relationships and maintain trust (Agranoff and McGuire 2001). The complications of shared authority give rise to the followingproposition: PROPOSITION 2: Even with centralized network governance, authorityis shared among members and subject to contention, weakening crisis response coordination. WORKING RELATIONSHIPS AND TRUST A recurring theme in network research is the role of trust in fostering coordination (Brass et al. 2004). Stability of network actors enables the development of working relationships that foster trust andeffectiveness (Provan and Kenis 2008). The case evidence suggests that having a consistent group of responders in crisis preparation allows for the development of mutual familiarity and trust. Such relationships are impossible to build once a crisis occurs and crucial to fostering coordination. Emergent members, by contrast, usually lack tieswith preexisting members. Although such actors may improve network outcomes, they reduce network stability. The ICS emerged from fighting forest fires. But even in this context, interpersonal trust was viewed by responders as a necessary supplement to the ICS. The firefighters interviewed by Rohde (2002, 224-5)highlighted the "'absolute' importance that positive relationships play in credibility, assessing needs and resource allocation, and commitment to action at all levels. Trust was a factor that was many times observed to be relationship driven; had the relationship not been created prior to the fire occurrence, the demands of the fire left little time for relationship building concurrent with firefighting ... Most respondents felt that the importance of relationships could not be overstated." A study of the 2003 Southern California fires is no less adamant: "Nearly universally, respondents reported the importance of trust, developed through established personal and professional relationships with peers and cooperators. During the initial chaos of these incidents and at the times when dispatch and ICSs were overwhelmed, these relationships became the primary means by which things got done, until the system could be brought online. These networks, enabled by these relationships, were frequently the primary force behind successful operations. Respondents also reported that networks of personal relationships minimized unproductive conflict. In situations where conflict did occur--sometimes under incredibly stressful conditions--it was often resolved by leaders who sought out their counterparts for face-to-face meetings" (MCS 2003, 12). The Oklahoma case also illustrates how responders can draw from existing social capital to facilitate working relationships. Shortly after the blast, a crucial meeting between the Mayor, the police chief,the fire chief, and a senior FBI agent occurred. This was a meeting between people who knew one another personally--three of the four were regular golfing partners. Basic responsibilities were quickly assigned, and potential conflict was averted in large part because of these relationships. Trust also facilitated problem solving. The Assistant City Manager "personally knew many of the players. He was able to call them at home or reach them on a direct line, which saved criticaltime. When there was conflict, the players were forced to sit down and work it out until compromise was achieved" (MIPT 2002, 58). Personal relationships helped to incorporate emergent aspects of the network. Southwestern Bell worked successfully with the Oklahoma City incident command because its director of external affairs knew the incident commander and contacted him to offer help. Southwestern Bell ultimately provided a location for the incident command, cell phones to responders, mobile cell phone units to manage call traffic, andphones for the family assistance center. The Pentagon case further underlines the importance of personal trust and previous working relationships. The after-action report argued that "it is difficult to overstate the value of personal relationships formed and nurtured among key participants long before the Pentagon attack" (Titan Systems Corporation 2002, A-31). One piece of evidence in particular is compelling, providing something akin to a natural experiment on the importance of working relationships. Fire Departments in Virginia and Washington DC were both familiar with the ICS. On 9/11, they both received calls asking for support and instructing them to locate at a staging point by the Pentagon. Both responded quickly. However, the Virginia firefighters followed instructions and integrated themselves with the ICS, whereas the DC Fire Department essentially formed their own command, failing to coordinate with the ICS. In explaining this variance in behavior, the after-action report contrasted the strong working relationships between the Arlington County Fire Department with other Virginia firefighters to the weak relationships maintained with the DC Fire Department (Titan Systems Corporation 2002, A-25-6, A-76). Trust facilitated coordination in other ways at the Pentagon. Trust between Pentagon and Arlington County Fire Department officials ledto an agreement to allow workers to return to the building and deal with the national security crisis at hand (Varley 2003). Trust of actors in other organizations gave the incident commander the confidenceto include them into a unified command. It was "close ties developedprior to this incident" that helped the FBI and other agencies to juggle the crime scene and other tasks on the site (Titan Systems Corporation 2002, A-50). This trust was supplemented by the use of an FBI liaison to the ICS who had worked with local fire departments for theprevious 3 years and was personally known to the incident commander. Relative to the other cases, the Katrina case is characterized by weak working relationships and a deterioration of trust as the crisisworsened. Under President Clinton, the FEMA Administrator was a former state emergency manager who built strong working relationships with state and local officials. The Bush administration weakened these relationships, in part because federal political appointees lacked state emergency backgrounds, but also because there was turnover among experienced career staff and because FEMA had less to offer state governments. After it was moved into the DHS, FEMA lost grant-giving authority for preparedness and lacked resources to build relationships through planning efforts. This loss of resources limited FEMA's abilityto influence state preparation and reduced contact with state responders. As the House Report (2006, 158) noted: "Numerous officials and operators, from state and FEMA directors to local emergency managers told the same story: if members of the state and federal emergency response teams are meeting one another for the first time at the operations center, then you should not expect a well-coordinated response." The case evidence suggests that for the IC S, working relationships and trust represent an essential complement to formal uses of authority, giving rise to the following proposition: PROPOSITION 3: Even with centralized network governance, positive working relationships and trust is a critical factor in fostering crisis response coordination. The case evidence suggests not only that trust and positive prior working relationships matter but also illustrates how they matter in a number of specific ways. The examples cited above show how trust fostered cooperation and problem solving between agencies, reducing conflict over authority and policy; eased the assignment of responsibilities, as trusted actors were provided with authority, resulting in quicker decisions and actions; encouraged information sharing between the incident commander and other actors; facilitated the flow of resources to trusted network members; and helped to incorporate new actorsinto the network. DISCUSSION The Implications for Crisis Response The previous sections suggest that the ICS, though often describedin hierarchical terms, is very much affected by the characteristics of the network it tries to govern. The implications for crisis response are spelled out in the three propositions provided above. These propositions are not only consistent with a network governance perspective, acknowledging both the centralized governance form represented by the ICS, but also the importance of network characteristics. A benefit of a network governance perspective is to move the debate beyond centralized versus collaborative approaches in crisis response. Tasks that demand a combination of network and hierarchical characteristics require a governance structure that can accommodate both. A failure to recognize the interplay of these two factors will lead to a partial diagnosis of crisis response operations. For example, theDHS assumption that a centralized command is sufficient to manage a network of responders overlooks some of the managerial challenges revealed in the cases and, as a result, may miss out on the following practical implications. Clarify the Basis of Command: The cases illustrate how confusion and contention characterized the question of who is tasked with network governance. The ICS structure itself does not make this clear. The basis for command needs to be better clarified. Maintain Working Relationships between Crises: The cases show thata durable basis for trust during crises is positive precrisis working relationships. Such relationships can be deliberately cultivated. Coordination among implementers is furthered when policies are designed to foster a sense of commitment and common interest (May 2003). Modes of network interaction can be modified to foster relationships andtrust (Agranoff and McGuire 2001). In the world of crisis response, this can be done by bringing together relevant actors for simulationsand other forms of cooperation, creating and ensuring the continuityof interorganizational liaisons who act as boundary spanners, and encouraging the mobility of organizational actors within the network. Incorporate Emergent Aspects of the Network: Many emergent networkactors can be identified prior to a crisis and have liaisons included in the core network through planning and training exercises. This makes it easier for the incident command to solicit their help during a crisis. There will, inevitably, be organizations and individuals that are not included in preplanning and who wish to help in any way they can. During a crisis, the ICS can facilitate their involvement by creating one central access point where volunteers can find out how to help. Improve Training of the ICS: Proponents of the ICS have urged responders to be better trained in ICS knowledge. A network governance perspective helps to explain why this matters. Given network diversity and the distinct backgrounds of responders, it is critical to have a common language and set of management concepts to bridge these differences. The Implications for Theories of Network Governance The case evidence illustrates the relevance of a network governance perspective for understanding crisis response. But can the study ofcrisis networks, in turn, inform network theory? The ICS operates under the unusual circumstances of crises: decisional urgency, high uncertainty, and threat (Rosenthal, 't Hart, and Charles 1989). These conditions limit generalizability to other network types. But the ICS provides evidence on the governance of whole networks, a topic about which we have "only a marginal understanding ... despite their importance as a macro-level social issue. Enhancing this knowledge is clearly a challenge that researchers in all sectors must take seriously" (Provan, Fish, and Sydow 2007, 512). The ICS is also particularly suited to informing certain questions about network governance posed by Provan and Kenis (2008) and Provan, Fish, and Sydow (2007). The Potential for Maladaptative Evolution of Mandated Network Governance Forms The history and diffusion of the ICS helps to inform two questionsthat Provan, Fish, and Sydow (2007, 508) ask about the evolution of network governance mechanisms: "How do governance forms emerge, and how do they become institutionalized?" and "And how is network performance affected when a particular governance form is mandated?" The evolution of the ICS shows that it was initially designed as a functional response to a specific problem and was adopted voluntarily by responders who felt it as useful for other contexts. National crisis policy now effectively mandates the ICS, assuming that it is generally applicable to all forms of crises. In truth, there is little empirical evidence as to whether this assumption is accurate. The case evidence shows the ICS working well in most situations but largely failed to assert control in the largest crisis. Although the ICS is flexible, the risk of a mismatch between governance form andtask has become greater in the mandatory diffusion period because responders no longer have discretion in choosing governance form, and many of those expected to use the ICS have little experience with it. The continuity of the ICS in national crisis policy after Hurricane Katrina appears to confirm an unwillingness to moderate it as a template for crisis response (FEMA 2007). Switching between Network Governance Forms The ICS also suggests a possible path of network evolution. Provan, Fish, and Sydow (2007, 509) ask "How do networks evolve from early birth to maturity and beyond? Does evolution occurs in predictable ways, either in specific evolutionary stages or based on environmental conditions and internal pressures and changes?" The ICS illustrates the possibility of switching between centralized and more decentralized forms of network governance consistent with task demands. During crises, network governance is highly centralized. But between crises, the ICS does not exist. Crisis response networks are loosely affiliated and follow a shared governance model. In precrisis periods, responders can build working relationships and trust and improve their understanding of mutual capacities and the principles of the ICS, thereby laying the groundwork for an integrated response during the actual crisis. Crisis networks therefore suggest the fluidity of network governance. In this policy area, network governance does not evolve incrementally but cyclically, changing in rapid fashion in response to environmental conditions that give rise to specific task demands. Tensions between Efficiency and Inclusiveness Provan and Kenis (2008) identify a tension between efficiency and inclusiveness in networks. The short-term coordination costs of incorporating new members discourage inclusiveness, even though new members may improve long-run outcomes. The frequent difficulty in incorporating emergent members of crisis networks illustrates this tension. Potential members are less likely than existing members to demonstrate network norms, understand the ICS as a governance mechanism, or have relationships with core members. As a result, they encounter and impose higher coordination costs, creating an incentive not to incorporate them. The evidence suggests that the bias against inclusiveness may increase under (a) conditions of mission urgency and (b) when the emergent component is relatively large. For urgent missions, NAOs will be sofocused on immediate task achievement that the coordination costs imposed by new members may appear prohibitive. As the number of potential members increase, the NAO can only incorporate so many. The resultis that a highly emergent crisis network is likely to be less formalized and coordinated. A bias against inclusiveness may reduce coordination costs for the ICS in the short run, but will see the loss of potential network resources and/or increase the number of uncoordinated"free agents" that are taking action in the ICS sphere of responsibility. The cases suggest three ways to reduce the bias against inclusiveness. First, previous working relationships between potential and coremembers of the incident command reduce coordination costs and facilitate inclusion, for example, the relationship between the incident commander and a representative of Southwestern Bell in Oklahoma City discussed above. Second, coordination costs can be reduced if many potential members self-organize to provide a single point of contact for the ICS. Construction contractors followed this approach in Oklahoma City and provided critical help in removing debris from the site. Third, the incident commander or network member may know that a potential member offers resources that have immediate value greater than the costs of including them and will reach out to incorporate this actor.For example, at the Pentagon, incident command officials contacted alocal Home Depot for flashlights and batteries. Home Depot then assigned a liaison to the site to coordinate the supply of other needed materials. The Relationship between Trust and Authority Provan and Kenis propose that NAOs offer a logical form of networkgovernance when there is limited trust between responders. The logicof design behind the ICS is consistent with this proposition, relying on centralized authority to direct a network where members may havelimited prior contact. However, the case evidence questions this trust, showing that responders view trust as a critical complement to authority. Authority, by itself, offers an inadequate basis for coordination without positive working relationships between core members. Although the NAO might be more suited than other governance forms to overcome low-trust scenarios, trust still appears to be positively related to NAO-governed network coordination. This finding supports the claim that formal control systems and trust are not necessarily mutually exclusive alternatives to fostering coordination, but can complement one another, especially in high-risk contexts (Schoorman, Mayer, and Davis 2007, 346-7). CONCLUSION This article has examined the network governance of crisis response via the ICS. A network governance perspective allows us to integrate two seemingly conflicting imperatives of emergency management--the need for interorganizational collaboration and the need for rapid coordinated response. Although the ICS seeks to coordinate multiple organizations using a hierarchical structure, the network setting significantly affects the operations of the ICS. Much of the study of whole networks has focused on health and social services, and more research is needed on other government functions (Provan, Fish, and Sydow 2007). Such research can generate or modify propositions about network governance in specific functional areas and under specific constraints, thereby helping to better distinguishbetween broadly applicable and more contingent propositions about how networks operate. In the case of this article, the ICS tells us a good deal about the functioning of "problemsolving" networks created to resolve pressing tasks (Milward and Provan 2006) and about how centralized network governance forms operate amid crisis conditions. FUNDING IBM Business of Government Endowment; the Graduate School of the University of Wisconsin-Madison. doi:10.1093/jopart/mun033 Advance Access publication on January 30, 2009 REFERENCES 9/11 Commission. 2004. Final report of the National Commission on terrorist attacks upon the United States. Washington, DC: Government Printing Office. Agranoff, Robert, and Michael McGuire. 2001. Big questions in public network management research. Journal of Public Administration Research and Theory 11:29--326. Bigley, Gregory A., and Karlene H. Roberts. 2001. The incident command system: High reliability organizing for complex and volatile tasks. Academy of Management Journal 44:1281-99. Boin, Arjen, and Paul 't Hart. 2003. Public leadership in times ofcrisis: Mission impossible. Public Administration Review 63:544-53. Buck, Dick A., Joseph E. Trainor, and Benigno E. Aguirre. 2006. A critical evaluation of the incident command system and NIMS. Journal of Homeland Security and Emergency Management 3: 1-27. Brass, Daniel J., Joseph Galaskiewicz, Henrich R. Greve, and Wenpin Tsai. 2004. Taking stock of networks and organizations: A multilevel perspective. Academy of Management Journal 47: 795-817. Bringer, Joy D., Lynne H. Johnston, and Celia H. Brackenridge. 2006. Maximizing transparency in a doctoral thesis: The complexities of writing about the use of QSR*NVIVO within a grounded theory study. Qualitative Research Journal 4:247-65. Brower, Ralph S., Mitchel Y. Abolafia, and Jered B. Carr. 2000. Onimproving qualitative methods in public administration research. Administration & Society 32:363--97. Cole, Dana. 2000. The incident command system: A 25-year evaluation by California practitioners. Emmitsburg, MD: National Fire Academy. Dynes, Russel R. 1970. Organized Behavior in Disasters. Lexington,MA: Heath Lexington Books. Drabek, Thomas E., and David A. McEntire. 2003. Emergent phenomenaand the sociology of disaster: Lessons, trends and opportunities from the research literature. Disaster Prevention and Management 12:97-112. Eisenhardt, Kathleen M., and Melissa E. Graebner. 2007. Theory building from cases: Opportunities and challenges. The Academy of Management Journal 50:25-32. Federal Emergency Management Agency (FEMA). 2007. National incident management system---Draft. http://www.fema.gov/pdf/emergency/nrf/nrf-nims.pdf (accessed July 12, 2008). George, Alexander L., and Andrew Bennett. 2004. Case studies and theory development in social science. Cambridge, MA: MIT Press. Hall, Thad E., and Larry J. O' Toole, Jr. 2000. Structures for policy implementation: An analysis of policy implementation 1965-1966 and 1993-1994. Administration & Society 31:667-86. Koppenjan, Joop, and Hans-Erik Klijn. 2004. Managing uncertaintiesin networks: A network approach to problem solving and decision making. New York, NY: Routledge. Laumann, Edward O., Peter V. Marsden, and David Prensky. 1983. Theboundary specification problem in network analysis. In Applied network analysis, eds. R. S. Burt and M. J. Minor, 18-34. Beverly Hills, CA: Sage. Leonard, Herman B., and Arnold M. Howitt. 2006. Katrina as prelude: Preparing for and responding to Katrina-class disturbances in the United States. Journal of Homeland Security and Emergency Management 3:1-20. May, Peter J. 2003. Policy design and implementation. In Handbook of public administration, eds. J. Pierre and B. G. Peters, 223-33. London: Sage. May, Peter J., Byran Jones, and Samuel Workman. 2008. Organizing attention: Responses of the bureaucracy to agenda disruption. Journal of Public Administration Research and Theory 18: 517-41. McGuire, Michael. 2006. Collaborative public management: Assessingwhat we know and how we know it. Public Administration Review 66 (special issue): 33-43. Miles, Matthew B., and A. Michael Huberman. 1994. Qualitative dataanalysis, 2nd ed. Thousand Oaks, CA: Sage. Milward, H. Brinton, and Keith G. Provan. 2006. A Manager's guide to choosing and using collaborative networks. Washington, DC: IBM Center for the Business of Government. Mission Centered Solutions. 2003. Southern California firestorm 2003. Report for the Wildland Fire Lessons Learned Center. Moynihan, Donald P. 2005. Leveraging collaborative networks in infrequent emergency situations. Washington, DC: IBM Center for the Business of Government. --. 2007. From forest fires to hurricane Katrina: Case studies" ofincident command systems. Washington, DC: IBM Center for the Business of Government. --. 2008. Combining structural forms in the search for policy tools: Incident command systems in U.S. crisis management. Governance 21:205-29. Oklahoma City National Memorial Institute for the Prevention of Terrorism (MIPT). 2002. Oklahoma City seven years later: Lessons for other communities. Oklahoma: Oklahoma City National Memorial Institute for the Prevention of Terrorism (MIPT). www.mipt.org/pdf/MIPT-OKC7YearsLater.pdf (accessed February 2, 2007). Oklahoma Department of Civil Emergency Management (ODCEM). n.d. After action report: Alfred P. Murrah federal building bombing 19 April1995. Oklahoma: Department of Central Services Central Printing Division. www.ok.gov/OEM/documents/Bombing%20After % 20Action%20Report.pdf(accessed March 3, 2007). Provan, Keith G., Amy Fish, and Joerg Sydow. 2007. Interorganizational networks at the network level: A review of the empirical literature on whole networks. Journal of Management 33:479-516. Provan, Keith G., and Patrick Kenis. 2008. Modes of network governance: Structure, management and effectiveness. Journal of Public Administration Research and Theory 18:229-52. Provan, Keith G., and H. Brinton Milward. 2001. Do networks reallywork? A framework for evaluating public-sector organizational networks. Public Administration Review 61:414-23. Raab, J6rg. 2002. Where do policy networks come from? Journal of Public Administration Research and Theory 12:581-622. Richards, Lyn. 2002. Rigorous, rapid, reliable and qualitative? Computing in qualitative method. American Journal of Health Behavior 26:425-30. Rohde, Michael. 2002. Command decisions during catastrophic urban interface wildfire: A case study of the 1993 Orange County, California, Laguna Fire. Thesis, Department of Occupational Studies, California State Univ. Long Beach, CA Rosenthal, Uriel, Paul 't Hart, and Michael T. Charles. 1989. The world of crises and crisis management. In Coping with crises: The management of disasters, riots and terrorism, eds. U. Rosenthal, M. T. Charles, and P.'t Hart, 1-33. Springfield, IL: Charles C. Thomas. Scharpf, Fritz. 1993. Coordination in hierarchies and networks. InGames in hierarchies and networks: Analytical and empirical approaches to the study of governance institutions, ed. F. W. Scharpf, 125-66. Frankfurt, Germany: Campus Verlag. Schoorman, F. David, Roger C. Mayer, and James H. Davis. 2007. An integrative model of organizational trust: Past, present and future. Academy of Management Review 32:34-354. Stallings, Robert A., and E. L. Quarantelli. 1985. Emergent citizen groups and emergency management. Public Administration Review 45 (special issue): 93-100. Tierney, Kathleen, and Joseph Trainor. 2004. Networks and resilience in the World Trade Center disaster. Research Progress and Accomplishments, vol. 6. Buffalo, NY: Multidisciplinary Center for EarthquakeEngineering Research. Titan Systems Corporation. 2002. After action report on the response to the September 11 terrorist attacks on the Pentagon. Report to Arlington County. http://wwwarlingtonva.us/Departments/Fire/Documents/after_report.pdf (accessed June 12, 2006). Trainor, Joseph E. 2004. Searching for a system: Multi-organizational coordination in the September 11th World Trade Center search and rescue response. DRC Preliminary Publication #343 Public Entity Risk Institute. http://www.udel.edu/DRC/Preliminary Papers/PP343-Searching%20for%20a%20System%20Paper.pdf (accessed May 23, 2007). US Department of Homeland Security (US DHS). 2004. National incident management system. Washington, DC: Government Printing Office. U.S. House of Representatives Select Bipartisan Committee to Investigate the Preparation for and Response to Katrina (House Report). 2006. A failure of initiative. Washington, DC: Government Printing Office. U.S. Senate Committee of Homeland Security and Government Affairs (Senate Report). 2006. Hurricane Katrina: A nation still unprepared. Washington, DC: Government Printing Office. Varley, Pamela. 2003. Command performance: County firefighters take charge of the 9/11 emergency. Cambridge, MA: Kennedy School of Government Case Program. Waugh, William L., Jr., and Gregory Streib. 2006. Collaboration and leadership for effective emergency management. Public Administration Review 66 (special issue): 131-40. Winter, Soren C. 2003. Implementation: Introduction. In Handbook of public administration, eds. J. Pierre and B. G. Peters, 205-11. London: Sage. Donald P. Moynihan University of Wisconsin-Madison (1) The other options identified by Provan and Kenis (2008) are shared governance and lead organization-governed networks. Shared governance networks are loosely affiliated and decentralized. Crisis response networks are akin to the shared governance model in between crises. But when a crisis actually occurs, networks become highly centralized via the incident command. Lead organization--governed networks are dominated by a single participating member, often the one with a formal mandate and the most task-specific resources. Unlike lead organizations, the incident command almost always contains representatives of multiple organizations, even if a single commander is in charge, and the incident commander might not be a representative of the organization providing most of the resources. (2) A case that was coded, the anthrax attacks of 2001, was subsequently excluded from the analysis because of a lack of sufficient documentation. A full list of the documentation used for the END case can be found in Moynihan (2005), and sources for the other cases are described in Moynihan (2007). (3) The process of interpreting complex and interactive concepts as they are represented in detailed texts about different contexts does not lend itself to intercoder reliability (Richards 2002). Intercoder reliability is appropriate if the researcher is employing a purelydeductive approach and a relatively simple coding instrument (e.g., May, Jones, and Workman [2008] coding of federal rules for specific characteristics). But computer coding does provide a documentary record of the researcher's definitions of codes, judgments about how to code text, and interpretations of those codes in memos. This offers an audit trail and a level of transparency that allows other researchersto critique and reinterpret the analysis (Bringer, Brackeuridge, andJohnston 2006). This article was initially presented at the 2007 Public ManagementResearch Association meeting at the University of Arizona, Tuscon. The author would like to thank the editor of Journal of Public Administration Research and Theory and three anonymous reviewers who offeredexceptional guidance in improving the manuscript. Address correspondence to the author at dmoynihan@lafollette.wisc.edu Table 1 The Impact of Network Characteristics on Crisis Response Network Diversity Shared Authority Fire cases Limited diversity and General agreement about most responders who was in charge, aided shared fire response by paramilitary culture of background. Problems firefighting. Occasional did occur in failure to incorporate incorporating local governments tied local governments, to jurisdictional disagreements. Oklahoma Difficulty in Local responders first on City incorporating scene and appointed bombing voluntary element, incident commander, FEMA and FBI had competing claims to control site. Initial conflict with FEMA about jurisdiction. Pentagon Difficulty in Local responders first on on 9/ 11 incorporating scene and appointed voluntary element, incident commander, FEMA and FBI had competing claims to control site. Incident commander later invited federal agencies to join unified command. END Limited diversity of Shared command between managers as response state and federal officials was dominated by state without major conflict, and federal vets. Initially state Limited conflict based jurisdiction, but state on professional officials kept federal background, for example, counterparts involved. When forest service officials federal declaration of and vets disagreed on emergency, federal actors application of ICS. had greater authority but kept state officials involved. Strong sense of cooperation among leaders. Hurricane Diversity of network Multiple commands, lack Katrina weakened coordination, of clarity about who was fostering delay, in charge, and who was confusion, responsible for specific and solo actions, roles. Confusion exacerbated by introduction of new roles such as the Principal Federal Official and lack of familiarity with ICS. Trust and Working Relationships Fire cases Strong trust and positive prior working relationships viewed as critical to minimizing conflict, sharing resources, and fostering coordination. Oklahoma Strong trust and positive City relationships facilitated bombing problem-solving, role allocation, facilitating coordination, and incorporating emergent actors. Pentagon Strong trust facilitated on 9/11 coordination, provision of discretion, role allocation, and willingness to move to unified command. Weaker relationships led to solo action. END Strong prior working relationships between hubs facilitated coordination. The Area Veterinarian in Charge (a federal employee of the Department of Agriculture permanently based in California) had a strong relationship with the state veterinarian. Hurricane Weakening of federal, state, Katrina and local relationships hampered coordination. Prior to Katrina, FEMA lost personnel and resources that had maintained working relationships with state and local responders.