Social Network Analysis for the Implementation of Sendai Framework for Disaster Risk Reduction in Iran

Background: Over recent years, the exposure of people and assets to disasters has been faster than reducing vulnerability in all countries. As a result, new risks have been formed and losses due to disaster are progressively increasing. Suffering from significant losses in the aftermath of disasters every year, Iran is no exception. Governmental and non-governmental stakeholders are jointly responsible for managing the risks of disasters. Hence, appropriate, collaborative and timely interactions of involved organizations will play an important role in their operation, especially during disasters. Methods: In this study, we used the Social Network Analysis (SNA) to analyze the network of stakeholders in disaster risk management in Iran. Our review of literature, laws, and regulations of disaster risk management plus brainstorming identified a list of 85 stakeholders. We used the Delphi method among purposefully selected experts to score the relationship between the stakeholders. We then used the modularity optimization method to identify groups with greater interaction. Organizations with key-roles in the network and the ones in need of stronger relationships were identified through centrality measurements. Results: The density of this network was 0.75, which represented that not all the stakeholders were connected. Among all organizations identified, the National Disaster Management Organization and Civil Defense Organization showed higher influences considering their responsibilities. Conclusion: To provide a visual and tangible picture of the status and interrelationships among the stakeholders, this method identified groups with better interaction using community/cluster detection and modularity optimization methods. Understanding the current structure of the network and strengths and weaknesses of the interactions among stakeholders may help improve disaster risk management in Iran. Results of this research

determine the role and importance of different organizations, their weakness, and strong points. Also, results help them to plan to strengthen their roles and solve their problems.

Background
Based on the World Disaster Risk Report, sustainable development is impossible without considering the disaster risk reduction approach (1). Ample evidence indicates that, through disaster management strategies, the risk of loss of life and assets increases faster than vulnerability reduction strategies all around the world (2). Statistics show that hazards such as earthquakes, tsunamis, tornadoes, and floods have caused annual economic losses of $250-330 billion globally. Moreover, from 1980 to 2012, 42 million years have been lost (YLL) due to disasters, over 80% of which occurred in Low and

Middle-Income Countries (LMICs) (3).
Due to its geographic location, topographical properties, and high structural and nonstructural vulnerabilities, Iran is prone to many hazards (4). Iran's World Risk Index was governments. The first step in this way is to identify stakeholders and determine their role and importance (2).
Given the complex nature of disaster risk management (DRM) and the diversity of the structure of relevant organizations, traditional methods of stakeholder analysis are less able to examine the dynamics and interaction of these links (6)(7)(8)(9)(10). Organizations are not important lonely in a network but their links and interactions with other stakeholders determine the importance of that organization. In these cases, methods such as the tools in network analysis approaches that examine the behavior of systems collectively would be an appropriate option (11)(12)(13). This is a matter of distinguishing and prioritizing the network analysis approaches over descriptive methods (14). Numerous studies have investigated the coordination and inter-organizational cooperation in disasters using these approaches. The flow of information among relief teams and the way of communicating in the response phase are evaluated in many studies and drawbacks are considered as lessons learned (7)(8)(9)(15)(16)(17)(18)(19)(20).
While coordination at all levels is an essential action in DRM, inconsistency among stakeholders has become one of the major recognized organizational challenges, (21).
Coordination prevents parallel work and loss of resources through planning, making mutual trust, combining capacity and timely use of resources in different organizations (7,11). Analyzing the stakeholders' roles and responsibilities are considered as the first steps in establishing effective coordination and cooperation before the hazards occur (8,21).
By identifying the internal and external coordination and communication challenges of systems, the challenges for planning and policy-making could be eliminated (10,22). For instance, a study was conducted to investigate the communication between the members of the incident command system (ICS) in the response process of six incidents (three firefighting operations due to hazardous chemical incidents and three police operations in response to emergency calls). The results showed that different types of communication affect the response phase quality in two types of events and, the best communication model was introduced accordingly (18,23). Another study was carried out based on reports of organizations to the Federal Emergency Management Agency (FEMA). The results showed that effective response and recovery require the optimal coordination among organizations and the trust between the governmental and non-governmental sectors at all levels as well as in the society. They emphasized that optimal preparation and response could be achieved within the framework of inter-organizational trust and coordination. (7,16).
To best of our knowledge, there are not many stakeholder analysis based on the network analysis method, most of which have merely been focused on the response phase of DRM in Iran (16,21). This study aimed to analyze the network of DRM stakeholders' as the first step for the implementation of the Sendai Framework in Iran. In addition to providing a visual and tangible picture of the status and interrelationships among the stakeholders, this method identified groups with better interaction using community/cluster detection and modularity optimization methods. Using centrality measurements (Degree, In-Degree, Out-Degree, Betweenness, Closeness, and Eigenvector), the organizations that play a key role in the network, as well as the organizations that need to strengthen their relationships were identified. Quantitative results were also used to interpret the interaction between the organizations and understand the blind spots of the links. Through understanding the relationships among various stakeholders in DRM, an opportunity will arise to plan for enhancing the level of stakeholders' participation with lesser roles.

Methods
There are two main organizations for DRM in Iran: the National Disaster Management Organization (NDMO) and the Civil Defense Organization (CDO). The NDMO has 14 specialized working groups with diverse members according to the title and description of their duties. Based on the members of these working groups, we developed a list of 52 organizations. Data triangulation was used to enhance the data credibility (24). A review of the literature, rules, and regulations related to DRM was then conducted and 11 organizations were added to the list. We then emailed the list to 32 experts of DRM. The experts were selected based on their experience or employment in one of the organizations involved in DRM. Following brainstorming, the list expanded to 85. The title of eight organizations was also corrected. The list remained unchanged after we received 19 response sheets. The name and the number assigned to the organizations for simplicity are presented in Table 1.
In the next step, we created a matrix of 85 actors using the Excel software. The matrix was then emailed to 32 experts. The social profile of the participants is described in Table   2. During this round of Delphi, the experts were asked to rate based on the Likert scale (0-5: 0 means the lack of interaction and 5 means the maximum connection), the extent to which organizations interacted with each other for DRM. This method is best known as the Full-rank Ordinal Measures of Relations (14).
The matrix was completed and returned by 17 people. Two reminder emails were sent to the remaining 15 experts within 10 interval days. The number of responses then increased to 22, and the network analysis was performed with 22 participants (68.75% response rate).
For data analysis, the responses were called in Python software and the corresponding networks were formed. We visualized the developed networks by Gephi software.
Stakeholders played the role of nodes (circles) in our networks and the relations between them were the links. Links were weighted and directed. The weight of the links shows the average score of each link in the 22 response sheets. Due to the long list of the organization, the numerical code was used instead of its name. The non-governmental sector was also identified through bolder margin circles. Figure 1 shows a schematic view of the process of data extraction.
Density is one of the popular indices to show the ratio of the presented links to the possible links (25). A network with n nodes can maximally have n (n − 1) directed links or n (n − 1)/2 undirected links. Density 1 represents a fully connected network and density 0 represents a fully sparse network with no link. Heterogeneity in the node connections and lack of some links in the network reduces the density index below 1.
Heterogeneity in the organizations' connections and weight of these connections (in range of 0 − 5) indicates the unequaled impact of organizations in the stakeholder's network.
Over the years, different centrality measurements have been introduced and used to quantify effective nodes in terms of their different roles in networks. We used four centrality measurements in our analysis; Degree, Betweenness, Closeness, and Eigenvector.
Degree centrality is one of the first centrality measurements which count the number of links by each node in the network. It indicates the reputation of nodes. The higher the degree of a node means the greater its authority (23). In a directed network, in-degree and out-degree can be different for each node. In-degree counts the number of the interaction of other nodes with a node and out-degree counts the number of the interaction of a node with other nodes. In our stakeholder's network, in-degree shows the frequency that other organizations refer to an organization for planning, coordinating and their actions. In contrast, out-degree shows the frequency that an organization contributes to other organizations in the network. As links in our network are weighted, we can also calculate Weighted Degree (WD), Weighted In-Degree (WID), and Weighted Out-Degree (WOD). Betweenness determines the position of a node within a network based on its ability to establish relationships among other couples (26). Organizations with a high betweenness play the role of the interface in the network, removing them might disturb the flow of communications and information in the network. These nodes are also known as brokerage (27). Closeness centrality is another important indicator that shows the closeness of a node to other nodes in the network (28,29). It is calculated as the inverse of the sum of the length of the shortest paths from each node to all the other nodes in the network. The nodes that are higher in this index make their connections with other network nodes more easily because they have fewer intermediaries.
Eigenvector centrality is the last index we used. It quantifies the ability of a node to get connected to nodes with a higher degree. The importance of this indicator is determined when the main node does not seem very important in terms of other centrality measurements; however, it is connected to other important nodes and is the bottleneck of communication for other nodes or clusters (30).
Heterogeneity of node's connections usually leads to the formation of clusters of nodes that mostly communicate with each other than the rest of nodes in the network (11,31,32). We find these clusters or communities by applying the Louvain (33) method with modularity optimization (34). In the beginning, the Louvain algorithm considers all the nodes in as single clusters. Iteratively, nodes join their neighboring clusters in the next steps to maximize modularity. Modularity is a scalar value in range −1 to 1 that quantifies the distance of detected clusters from the ones in the network with randomized links.
Negative modularity refers to the situation that nodes are assigned to the wrong clusters, zero occurs when the number of detected clusters is equal to the number of nodes, and higher positive values represent more optimal partitions (35,36).
In order to view the visual details of the network in denser networks, we eliminated weaker links with the network reduction technique in the sequential stages (11). This was carried out to identify organizations with weaker communication and design interventions in order to ensure their cooperation. The comparison of these measures identifies the organizations with a key role in the network, plus the relationships in need of strengthening. Figure 1 summarizes the whole process of data preparation and analysis in this research.

Results
Our findings revealed 85 organizations involved in DRM in Iran, of which 58 are governmental agencies and the rest are non-governmental actors. Figure    The Eigenvector centrality score nodes based on their connections to high degree nodes.
The significance of this measurement is determined by the fact that sometimes an organization does not seem to have good connections to other organizations, but their main connections are with core organizations.  In the event of a disaster, one of the most important issues is the timely performance of the organizations in order to provide the best services in the coordinated space. This coordination has begun before the occurrence of hazards, and all actors in this area must establish appropriate communication with other stakeholders in accordance with their duties (8,10,12,15,17,20,37,38). The stakeholders are individuals, groups, and organizations that are influenced by decisions and plans of an organization, or affect its decision-making (4). Accordingly, an analysis of the network of DRM stakeholders in Iran was conducted with 85 organizations. The appearance of the network of DRM stakeholders shows acceptable consistency with a density of 0.75. This is the strength of the network, which is necessary for planners to set up a participatory program by identifying key network entities and institutions (39). The network has three clusters that are different not only in the number of cluster members but also in the importance of the organizations within each cluster. The main node of each cluster is usually the node that has the largest number of connections to other nodes and is considered to be the core organization (30).
These cores have also good connections to the nodes in other clusters.
The NDMO (30) (6), which is considered a vital infrastructure, might have a significant impact on community resilience (2). Schools have special operational capacities and play an effective role as educational poles, which could be used as evacuation places at the time of disaster (40). At the Bangkok Summit in 2007, it was suggested that disaster risk reduction should be incorporated into educational policies, to be used to promote educational systems to transfer training to the community (41). The IRCS, as the responsibility of the Public Education working group, might plan activities in this regard, using its high betweenness capacity in cooperation with the NDMO (Figure 2-D).
The lack of coping capacities of Iran is 80.35%, which is categorized among high-risk countries. In terms of vulnerability and lack of adaptive capacities, it is classified as 47.78% and 43.81% (5)

Conclusions
Although reducing the risk of disasters has been conventionally the responsibility of governments in Iran, this, in reality, is a shared responsibility of the government and relevant stakeholders. We used different centrality measurements (degree, in-degree, outdegree, betweenness, closeness, and eigenvector centralities) to find effective organizations in the DRM network. In particular, CBOs, as enablers, play an important role in helping the government comply with national policies and laws at the local, national, regional and global levels. Governments can use social network analysis to make policy, regulation, and culture-building. Our assessment of the stakeholders in the DRM network in Iran demonstrated a good coherence. The network includes three clusters of organizations with different responsibilities (the theoretical basis of management, operational and executive affairs, and policy affairs), most of whom collaborate with their members strongly. Removing less weighted links in the network reveals that there are some more coherent clusters inside the main three clusters. In order to strengthen DRM (especially prior to the occurrence of hazards), and implement the Sendai Framework, it is necessary to define the organizational tasks and the participatory plan. Needless to say that transparent process of accountability, monitoring, evaluation, and reporting on the progress of work need to be also explained. To improve the analysis, it would be worthwhile to monitor and compare the network of organizations before and during disasters, periodically.
In addition, SNA could provide a visual and tangible picture of the status and interrelationships among the stakeholders and identify groups with better interaction using community/cluster detection and modularity optimization methods in other issues. Participants were also reassured that the information obtained was for research purposes. The approval code to do the research is Ir.tums.sph.rec.1396.4315 on 22 January 2018.

Consent for publication
Not applicable Availability of data and materials Supplementary data is Centrality indices in the Organizations (Table 3).

Competing interests
We declare that none of the authors or their organizations has any conflict of interest in the publication of this paper.

Funding
The authors received no fund for this study.

Kind of Reviewed Organization
Governmental Non-governmental Figure 1 A schematic view of the process of data preparation and network analysis.   Supplementary Files