Longitudinal multiplexity of online ECNs

Multiplexity denotes the extent to which organizations are committed to diverse collaborative activities [22] as well as the strength of ties between them [23]. The emergency management cycle, one of FEMA’s most frequently used frameworks, consists of four phases: mitigation, preparedness, response, and recovery. Specifically, mitigation and preparedness occur before the disaster ever strikes. In these two phases, organizations prepare for potential disasters and hazards through formulating emergency plans, raising public awareness, and enhancing infrastructure quality. During the response phase, organizations provide emergency assistance, maintain transportation and communication, and operate other support functions. Immediately after the disaster, organizational actors enter the long-term recovery by restoring public services to full or sufficient functional capabilities and rebuilding infrastructures and institutions damaged in the devastation. In practice, effective disaster responses typically rely on the synergy of multiple ECNs generated and sustained across different phases of emergency management, while the coupling failure of various relational types is likely to compound ECNs’ overall vulnerability.

As the disaster evolves, participating actors’ engagement in the social media environment may exhibit diverse communication patterns, and collaborative network structures constructed online are also likely to change in accordance with how actors leverage various social media features to achieve their collective goals [24]. Messages sent during emergency response, for instance, incorporate more representations of information broadcasting and brokerage (e.g., retweets) while fewer displays of interactive activities (e.g., replies) when compared to ordinary Twitter use [4]. The engagement of organizational accounts may reveal a different communication pattern. A content analysis of the organizational Twitter use shortly after the Haiti earthquake in 2010 revealed a decreased use of information dissemination strategies (e.g., links) and a consistent use of techniques to involve other organizations (e.g., retweets, mentions) [25]. In the context of China, social organizations such as NPOs and NGOs were found to be more likely to adopt WeChat groups as a major platform for disseminating and sharing disaster information, whereas government agencies and private enterprises used WeChat generally as a supplementary communication tool [26]. This illustrates the fact that as the emergency unfolds over time, multiple ECNs built upon diverse technological features of social media may present distinct structural patterns. Hence, this study scrutinizes three types of online ECNs that can be identified in line with the emergency management cycle: online preparedness networks, online response networks, and online recovery networks.

Within the online preparedness network, organizational actors keep regular and informal online interactions with other public, private, and non-profit organizations staying active in or beyond the domain of emergency management. This type of network involves scant formal collaborative activities and programs designed to cope with disastrous situations. To take advantage of unexploited personnel, expertise, resources, and information, organizations need to better grasp the informal network structures in order to align them with organizational goals and to comprehend the underlying logic of informal networks that have not been acknowledged in any form of cooperative agreement [27]. Moreover, informal networks formed during this phase may mimic friendship networks in rendering venues for organizations to channel resources, share information, and strengthen capacity [10, 28]. Such informal networks are likely to formalize over the long haul and further facilitate the information sharing and resource mobilization when the disaster really strikes [29].

Previous research on structural embeddedness demonstrates how actors’ network positions in terms of preexisting or shared third-party ties contribute to the formation of new ties and economic cooperation [30–32]. Embedded in a densely multiplex network, organizations seek a variety of means to access fit links [33], obtain relational benefits [34], and minimize the uncertainty of collaborative activities [35]. They thus tend to sustain collaborative relationships with other organizations such that all participating actors present interdependency in sharing resources and information [36]. Similarly, empirical efforts on collective action networks (e.g., social movement coalition) suggest that the preestablished connections between organizational members help promote the communication, trust, and common values, which in turn encourage collaborative behaviors in collective actions [37].

In the context of emergency management, researchers have also stressed the importance of preexisting relationships in intergovernmental and cross-sectoral collaboration and coordination [38]. Confronted with disasters, organizations come to the rescue of afflicted areas and communities, frequently resorting to preexisting ties for prompt emergency planning and coordination. These ties are often built and developed in the preparedness phase, when a broad spectrum of organizations have been mobilized to take action to support and enhance the emergency response (e.g., trainings, drills, joint exercises). An examination of the partnership between public and non-profit organizations revealed that pre-disaster ties would enable affected communities to perform better, owing to the fact that mutual trust and commitment can be built and enhanced among network actors in their daily working relationships [39]. This study argues that the importance of structural embeddedness remains constant online such that the symbolic forms of interorganizational relationships rely on organizations’ actual connections in the real world [40]. For example, the hyperlink network constructed by Chinese environmental organizations reveals high density, which moves beyond the general features of Internet-based technologies yet reflecting the close offline connections between organizational actors in the domain of environmental protection [41]. Therefore, the first set of hypotheses is proposed as follows:

1. Hypothesis 1 (H1): Organizational actors’ online preparedness networks positively correlate with their online response networks.
2. Hypothesis 2 (H2): Organizational actors’ online preparedness networks positively correlate with their online recovery networks.

Unlike informal preparedness networks, online networks built in the response and recovery phases usually incorporate more formal collaboration and coordination. During actual disasters, interorganizational ties constructed through various social media features may enable organizations to share common goals and values on account of varying levels of participation in the collaborative network. Network ties formed at the preparedness stage have also been enhanced and developed because actors prefer to seek new partners to address emergent needs and achieve immediate goals [42]. Therefore, online preparedness ties are not necessarily transformed into effective working relationships in response to disasters since preexisting connections may not always be exploited in line with the resource interdependency shaped in disastrous scenarios.

Nonetheless, the collaborative approach to sustainable recovery after disasters indicates that the post-disaster development of impacted communities and areas is a rather complex and non-linear social process driven by both political-administrative arrangements [43] and multiple accessible resources [44]. The involvement of various organizations and sectors forms an intricate post-disaster ECN maintained by common interests and shared resources, in which external aid and assistance is embedded. As a result, the structural inertia of online ECNs may persist even after external assistance withdraws from the affected areas in a geographic sense. In other words, organizations’ needs for resources and information at the post-disaster recovery stage can be better accommodated through response networks. Therefore, the following hypothesis is posited to investigate the correlation between online ECNs built during disaster response and those sustained in the post-disaster recovery:

1. Hypothesis 3 (H3): Organizational actors’ online response networks positively correlate with their online recovery networks.

As discussed earlier, the sustainability of external aid and assistance poses the principal challenge for achieving a sustainable recovery; that is, the doubt remains about whether the long-term disaster recovery of impacted areas and communities can be sustained once the outside help subsides. Prior research accentuates the combination of local reliance on internal capacity and flexible management of external assistance in addressing post-disaster needs (e.g., [45–47]). A systematic review of the disaster assistance framework in the U.S. concluded that post-disaster aid and assistance should rely on response operations with numerous accessible resources rather than on plans and arrangements stipulated prior to the disaster [44]. It also questioned the current U.S. disaster recovery policy because involved organizations have conflicting assistance plans on the process before the focal event, thus lacking sufficient collaboration and coordination among public, private, and non-profit sectors. Following this stream of research, this study contends that online collaborative ties built in response operations are more likely to contribute to facilitating sustained interorganizational interactions during post-disaster recovery because the structural inertia may come into existence between two types of formal connections, and more importantly, further reinforce as collaborative efforts during response and recovery are both made to address actual disastrous situations. Hence, a hypothesis comparing correlations among three types of online ECNs is proposed as follows:

1. Hypothesis 4 (H4): The correlations between online response networks and online recovery networks are higher than the correlations between online preparedness networks and online response networks and between online preparedness networks and online recovery networks.

Structural constraints on network formation and evolution

In a disaster setting, each EMS entails a certain level of authority in constraining the autonomy of organizational actors and tackles collaboration dilemmas differently. For example, actors working under a coercive EMS tend to forgo more localized autonomy than those operating under a more voluntary and self-organized system [48]. Although actual response networks may differ substantially from planned networks institutionalized in political-administrative arrangements, this gap can be more salient in a centralized and coercive system than in a decentralized and voluntary system [26].

In general, there are three types of EMSs: centralized/vertical, decentralized/horizontal, and a hybrid of both [10, 49]. In the U.S., there are two major systems operating across different levels of the federal structure: the Incident Command System (ICS) and the Emergency Support Function (ESF) System. The ICS-based approach represents a centralized and command-and-control coordination mechanism, which highlights the hierarchy of authority and standard operations for managing emergencies. When a disaster occurs, the system allows the federal government to exert direct control over local emergency management actions [50, 51]. State and local governments are required to conform to the unified command framework in managing resources, assigning tasks, and making decisions. Despite its capacity to swiftly issue orders and return real-time situations [52], the ICS has long been criticized for its lack of flexibility in managing interjurisdictional and cross-sectoral collaborations as well as the utilization of community resources [53]. The delayed rescue efforts in conjunction with the collapse of communication and coordination among the U.S. government agencies after Hurricane Katrina have manifested the inadequacies of such a centralized coordination mechanism [54]. In contrast to the ICS, the ESF-based approach embodied in the National Response Framework (NRF) represents a decentralized coordination mechanism emphasizing the crucial role of state and local government agencies, as well as private and nonprofit sectors. Within the ESF-based system, organizational actors operate according to 15 ESFs, which necessitate a more horizontal and collaborative approach to an enhanced coordination across public, private, and nonprofit sectors [51, 55].

According to the ICA framework, coordination mechanisms are most effective when they overcome collective action problems while incurring the lowest transaction costs [20, 21, 56]. By and large, the centralized EMS operates a typical command-and-control approach, which categorizes response organizations and emergency tasks into ordered levels and forms a hierarchical power structure to ensure the efficiency of workflow at all levels [53, 57]. Although this system incurs high autonomy costs on actors, it can mitigate the uncertainty of disastrous situations and the associated defection risks (i.e., the probability of commitment violation; [15]). Furthermore, the coercive nature of the centralized system may reinforce the enforcement mechanism for mutual trust and commitments, thus limiting the possibility of actors engaging in opportunistic behavior. As thus, the centralized EMS can be more effective in facilitating network changes during the transition from disaster preparedness to response, which is characterized by an abrupt increase of complexity and uncertainty.

Compared with the centralized approach to emergency collaboration, the decentralized structure demonstrates more flexibility and resilience because it eases interactions between organizations across different sectors and jurisdictions, allowing actors to reach out to organizations of partnership and adapt to one another in alignment with agreed goals and shared values [57]. In addition to the low autonomy costs incurred on actors, the decentralized system can mitigate resource insufficiency and the associated coordination risks (i.e., probability of inaction; [21]) by securing access to nonoverlapping resources and information and helping actors identify mutually beneficial situations. Hence, the decentralized EMS appears more effective to facilitate network changes during the transition from disaster response to recovery, which is characterized by resource scarcity and interdependency. Yet so far, there have been few comparative studies on network development and evolution across different EMSs, not to mention those focusing on structural constraints on the social media landscape. The following set of hypotheses is thus proposed:

1. Hypothesis 5 (H5): Online preparedness networks within the centralized EMS have stronger predictive power in the formation of online response networks than do their counterparts within the decentralized EMS.
2. Hypothesis 6 (H6): Online response networks within the decentralized EMS have stronger predictive power in the formation of online recovery networks than do their counterparts within the centralized EMS.

Attribute-based network homophily

Prior interorganizational network research suggests a salient homophily effect that actors sharing similar attributes are more likely to build and sustain connections with each other [58]. Due to differences in organizational attributes, the internal structure, mobilizing capacity, and designated roles and responsibilities can vary greatly across organizations and sectors [39]. Hence, the amount and type of accessible information together with rescue resources are likely to cluster actors throughout different emergency phases. From the ICA perspective, organizational attributes can influence how actors perceive collaboration risks [48]. Collaboration and coordination among similar actors can reduce their perceived risks while raising their estimation of net expected benefit (i.e., the difference between expected benefit and expected cost) because previously shared behavioral norms and authority may reinforce interorganizational trust and operational relationships in response to a disaster [52]. The openness and transparency of online communication activities make it easier for organizations to locate those with similar interests and more difficult to conceal their preferences, thereby strengthening the homophily effect [59].

This study has identified organization type, geographic scale, and geo-location as three organizational attributes with the potential to contribute to network homophily. In response to major disasters, organizational actors of the same type share common goals and network constraints, which are usually considered as prerequisites for interorganizational alliance [37, 60]. For instance, governments at all levels play a leading role in the institutionalized emergency management. On account of the frequent flow of political-administrative arrangements and commands, government agencies are more likely to build connections with one another, rather than contacting other types of participating organizations. Similar geo-locations of two organizations indicate that they may face common external environment and similar vulnerability to natural disasters, which prompt them to form interlocal collaborative ties to exchange disaster experiences and lessons in response to floods or hurricanes [15, 17, 61]. Conversely, a greater geographic distance may increase the cost of communication and coordination, thus impeding the creation and development of network relationships between actors. Moreover, network links between organizational actors with the same geographic scale are considered fit because they share similar authorities, commands, identities, and legitimacy. Nevertheless, technological features of social media may afford a different communication pattern that transcends geographic boundaries through more accessible connections and relationship management [62]. Also, it remains unclear whether the homophily effect of geographic scale applies because collaboration in disastrous situations is largely fortified by intergovernmental and cross-sectoral partnership. Due to the inconclusive evidence, an exploratory research question is proposed as follows:

Centralized EMS in Shenzhen

Located on the eastern shore of PRD, Shenzhen was established as the first special economic zone in China and has become a fast-growing city known for its vibrant economy and technology industry. The EMS in Shenzhen complies with the One Planning plus Three Systems (一案三制) framework, which requires one preceding emergency response plan to be detailed in the later emergency legislative, institutional, and regulatory systems. Within a hierarchical structure at national, provincial, municipal, and district/county levels, roles and responsibilities of a broad range of key stakeholders were defined in the emergency operation plans. From the network perspective, the centralized EMS in Shenzhen has two distinguishable characteristics. Horizontally, government agencies play a dominant role in building and maintaining network relationships with other types of organizations. Public institutions and state-owned enterprises are in practice affiliated to government agencies regardless of whether they have been literally included in the pre-disaster emergency plan. Hierarchically, superior governments have absolute authority over government agencies at lower levels. This means that once the emergency response is activated nationwide, the district/county-level government at the place of emergent incident shall instantly take prompt action to control the potential social damage and immediately report the operational status to the government of the next higher level in a timely manner. As a result, governments at all levels, state-owned enterprises, and public institutions have been bundled into an institutionalized emergency network.

Decentralized EMS in Hong Kong

Bordering Shenzhen to the north, Hong Kong is a special administrative region of China (HKSAR) with separate economic and political systems from that of mainland China under the “one country, two systems” principle. Serving a population of around 7.4 million, the HKSAR government is led by the Chief Executive (CE), who nominates other principal officials for appointment of the central government. Immediately after the handover in 1997, the HKSAR government developed a Three-Tier EMS to regulate emergency preparedness and response in line with the scope and severity of a given natural disaster. Despite the consistent guidance provided throughout different emergency phases, the Three-Tier System is mainly operated on the basis of intergovernmental coordination and liaison. Specifically, it has explicitly defined the roles and responsibilities of involved emergency response organizations. It resembles the ESF-based system which takes a more collaborative approach to disaster response and allows for more flexibility enabling response organizations to reach out to other interest-related actors in the network to mobilize resources and share information. Moreover, a departmental liaison officer system is executed to ensure direct and proper communication between the Emergency Monitoring and Support Centre (EMSC) and key departments at all levels. Liaison officers who have ample knowledge and expertise within their areas will be called into the EMSC as necessary.

Compared with Shenzhen, the EMS in Hong Kong exhibits a more decentralized and inter-departmental collaborative structure. Nonetheless, it should be noted that these two systems are both conceptually and practically intertwined rather than mutually exclusive. Even the centralized system in Shenzhen has embraced an increasingly collaborative approach to emergency management since the SARS outbreak in 2003. As thus, the contrast between two systems is not absolute—it is rather a continuous than a dichotomous differentiation defined by the extent to which a region’s political-administrative arrangements in emergency management exhibit more coerciveness or more voluntariness. Table 1 presents the context of Typhoon Mangkhut in Shenzhen and Hong Kong.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. The context of Typhoon Mangkhut in Shenzhen and Hong Kong.

https://doi.org/10.1371/journal.pone.0289277.t001

Ethics statement

All the data collected including government documents, newspaper articles, and social media data are open for the public. Due to the relatively modest data size, data obtained from Sina Weibo and Twitter were manually captured and analyzed complying with the terms and conditions of both platforms.

Identifying network actors

This study constructs online ECNs with Sina Weibo data of 71 organizational actors in Shenzhen and Twitter data of 38 organizational actors in Hong Kong. As network researchers have emphasized the importance of both theoretical aims and case specificity in delineating social network boundaries [24, 65], online network actors were identified with the following few steps.

First, a content analysis of secondary data obtained from publicly available government documents and six weeks of newspaper articles was conducted to identify organizations involved in the ECN during the response to Typhoon Mangkhut. Most government agencies across the world, either central or local, have executed strict policies of information disclosure over emergency and disaster news. Moreover, news reports of natural disasters are of high transparency and continuity, thus providing opportunities for researchers to identify multiple types and dynamics of ECNs [39, 54, 55].

Government documents including situation reports and after-action reports were captured from government news websites of Shenzhen (www.sz.gov.cn) and Hong Kong (www.news.gov.hk). After carefully reading through these documents, 113 reports were identified from sz.gov and 46 from gov.hk as relevant for the emergency management of Mangkhut (see S1 Table). With respect to newspaper articles, Southern Metropolis Daily-Shenzhen Edition (SMD-SZ) and Shenzhen Evening News (SEN) were selected to include media coverage from both market-driven and party-affiliated media outlets in Shenzhen; while Ming Pao (MP) and Oriental Daily News (ODN) were chosen to obtain a representative sample from Hong Kong newspaper market. WiseNews, a widely used Chinese-language database that provides full-text newspaper articles, was employed to search articles on Typhoon Mangkhut that were published in a six-week period starting from two weeks before the disaster to four weeks after (September 2, 2018 to October 13, 2018). The keyword “Mangkhut” (山竹) was used for search queries and after removing duplicates and irrelevant contents, 116 newspaper articles in Shenzhen and 301 in Hong Kong remained for further analysis (see S2 Table).

Second, by means of a systematic content analysis of government documents and newspaper articles, a list of 330 organizations in Shenzhen and 195 in Hong Kong were identified to have participated in the on-the-ground ECN during and immediately after the disaster. Third, among these response organizations, 100 were found to manage verified accounts on Weibo while 43 were found to manage Twitter accounts with a moderate to considerable number of followers (see S3 Table). Finally, after data screening, those accounts engaging in no communication activities in any direct or indirect forms (i.e., tweets, hashtags, mentions, retweets) during the response and initial recovery were further excluded, resulting in a total of 71 organizational accounts in Shenzhen and 38 in Hong Kong. These accounts represent a cross-section of public, for-profit, and non-profit sectors, as well as a collection of emergency collaborative relationships managed by organizations.

Data collection

The next step is to obtain the research data. All tweets posted by the 71 Shenzhen-based Weibo accounts and the 38 Hong Kong-based Twitter accounts over the aforementioned six-week period were manually captured. These social media data were then broken into three sub-datasets for both regions, each of which represents a two-week observation of online interactions and collaborative activities (T1-preparedness phase: September 2, 2018 to September 15, 2018; T2-response phase: September 16, 2018 to September 29, 2018; T3-recovery phase: September 30, 2018 to October 13, 2018).

During actual disasters, multiple interorganizational ties constructed through various social media features may enable participating actors to share common goals and group norms on account of different levels of engagement in online ECNs. One established means of observing network ties is based on affiliations of organizational actors, for example, shared membership in boards of directors [66], co-authorship in academic publications [67], common references in media reports [68], and collaborative partnership in interlocal government agreements [69] or cooperative initiatives [70]. Following several recent empirical efforts that have applied such an affiliation approach to observing networks in online emergency collaboration [8, 24], this study captures the evidence of interorganizational communication activities operated among network actors through two types of affiliation network relationships: shared hashtags and mention contacts. In each case, the affiliation network includes two modes: organizational actors and hashtags/mentions, and network ties between any two actors are measured in terms of the observed hashtags/mention contacts they share in common. Specifically, hashtags represent actors’ communication practice in organizing words or multi-words phases that categorize and track topics in a tweet (i.e., a tag preceded by the hash mark “#” on Twitter and two hash marks between which a tag presents on Weibo). As thus, links observed through shared hashtags may reflect actors’ intention to follow the same discussion as well as similar online communities discussing the marked topic that interest them [71]. Two organizations were thus connected by means of using common hashtags. On the other hand, organizational actors can mention other users by using the symbol “@” immediately followed by the target username within the text. Links observed through shared mention contacts may refer to actors’ intention to engage in the same conversation, build common connections, or draw awareness within the social media community. Similarly, two organizations were considered connected provided they share common mentions.

It is worth noting that the present study focused on investigating shared contacts as indirect forms of interorganizational communication activities for several practical reasons. First, sharing common contacts with other organizations can reflect how actors promptly take advantage of various social media features to facilitate information flow, enhance the likelihood of substantive collaboration, and engage with the larger stakeholder community [24]. This form of communication can be addressed as strategic communication since it demonstrates “the purposeful use of communication by an organization to fulfill its mission” ([72 p. 3]). Second, although actors sharing common contacts may not be aware of each other, the publicly exhibited mediators provide opportunities for actors to engage in a series of representational collaborative networks to draw awareness from the broader organizational community [73]. Moreover, the inspection mechanism afforded by the social media application ensures that contacted users receive a notification from social media alert when they are mentioned or replied to. It can be therefore assumed that contacted users are more likely to notice those communication activities which give rise to sustained inspection and continuous information exchange between organizational accounts. Third, online ECNs established on the basis of direct ties among organizational actors using a similar set of social media features (e.g., mentions, retweets) were found to be loosely connected with an extremely large proportion of isolated nodes. The highly skewed network structure together with the low network density were likely to bias the results of data analysis.

Several steps were then carried out to generate network matrices for two regions, respectively. First, all unique hashtags (N_SZ = 740; N_HK = 1244) and mentioned contacts (N_SZ = 974; N_HK = 373) used by organizational accounts were identified from the raw data (see S4 Table). The next step was to generate six two-mode matrices (3 × 2) for both regions to measure each established online ECN over a two-week period of observation. In these matrices, the rows represented organizational accounts while the columns represented unique hashtags and mentioned contacts, respectively. For all 12 binary network matrices, a “1” was entered into the cell when a given organization used a particular shared contact in their tweets during the response to Typhoon Mangkhut and a “0” was entered into the cell otherwise. In addition, the aggregate of shared contacts was not accounted due to the potential bias of multi-valued networks and the relatively short period of network observation. Finally, 12 two-mode matrices were converted to 24 one-mode square matrices. For the current concern, only those one-mode organization-by-organization matrices representing the shared communication relationships between two network actors were used for further analysis (see S1 and S2 Files).

Data analysis

Organizational attributes.

Three categorical attributes were defined and manually assigned to each sampled organizational account, including organization type, geographic scale, and geo-location. Specifically, network actors in both regions were categorized as to whether they belong to a government agency, a public institution (e.g., university, hospital, blood station), a for-profit business (e.g., state-owned, private, and foreign enterprises), an armed force (e.g., PLA Hong Kong Garrison, People’s Armed Police Forces at local levels), or an NGO/NPO. The addition of armed forces to this categorization reflects unique features of China’s EMS that were not considered in prior network studies conducted under the federal system. Geographic scale was coded as the coverage within which an organization operates following five categories: central-level, provincial-level, municipal-level, district/county-level, and below district/county-level. As an ECN incorporates social sectors other than government agencies, geographic scale proves to be more suitable than government jurisdiction in describing the footprint of a network actor’s emergency management practice [8, 74]. It should be noted that the geographic scale was not accounted for the Hong Kong data because nearly all identified actors operate within the HKSAR. In addition, the geo-location was measured by the place in which organizational actors are located or headquartered. Network actors were classified into five groups in line with the jurisdictional division of Shenzhen: Shenzhen East (Longgang District, Yantian District, Pingshan District, Dapeng New Area, and Shen-Shan Special Cooperation Zone), Shenzhen West (Baoan District, Guangming District, and Nanshan District), Shenzhen South (Futian District and Luohu District), Shenzhen North (Longhua District), and non-local; and five groups according to the jurisdictional division of Hong Kong: Hong Kong Island, New Territories, Kowloon Peninsula, Islands District, and non-local.

The organizational attribute files were then transformed to network matrices using the attribute/matrix converting function afforded in UCINET [75]. Table 2 presents the attributes of identified organizational actors in Shenzhen and Hong Kong.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Attributes of online participating organizations in Shenzhen and Hong Kong.

https://doi.org/10.1371/journal.pone.0289277.t002

Descriptive statistics

Whole network characteristics of online ECNs across both EMSs were firstly reported. As shown in Table 3, both hashtag networks and mention networks within the centralized EMS were consistently denser than their counterparts in the decentralized EMS as the disaster evolved over time, suggesting that in Shenzhen, network actors forged more online communication ties with other organizations throughout the emergency management cycle. In both EMSs, online recovery networks were sparser than online preparedness networks and online response networks. In the centralized EMS, the degree centralization index of online recovery networks was lower than that of preparedness and response networks. In contrast, the opposite was true for the decentralized EMS wherein the centralization score of the recovery network characterized by shared hashtags was slightly higher than that of its two counterparts. This implies that the network tendency to center around a couple of key actors may be subject to structural variations.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Descriptive statistics of online ECNs.

https://doi.org/10.1371/journal.pone.0289277.t003

Structural characteristics of online ECNs were visualized for Shenzhen data and Hong Kong data in Figs 1 and 2, respectively. Longitudinal variations can be more explicitly observed that online preparedness and response networks were denser than recovery networks across both EMSs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Structural characteristics of two types of online networks in the centralized EMS in Shenzhen.

Isolated nodes without any shared hashtag or mention contact with other network actors are excluded for visualization parsimony.

https://doi.org/10.1371/journal.pone.0289277.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Structural characteristics of two types of online networks in the decentralized EMS in Hong Kong.

Isolated nodes without any shared hashtag or mention contact with other network actors are excluded for visualization parsimony.

https://doi.org/10.1371/journal.pone.0289277.g002

QAP correlations among multiple networks

QAP correlations among online preparedness networks, response networks, and recovery networks were reported to cast a preliminary spotlight on the longitudinal multiplexity of online ECNs. Since the network data (i.e., one-mode matrices) analyzed in this study are binary, Jaccard correlation coefficients were used to evaluate the QAP correlations among multiple networks. Specifically, Jaccard (similarity) coefficients measure the similarity of two networks by comparing the total number of common dyads to the total number of distinct dyads [76]. The coefficient values range from 0 to 1 and the higher the coefficients, the more similar the two networks. As shown in Table 4, online preparedness ties were positively associated with online collaborative ties during disaster response and immediate recovery in both EMSs. Within the centralized EMS in Shenzhen, the Jaccard correlation coefficients between preparedness and response networks and between preparedness and recovery networks were 0.571 (hashtag)/0.370 (mention) and 0.255 (hashtag)/0.339 (mention), respectively. Correlations between response and recovery networks were 0.265 (hashtag)/0.255 (mention). All these coefficients were significantly positive at .001 level. Similarly, the Jaccard correlation coefficients were significantly positive between preparedness networks and response networks (0.614/0.618), between preparedness networks and recovery networks (0.422/0.382), and between response networks and recovery networks (0.381/0.531) within the decentralized EMS in Hong Kong. By and large, compared with the centralized EMS, correlations between multiple networks were relatively stronger in the decentralized EMS. The relationship between online preparedness and response networks was consistently higher than two other comparisons of network similarities. Nonetheless, the contrast of network similarities varied depending on the forms of online communication activities (i.e., shared hashtag or mention contacts).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. QAP correlations between three types of online ECNs.

https://doi.org/10.1371/journal.pone.0289277.t004

MR-QAP analyses

Although QAP correlations appear easy and straightforward to interpret, they may present erroneous results due to their sensitivity to small sample sizes. Therefore, MR-QAP was conducted to further test the evolving correlations among multiple online ECNs. By doing so, the effects of correlations between preparedness and response networks can be separated and more importantly, the control variables (i.e., homophily effects) can be considered to obtain a more comprehensive picture of network dynamics. Double Semi-Partialing (DSP) permutation method, which is robust against a broad spectrum of statistical biases [77], was used to reduce the sensitivity of MR-QAP analyses to network autocorrelation and collinearity.

H1 posits that organizational actors’ online preparedness networks positively correlate with their online response networks. As presented in Table 5, online collaborative ties within response networks were set as dependent variable in Model I. The regression coefficients of online preparedness ties for predicting response ties were 0.643 (hashtag)/0.463 (mention) within the centralized EMS and 0.724 (hashtag)/0.748 (mention) within the decentralized EMS. All coefficients were significantly positive at .001 level, thus supporting H1 that online collaborative ties forged at the preparedness phase were conducive to the formation of collaborative ties during the response to Typhoon Mangkhut.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Model I: MR-QAP results for predicting online response networks.

https://doi.org/10.1371/journal.pone.0289277.t005

H2 examines the association between online preparedness networks and recovery networks. As shown in Table 6, online collaborative ties within recovery networks were set as dependent variable in Model II. The regression coefficients of online preparedness ties were 0.169 (hashtag)/0.357 (mention) within centralized networks and 0.408 (hashtag) within decentralized networks and significantly positive at .001 level. However, the coefficient of preparedness ties built upon shared mention contacts was not statistically significant for the Hong Kong data. Hence, H2 was partially supported.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Model II: MR-QAP results for predicting online recovery networks.

https://doi.org/10.1371/journal.pone.0289277.t006

H3 assumes a positive relationship between online response networks and recovery networks. As Table 6 shows, the regression coefficients of collaborative ties formed at the stage of disaster response for predicting the collaborative ties during immediate disaster recovery were 0.214 (hashtag)/0.182 (mention) within the centralized system and 0.233 (hashtag)/0.632 (mention) within the decentralized system. All reported coefficients were significantly positive at .001 level. Therefore, H3 was supported that organizational actors’ online response ties contributed to the formation of post-disaster recovery ties.

H4 assumes that the correlations between online response and recovery networks are higher than those between the other two pairs of online networks. As shown in Model I together with Model II, the regression coefficients of preparedness ties for predicting response ties were consistently higher than those of preparedness ties for predicting recovery ties and of response ties for predicting recovery ties. H4 was thus unsupported.

Structural constraints represented by political-administrative arrangements were further examined as a context-specific factor in uncovering the logics of online network formation and evolution. Descriptive statistics presented in Table 3 showed that with the exception of the hashtag network built during disaster recovery, centralization scores of all other five types of online networks were higher for the Shenzhen data than those for the Hong Kong data. This contrast corroborated the current division of centralized versus decentralized coordination mechanism against the social media context. That is, the structural constraints for emergency collaboration on the ground can apply to online interorganizational interactions such that Shenzhen demonstrated a more centralized online network structure, whereas Hong Kong exhibited a more decentralized network pattern for emergency management.

More specifically, H5 assumes that online preparedness ties within the centralized EMS have stronger predictive power in the formation of online response ties than do their counterparts within the decentralized EMS. As presented in Table 5 (Model I), the regression coefficients of preparedness ties for predicting response ties within the centralized system were 0.643 (hashtag)/0.463 (mention), which were lower than their counterparts within the decentralized system (0.724 for the hashtag network; 0.748 for the mention network). Therefore, H5 was not supported.

H6 posits that online response networks within the decentralized EMS have stronger predictive power in the formation of online recovery networks than do their counterparts within the centralized EMS. As shown in Table 6 (Model II), compared with the centralized system within which regression coefficients were 0.214 for the hashtag network and 0.182 for the mention network, online collaborative ties forged at the response phase had higher predictive power in the formation of post-disaster recovery ties within the decentralized system (regression coefficients were 0.233 for the hashtag network and 0.632 for the mention network). As thus, H6 was supported.

Regarding the homophily variables (RQ1) which were controlled for the current concern, regression coefficients of the organization type similarity with the centralized EMS were 0.082 (hashtag)/0.155 (mention) for Model I and 0.141 (hashtag)/0.117 (mention) for Model II. All these coefficients were significantly positive at .01 level, suggesting that organizational actors of the same type may share common goals and network constraints and thus are more likely to collaborate with one another in mobilizing resources and exchanging information in the disaster response and post-disaster recovery. The openness and transparency of communication activities on social media applications may make it easier for communication managers to seek partners sharing similar organizational values and identities, therefore strengthening such homophily effects. However, the opposite was evidenced for the Hong Kong data that the regression coefficients of the organization type similarity within the decentralized system were mostly negative. This indicates that throughout the Typhoon Mangkhut, online interorganizational networks in Hong Kong may exhibit a more horizontal structural pattern with more cross-sectoral information sharing and resource mobilization than their counterparts in Shenzhen.

Practical implications

Emergency managers and practitioners can also find some takeaways from this study. First, the significance of preexisting relationships also extends to emergency management practices. After the Great Flood of 1993 that occurred in Midwestern areas of the U.S., a comparison of five communities indicated that immediate aid and assistance at the response stage were available from public and private sources in all communities, but the situation differed greatly during disaster recovery [82]. In communities with preexisting interorganizational networks, local leaders and organizations actively shouldered the responsibility of reconstruction and rebuilding, thus ensuring a sustainable post-disaster recovery. In the absence of preestablished collaborative ties, however, self-organized emergency management teams failed to retain residents’ enthusiasm of collaboration as the flood crisis decayed because they lacked collective learning experiences which essentially contributed to the reconstruction progress. Results from the present study once again uphold the importance of playing out in early emergency preparedness because a wide variety of joint planning activities, training programs, and drills may help enhance collective learning and develop online collaborative ties during disaster response and recovery.

Second, there has long been a heated debate among emergency researchers and practitioners about whether a centralized/coercive power structure is more appropriate for managing emergencies or a decentralized/voluntary power structure prevails (e.g., [53, 83]). This study complements prior research in clarifying the comparison between centralized and decentralized structural constraints by furthering a contingency perspective to explore their varying effects on online network changes as the disaster evolves over time. The decentralized network structure appears more suitable to manage the disaster concerned in this study because it incurs lower autonomy costs while illustrating more effectiveness in facilitating online networks changes during both transitional periods. Nonetheless, the findings exhibit that the correlations between preparedness and response networks and between response and recovery networks are strong in both EMSs, even though they are stronger in the decentralized system. Also, it remains neither clear nor systematic about the extent to which the cross-platform divergence between Twitter and Weibo would influence the results of this study. Hence, more efforts are expected to be made to explicate how each system copes with collective action problems differently against the social media environment and what contributes to the priority of one system over the other.

Moreover, global crises represented by the Covid-19 pandemic have also posed rapidly shifting challenges for leadership, which refers to the ability of top executives and senior managers to unite and persuade a group of individuals and organizations to collaborate and coordinate effectively in the pursuit of collective goals. These emerging challenges, which require leaders in all industries to strategically pivot amid disrupted operations, to promptly embrace a social media mindset, and to effectively manage scattered and remote workforce, are likely to widen the leadership gap between leaders who are purposefully prepared for constantly changing circumstances and those who are not. Findings obtained from this study have crucial implications for leaders to reassess priorities across different phases of a crisis and continuously adapt to a climate of purpose and commitment. Learning and development (L&D) professionals and teams are also expected to help cultivate specific capabilities and skills for response organizations to close the leadership gap in the face of extreme complexities and uncertainties.

Limitations and future directions

Several limitations of this study should be pointed out. First, the temporal dimension represented by decisions about how to demarcate three disaster phases may bias online ECN boundaries. Although the temporal cutting points were set in line with the case specificity and personal observations of Typhoon Mangkhut, it should be recognized that network dynamics are subject to how online ECNs are operationalized over time. For example, it is plausible that a in a catastrophic event with a much larger scale than Mangkhut, the frequency, durability, and distribution of online interorganizational interactions across different disaster phases would differ greatly such that actors tend to forge ties to foster a sustainable post-disaster recovery, and more importantly, the correlation between formal networks built during disaster response and recovery would be greater than that between informal and formal networks. Future studies are expected to either conduct sensitivity tests to remedy this limitation or employ a more innovative research design in collecting longitudinal online network data.

Second, the current investigation primarily focuses on collective efforts at (inter)organizational levels, while overlooking individual desires and strategic motivations of emergency managers, especially frontline communicators who are actually responsible for managing routinized emergency operations and making improvised decisions on behalf of their organizations and departments on social media platforms. Apart from organizational preferences and response capacities, it merits attention that communication managers’ personal mindset in aligning communication objectives with collaborative goals during emergencies may influence the types of strategy they use to shape collective actions [84]. In this regard, further studies can benefit from retaining the present explanatory pattern but maintaining a certain level of openness and adaptability to incorporate multiple individual motivations.

Finally, despite the efforts made to examine the longitudinal multiplexity and exogenous factors (i.e., EMSs, homophily effects) possibly influencing actors’ coordination mechanism choices, the present investigation suffers from lacking empirical evidence of the effectiveness and performance of collective action, and accordingly, how network outcomes can be improved. To address challenges posed by data availability and an overall lack of direct measures, an emerging body of literature has utilized simulation-based approaches to evaluate network performance and outcomes with crucial indicators such as frequency, speed, and efficacy of information dissemination [85, 86]. Such techniques can broaden the scope of operationalization of network performance and lend researchers methodological support that supplements traditional approaches (e.g., structured/semi-structured survey, text analysis, interview).