Unpacking the Network Processes and Outcomes of Online and Offline Humanitarian Collaboration

Abstract

Employing a bona fide network perspective, this study investigates the network processes and outcomes of organizational collaborative networks before and following Typhoon Haiyan, taking into account the influences of network factors, organizational attributes, and environmental exigencies. The analysis from an online survey with relief organizations and those organizations’ Twitter data showed the consistent influence of past relationships on the formation of subsequent relationships after the disaster. In the on-the-ground network, a few highly active organizations stood out and engaging in multiple modes of communication with resource contacts was seen as an adaptive practice that helped organizations to build resource ties after the typhoon. In the online domain, organizations developed post-typhoon networks by means of becoming directly linked to one another and becoming equally resourceful in building their ties. In addition, different forms of resilience were observed as outcomes of collaborative networks. Findings of this study present theoretical and practical implications by unveiling the network dynamics of contemporary humanitarian actions.

Keywords

organizational collaboration network analysis disaster response social media voluntary sector

Disaster represents a condition under which both formal and informal methods of collaboration are likely to materialize. This is shown to be true for the voluntary sectors, which are part of the civil emergency response system for pre-disaster planning, disaster response, and post-disaster aid (Adger, Quinn, Lorenzoni, Murphy, & Sweeney, 2012). Voluntary sectors may include societal institutions (e.g., religious groups, professional associations), non-governmental organizations (NGOs), non-profit organizations (NPOs), media, and/or business communities (Taylor & Doerfel, 2011). Citizens may resort to humanitarian aid by voluntary sectors due to a lack of confidence in government support. In addition, online and social media have been an important foundation of various self-organized and voluntary emergency efforts in social movements and disaster contexts (Starbird & Palen, 2012). After Hurricane Katrina, for example, a wiki was used by people around the world to mobilize and coordinate rescue efforts (Majchrzak, Jarvenpaa, & Hollingshead, 2007). Similarly, NPOs and news organizations also adopted social media to disseminate information after the 2010 Haiti earthquake (Muralidharan, Rasmussen, Patterson, & Shin, 2011).

While these aforementioned studies have documented the usefulness of online and social media for collaborative emergency responses, most research on disaster and social media is focused on how individuals and relief organizations adopt new technologies for online emergency response (Landwehr & Carley, 2014). Little is known about the process of how these collective relief efforts come into being and change over time outside the online domain. Moreover, most studies on organized collaboration in the voluntary sectors rely on one-time observations (e.g., Guo & Acar, 2005; Nolte & Boenigk, 2013), with a few exceptions such as Shumate, Fulk, and Monge (2005) and Lee and Monge (2011). A longitudinal and systematic understanding is imperative because it can identify whether and how the organized efforts are robust and sustainable beyond a one-time event, across both online and offline domains.

To address these gaps, the study examines the relief organizations involved in Typhoon Haiyan in the Philippines, with the aim of understanding the evolving processes of those relief organizations’ different forms of collaborative networks across online and offline domains. Typhoon Haiyan hit the Philippines on November 6-9, 2013, causing physical devastation as well as 6,000 deaths; more than 16,000,000 people and 57 cities across the nation were affected (National Disaster Risk Reduction and Management Council, 2014). In the disaster-prone Philippines, international or cross-geographic relief efforts are particularly salient (Luna, 2001). Following the 2011 Haiti earthquake and 2012 Hurricane Sandy, the use of social media has become an important source for organized relief efforts (Landwehr & Carley, 2014). Together, these facts establish Typhoon Haiyan as a solid case to investigate the contemporary form of organizational collaborative networks for humanitarian response actions.

This study aims to answer the following questions. First, what contributes to the formation of organizational collaborative networks in the disaster response context, and how are these collaborative networks sustained over time? Second, what are the consequences that result from sustained collaborative networks? Because of the study’s consideration of both environmental exigencies and network factors in organizational collaboration, we use Cooper and Shumate’s (2012) bona fide network perspective as the conceptual foundation to examine the network processes and outcomes of relief organizations’ collaborative networks across online and offline domains, taking into account the influences of network factors, organizational attributes, and environmental exigencies. Specifically, in this study, we examine two types of organizational collaborative networks in the disaster response context: the resource network constructed by the relief organizations following Haiyan and those relief organizations’ online humanitarian networks built and maintained on Twitter before and after Haiyan. In developing the hypotheses and research questions, we differentiate these two networks as the resource network on the ground and online humanitarian networks, respectively.

This study contributes to communication research and practice in the following ways. First, drawing on the bona fide network perspective and the concepts of embeddedness and homophily, this study examines both the structural and attribute-based influences on the formation and evolution of disaster relief networks across online and offline domains. Rather than focusing on the attribute- or geography-based enablers or constraints, findings of this study suggest the importance of relationship building in the organizing of post-disaster relief networks. Second, by integrating the bona fide network perspective and the theory of media multiplexity, this study enriches the conceptualization of multimodal communication as an emergent and robust practice enacted by relief organizations after a disaster. Findings of this study revealed that engaging in multimodal communication with external resource contacts facilitated connections to the same resource contact types. Third, this study proposes to examine resilience as a type of outcome for humanitarian collaborative networks, which is especially salient in disaster-prone regions and countries. Last, this study makes a major methodological contribution by using a mixed-methods approach and encompassing cross-sectional and longitudinal data in examining bona fide humanitarian organizing networks.

This article first reviews the bona fide network perspective and related theories, then explains the development of hypotheses and research questions. Detailed procedures of how to obtain and process multiple sources of data are provided in the “Method” section. The results of the analyses are presented, with the aim of enhancing theoretical and societal understandings about the evolving collaborative networks across online and offline domains in the context of disaster response.

Network Mechanisms and Outcomes of Humanitarian Collaboration

As an extension of the bona fide group perspective (Putnam & Stohl, 1990), the bona fide network perspective emphasizes the following characteristics: (1) network boundaries are blurred; (2) inter-organizational networks are embedded in a complex environment; (3) inter-organizational networks are multifaceted and multiplex, and different relationships influence one another; (4) networks are dynamically restructured over time; and (5) outcomes of networks occur at both individual organization and whole network levels. Essentially, the bona fide network perspective posits that inter-organizational collaboration is influenced by environmental exigencies and network factors (Cooper & Shumate, 2012). Disaster can be a type of triggering event to affect these collaborations. Similar to the situation of social movements, in the disaster context, having access to tangible (e.g., goods, money, volunteers) and intangible (e.g., information) resources is necessary, and the ability to mobilize these resources usually comes in the form of connecting to pre-existing networks and reliance on the timely support from other organizations (McCarthy & Zald, 1977).

Applying this bona fide network perspective, Cooper and Shumate (2012) investigated the antecedents (types of NGOs), processes (environmental constraints, multiplex relationships), and outcomes of collaboration at organizational (centrality, offering more and varied services to clients) and network levels (gains in systemic capacity). Yet, as much as the bona fide network perspective offers a flexible and systematic way of accounting for diverse forms of organizational collaboration, it does not delve into or provide theoretical reasons for the underlying network processes of organizational collaborative networks. To fill the void, we draw on the concepts of embeddedness, homophily, and media multiplexity to explain the network factors that influence the formation of bona fide collaborative networks among disaster relief organizations. We follow this with the delineation of network outcomes as a result of collaboration.

Embeddedness and Organizational Collaborative Networks

In the disaster context, prior network relationships among organizations help the mobilization of emergent disaster response networks even though those relationships are not emergency related (Robinson, Berrett, & Stone, 2006). The influence of network relationships on social and organizational behaviors corresponds with the concept of embeddedness, which includes relational, structural, and positional embeddedness (Granovetter, 1992; Gulati, 1995; Gulati & Gargiulo, 1999). Essentially, embeddedness is closely linked to the notion of social capital, which refers to the resources inherent in social relationships that are accessed or mobilized for purposive actions (Lin, 1999). Similar to embeddedness, social capital may embody relational and structural dimensions of connections—how well organizations know each other and who organizations know and in what ways (Nahapiet & Ghoshal, 1998). Because of the lack of direct measures, we focus on structural and positional embeddedness in this study. We define these two types of embeddedness as follows, and make a theoretical linkage between each type of embeddedness and the relevant communicative and relational dynamics involved in organizational collaborative networks.

Operationally, structural embeddedness represents the extent to which an organization’s structural position presents opportunities for new tie formation. In response to a disaster event, organizations often tapped into their previous networks; knowing that someone in the network was safe and ready to resume operations motivated them to rebuild their own operations as well (Doerfel, Lai, & Chewning, 2010). Research has consistently found that organizations’ past relationships are likely to predict the formation of future ties (Lee & Monge, 2011; Shumate et al., 2005). The first two hypotheses thus examine the influence of relief organizations’ structural embeddedness in the post-disaster network, or the influence of previously established relationships on subsequent relationship building.

Hypothesis 1 (H1): The relief organizations that have established ties with resource contacts on the ground are more likely to continue those resource ties in the later phase of disaster relief than change to new resource ties.

Hypothesis 2 (H2): The relief organizations that have connected to each other online are more likely to continue those connections online in the later phase of disaster relief than change to new connections.

Structural embeddedness is also observed through the dynamics of closure, that is, organizations with shared partners from past ties become embedded and central in a densely clustered group (Granovetter, 1992; Gulati & Gargiulo, 1999). In particular, being embedded and central in a densely clustered group has been linked to corporate success and the ability of an organization to face crisis or uncertainty (Gabbay, Talmud, & Raz, 2001). Such a clustering network structure is also believed to help a focal organization guard against threats to survival (Wells, Lee, & Alexander, 2001). In disaster response, certain types of resources such as materials, volunteers, or money are often in high demand and a few entities are likely to fulfill the need for these resources. As such, relief organizations tend to be connected to those similar few entities for operations. We thus hypothesize that a small group of active relief organizations are likely to cluster together through their pattern of building connections to a few common popular resource contact types.

Hypothesis 3 (H3): The relief organizations’ resource network on the ground exhibits a structural tendency toward closure such that active relief organizations are more likely to be connected to a few popular resource contact types than would be predicted by chance alone.

When organizations decide who to connect with, they usually first resort to their current partners, which represents the dynamics of triadic closure. Atouba and Shumate (2010) showed that in the international development organization network, organizations tended to collaborate with their partners’ partners. Shumate et al.’s (2005) study found that the patterns of the alliances of HIV-AIDS networks were predicted by common ties to inter-governmental organizations (IGOs) such as the United Nations, as well as prior relationships among organizations. The use of information and communication technologies (ICTs), especially social media, may create opportunities for NPOs and NGOs to go beyond their specific regions and easily engage in conversations with other organizations directly. But relatively little empirical evidence has been provided concerning whether and how the pattern of transitivity manifests in the online humanitarian networks. One of the few exceptions close to the topic of this research is Shumate’s (2012) work, which showed that NGOs tended to hyperlink to other NGOs that shared a common partner. In light of this, we hypothesize that the argument of triadic closure also applies to the online humanitarian network.

Hypothesis 4 (H4): The online humanitarian network exhibits a structural tendency toward triadic closure such that relief organizations with shared partners are more likely to be directly connected to one another than would be predicted by chance alone.

Organizational collaboration may also be driven by positional embeddedness, that is, the positions or roles an organization occupies in the network (Gulati & Gargiulo, 1999). For example, centrality is usually associated with informational benefits, because organizations having more connections are likely to have a better understanding of the network situation (Freeman, 1979). These central organizations are relatively visible, signaling their attractiveness as potential partners to others in the network (Gulati & Gargiulo, 1999). In fact, organizations with a higher level of centrality are likely to form different types of relationships with one another (Lee & Monge, 2011). Central organizations are perceived as more influential and important than peripheral organizations (Doerfel & Taylor, 2004). As the central players dominate the opportunities, the network may evolve into a core-periphery structure, where a few nodes are highly active and connected while the rest are loosely connected to one another.

In sum, when an organization decides who to follow in a network, they may look for the most central and popular organizations. Translating the principle of positional embeddedness into relief organizations’ collaborative networks, it is possible that certain active relief organizations are more likely to seek out and acquire resources from more contacts, while others focus on a limited set of resource contacts. Moreover, a core-periphery network structure reflects the patterns of structural differentiation in a network, which was found to be useful for the formation of new inter-organizational ties (Gulati & Gargiulo, 1999). In the disaster context, such core-periphery network structure is not uncommon. Emergent disaster relief networks tended to exhibit a two-tier system, where a relatively core and stable sector (governmental organizations) coexisted with a relatively transient and peripheral sector (NGOs, NPOs, and private organizations; Robinson, Eller, Gall, & Gerber, 2013). Moore, Eng, and Daniel (2003) studied NGOs involved in the flood relief operations in Mozambique and found that international NGOs tended to occupy more central positions in the inter-organizational relief network while local NGOs were relatively peripheral in the relief network. The following three hypotheses are developed to examine the resource and online humanitarian networks’ tendency toward a core-periphery structure.

Hypothesis 5a (H5a): In the resource network on the ground, a few active relief organizations that have connections with different types of resource contacts are more likely to emerge than would be predicted by chance alone.

Hypothesis 5b (H5b): In the resource network on the ground, a few popular resource contact types that have connections with different relief organizations are more likely to emerge than would be predicted by chance alone.

Hypothesis 6 (H6): In the online humanitarian network, there is a concentration of a few active relief organizations while others are more equally distributed.

Attribute-Based and Socio-Technical Homophily

In addition to relational factors, the homophily effect is salient in organizational collaboration because similarity in organizational attributes breeds connections (McPherson, Smith-Lovin, & Cook, 2001). For example, Atouba and Shumate (2014) examined homophily manifested in attribute-based (same status, similar founding dates), geography-based (headquartered in the same regions), and institutional/structural aspects (common funding partners), and studied how these influenced the formation of inter-organizational collaborative ties among infectious diseases international non-governmental organizations (INGOs). Their study supported the principle of homophily because those INGOs tended to collaborate with one another when they shared the same status, had similar founding dates, were headquartered in the same geographic regions, and had common funding partners. Similarly, Kim and Barnett’s (2000) study on the international telecommunications industry also showed the influence of geographic proximity on the formation of the international telecommunications system. Singer and Kegler (2004) investigated the role of inter-organizational collaborative networks in enhancing community capacity for community health intervention. They found that the intensity and density of inter-organizational networks were high among organizations of similar types, yet decreased as the collaboration moved through stages.

In the meantime, a growing line of work attempts to draw connections between the online networks of voluntary sectors and their offline collective actions. For example, Gillan’s (2009) study on the U.K. anti-war movement showed that the online network pattern tended to be collaborative and diverse, while the offline counterpart was more divided and limited. Furthermore, international NGOs are usually divided into two groups: North NGOs and South NGOs. The former refers to those INGOs headquartered in developed countries (e.g., North America, Western Europe, Australia/New Zealand) while the latter represents those groups headquartered in developing countries (e.g., Africa, Central and South America). Yet so far, no conclusive evidence shows that technology use overcomes the North/South divide (Shumate & Dewitt, 2008).

Moreover, most of these studies linking online and offline domains focus on the hyperlink network among NGOs’ websites (e.g., Yang, 2013). It is expected that the organizing networks on social media may exhibit a different pattern of overcoming these existing divides among humanitarian organizations. For example, the feature of following other users on Twitter makes it easier to build connections despite organizational or geographic differences, which is social media’s affordance of association (Treem & Leonardi, 2012). Due to the lack of conclusive claims about the effect of geographic or type similarity on online organizational collaborative networks, we develop a general research question.

Research Question 1 (RQ1): How does the principle of homophily in the form of similarity in organizational type and geographic location affect the evolved online humanitarian network?

While the bona fide network perspective suggests that technology plays a critical role in facilitating the process of collaboration between organizations, no specific conceptualizations of how technology is related to the network processes are proposed. In this regard, the theory of media multiplexity helps explain the process of technology use in the organizing of bona fide networks. Extending the concept of relational multiplexity to technology use, the theory of media multiplexity suggests that the more frequently two people communicate, the stronger the tie, and the more types of media they use (e.g., face-to-face talk, email; Haythornthwaite & Wellman, 1998). Instead of focusing on the attributes of media or individual characteristics, the framework of media multiplexity emphasizes the social context in which a medium is used (Haythornthwaite, 2002).

In this article, the use of multiple technologies for communication is termed “multimodal communication.” In the disaster context, using multiple means of communication to create or maintain relationships can be seen as a type of adaptive relational behavior initiated by organizations in response to environmental exigencies. Because multimodal communication with resource contacts involves engagement in technology use and the social context, we consider organizations with similar levels of multimodal communication with external contacts as exhibiting socio-technical homophily. Specifically, because resource contacts have preferred modes of communication, in consideration of such preferences, relief organizations may adopt this practice of multimodal communication and exhibit similarity in the level of multimodal communication when connecting to similar types of resource contacts.

Hypothesis 7 (H7): Relief organizations on the ground that have similar levels of external multimodal communication are more likely to share similar resource contact types than organizations that have different levels of external multimodal communication.

Along with relational factors, the bona fide network perspective emphasizes the environmental influences on organizational collaborative networks. In emergent and disaster contexts, organizations directly affected by the exigencies are likely to mobilize their accumulated social capital in order to receive resources (Doerfel et al., 2010). In particular, response organizations in the affected area are likely to reach out and be reached by organizations of different types (Kapucu, 2005). It is expected that relief organizations in the affected country are more likely to build connections with resource contacts. Similarly, if organizations cover a wider scope of area for relief operations, they are inclined to seek out resource support. Furthermore, relief organizations located within the affected country may have similar resource needs, and are thus likely to reach out to similar entities that can provide resources needed for relief operations.

Hypothesis 8a (H8a): Relief organizations operating in the affected country are more likely to engage in resource ties with other entities than those organizations operating outside of the country.

Hypothesis 8b (H8b): Relief organizations covering a larger scope of relief operations on the ground are more likely to engage in resource ties with other entities than those organizations covering a smaller scope of operations.

Hypothesis 9 (H9): Relief organizations located within the affected country are more likely to have ties with similar resource contact types than those organizations operating outside of the country.

Outcomes of Collaborative Networks

The bona fide network perspective examines the outcomes of inter-organizational collaboration realized at organizational and network levels (Cooper & Shumate, 2012). In the disaster context, Nolte and Boenigk (2013) proposed a process model examining public–non-profit collaborative networks as well as looking at operationalized network outcomes in forms ranging from stronger inter-organizational relationships, expanded scope of services, and growth of the network. Moore et al. (2003) studied NGOs involved in the flood relief operations in Mozambique and found that international NGOs tended to have a higher impact in terms of having more beneficiaries, while local NGOs were relatively peripheral in the relief network and had a lower number of beneficiaries in the affected communities. Simo and Bies (2007) examined the usefulness of cross-sector collaborative relief efforts between non-profits and other organizations (e.g., government, business, the public) in helping community recovery and rebuilding after disasters.

In this study, we focus on examining network outcomes in the form of resilience. Resilience refers to the ability of individuals, organizations, and a system of organizations to communicate and reorganize when faced with disruptive challenges (Chewning, Lai, & Doerfel, 2013). Resilience is a suitable concept for observing network outcomes of organizational collaboration in the disaster context because organizations’ sustainable collective relief actions may help the affected communities to respond and recover more effectively after a disaster (Simo & Bies, 2007). The last research question is developed to address the network outcomes of relief organizations across online and offline domains.

Research Question 2 (RQ2): What are the realized network outcomes of relief organizations’ collaborative networks in the form of resilience across online and offline domains after a disaster?

Method

This study uses a mixed-methods approach, consisting of an online survey with relief organizations and the organizations’ longitudinal Twitter data before and after Typhoon Haiyan. Organizations were chosen mainly based on a news article by CNN and the United Nations Office for Coordination of Humanitarian Affairs’ (OCHA) website, which provides a complete list of organizations involved in Haiyan relief efforts (N = 264).¹ After receiving the institutional review board (IRB) approval from the first author’s affiliated institution, we filtered out duplicate organizations and sent email invitations to all 180 organizations on the OCHA list to participate in the survey from March to May 2014. In OCHA’s directory, details about an organization’s contact person (e.g., email, job title) were provided. To ensure the reliability and validity of the response, we intentionally selected the organizational contacts who were in charge of the relief efforts on the ground (e.g., country director, humanitarian aid coordinator, emergency coordinator) to complete the survey. In the invitation email, we provided the scope and the type of the questions covered in the survey and asked the contact to forward it to appropriate personnel if he or she was not directly in charge of the emergency management and community engagement on the ground.

Among the 180 invitations sent out, 41 organizations responded (response rate = 22.8%).² The Twitter data of these 41 organizations were also retrieved. Using the additional organizational contacts mentioned by these participating organizations in the survey,³ the scope of the Twitter data was expanded accordingly. After this adjustment, the final Twitter dataset consisted of 70 organizations. Table 1 presents the details about the organizations under study for both the survey and Twitter data.

Table 1.

Information About the Organizations for Survey and Twitter Data.

Type	Survey	Twitter
Government agency	1	1
Business	1	3
NPO	9	14
NGO	29	45
IGO	1	6
Coalition	—	1
Location
Philippines	17	17
Asia Pacific region	3	1
Europe	10	16
North America	11	36
Total	41	70

Note. NPO = non-profit organization; NGO = non-governmental organization; IGO = inter-governmental organization.

Online Survey With Relief Organizations

All 41 participating organizations were asked whether they received resources of any type (e.g., informational, financial, physical) from the following 10 sources in each of the two phases of their relief and recovery support with the community (0 = no, 1 = yes): individuals from the affected community, NPOs/NGOs from the affected community, individuals outside of the affected community, NPOs/NGOs outside of the affected community, businesses from the affected community, businesses outside of the affected community, local government or public agencies, provincial government or public agencies, news media from the Philippines, and news media from outside of the Philippines. These 10 resource contact types were developed by adapting existing research (e.g., Doerfel et al., 2010; Lai & Chewning, 2014). Based on the responses, two rectangular two-mode matrices were created where the rows represented the 41 relief organizations and the columns were the 10 resource contact types. The matrices were to examine the relief organizations’ resource network immediately after Haiyan (Time 1) and within three months after Haiyan (Time 2). We chose to focus on resource receipt in this study because for relief organizations, receiving resources from both inside and outside the affected area is particularly important for relief operations on the ground (Shepherd & Williams, 2014). Table 2 presents the descriptive statistics of the 10 resource contact types.

Table 2.

Descriptive Statistics of the 10 Resource Contact Types.

Frequency distribution of the resource contact types	Time 1	Time 2
Individuals from the affected community	19	18
NPOs/NGOs from the affected community	13	12
Individuals outside of the affected community	22	18
NPOs/NGOs outside of the affected community	23	20
Businesses from the affected community	9	12
Businesses outside of the affected community	16	16
Local government or public agencies	26	23
Provincial government or public agencies	21	22
News media from the Philippines	15	11
News media from outside of the Philippines	19	13
M	18.3	16.5
SD	5.10	4.38
No. of resource contacts each organization had
M	4.46	4.02
SD	3.27	3.81

Note. NPO = non-profit organization; NGO = non-governmental organization.

Four attribute factors were measured in the following ways. The two-mode network at Time 1 was considered as a dyadic attribute to assess the effect of the resource network at an earlier time on the network at a later time. In measuring the binary variable of geographic location, organizations whose headquarters are in the Philippines were coded 1 (n = 17); those not in the Philippines were coded 0 (n = 24). For categorical estimation, we recoded the variable of geographic location, where those organizations with headquarters in the Philippines were coded 1 and others coded 2. In creating the binary variable of scope of relief action, two steps of data processing were implemented. Typhoon Haiyan mainly affected five regions of the Philippines (Regions 4-8). In the survey, organizations were asked to report the region(s) where they provided relief support. The identified regions were counted, and classified organizations whose operations covered only one judiciary region were coded 0 (n = 24) and organizations covering two or more regions were coded 1 (n = 16).

Organizations’ external coordination was measured by the use the following means of communication: face-to-face, email, texting, voice phone calls, social media, websites, and other (e.g., bank transfers, Skype; 0 = no, 1 = yes). A summed scale of level of external multimodal communication was created, ranging from 1 to 7 at Time 1 (M = 4.81, SD = 1.94) and Time 2 (M = 4.83, SD = 1.86). The variable of multimodal communication was then dichotomized where code 1 described the organizations with a lower level of multimodal communication (those using five or fewer modes of communication with external contacts; Time 1, n = 15; Time 2, n = 14) and code 2 represented those with a higher level of multimodal communication (using more than five modes of communication; Time 1, n = 22; Time 2, n = 17). Two categorical attributes, external multimodal communication at Time 1 and Time 2, were created.

Collection and Processing of Twitter Data

To examine the evolution of organizations’ network on social media, the official Twitter pages of these 70 organizations for a 4-month period starting one month before the typhoon to three months after (October 12, 2013, to January 31, 2014) were crawled and downloaded (total messages = 44,280). This study measures two types of organizing network relationships on Twitter through the features of mentions and hashtags. Twitter users can mention other users in a message (tweet) by adding the “@” symbol before the other user’s username. Mentions can be used to indicate connections and conversational dynamics between Twitter users (Highfield, Harrington, & Bruns, 2013). The other type of network structure was based on the shared hashtags between users. Hashtags represent Twitter users’ common practice of organizing the information in a tweet with the pound sign or hash mark (#) followed by a keyword (e.g., #TyphoonHaiyan). Use of hashtags represents the intention to follow the discussion related to the marked topic within the Twitter community (Larsson & Moe, 2012). Accordingly, two Twitter users are deemed connected by virtue of their use of common hashtags.

In generating network matrices, the following four steps were performed. First, the raw data were processed using regular expression in Python to generate the contacts mentioned in the tweets and hashtags used by these 70 organizations. Second, the dataset was broken into eight sub-datasets, with each representing a two-week observation of activity (e.g., T1 = October 12, 2013, to October 25, 2013). Third, for each time point, two two-mode binary matrices were generated. The first one was developed to measure the “mentioned” network, where the rows represented the 70 focal organizations and the columns were populated by the 15,634 contacts mentioned by these organizations in the tweets. The other matrix was created to measure the shared hashtag network, where the rows represented the 70 organizations and the columns were the 5,983 hashtags used by these organizations. In both matrices, a “1” was entered in the cell when the focal organization mentioned the particular contact or used a particular hashtag in the tweet during that period, regardless of the number of mentions or the number of hashtag uses.

Fourth, the two binary matrices created in the earlier step were converted into four square one-mode valued matrices (70 × 70 and 15,634 × 15,634; 70 × 70 and 5,983 × 5,983). In total, 48 matrices were created (2 × 8 time points + 4 × 8 time points = 48). For the purpose of this article, only the one-mode org-by-org matrices were used for analysis. These org-by-org matrices represent the common connections between two organizations as they are linked to the same entities or topics. This is similar to the concept of the affiliation network, where two persons are considered connected when they both belong to the same group (Borgatti & Halgin, 2011).

Two categorical attributes were created for the Twitter data, including organizational type and geographic location. All the 70 organizations were coded using the following six categories: business, government agency, NPO (e.g., a group for Filipino students overseas, a volunteer group for Haiyan), NGO (e.g., Red Cross National Societies, CARE), IGO (e.g., United Nations), and coalition (e.g., national response network). Unlike NGOs, NPOs usually engage in a smaller scope of services contributing to the well-being of a community or a society, and they do not engage in the distribution of surplus of funds to their members or stakeholders (Bromideh, 2011). It is important to distinguish NPOs and NGOs in this study because of the emergent and adaptive practices often carried out by the NPOs in response to disasters (Atkinson, 2014). These practices include leveraging different types of communication channels and social relationships, which corresponds to the variable of multimodal communication examined in this study. The region where organizations are headquartered was used to measure similarity in geographic location (Table 1). In addition, two attribute files (geographic and type similarity) were converted into matrix formats using the “Attribute to matrix” option available in UCINET (Borgatti, Everett, & Freeman, 2002).

Data Analysis

Hypothesis testing and research questions were answered using exponential random graph modeling (ERGM) and Multiple-Regression Quadratic Assignment Procedure (MR-QAP). The ERGM method is suitable for this study because it permits the simultaneous examination of attributes, network factors, and other relations on the formation of a network relationship (Shumate & Palazzolo, 2010). PNet is a program used for ERGM analysis through Monte Carlo Markov chain likelihood estimation to maximize the probability of generating the network that fits the observed network by producing convergent estimates (Robins, Snijders, Wang, Handcock, & Pattison, 2007). Each hypothesized parameter is estimated and parameter estimates more than twice their standard errors are considered significantly different from zero (Robins et al., 2007). Positive and significant parameter estimates indicate that the tested structural or attribute effects happen more than by chance alone. Negative and significant parameters indicate that the hypothesized structural or attribute effects occur less than would be expected by chance.

In a well-fitting model, the convergence t ratio for each parameter should be less than 0.1 in absolute value (Robins et al., 2007). After the model converges (with all t ratios less than 0.1), the goodness-of-fit estimates are then obtained by comparing the observed network with the simulated models using the estimated parameter values (Goodreau, 2007). The goodness-of-fit indicates the degree to which the estimated model provides a plausible or good explanation for the observed network as well as for additional network dimensions not included in the model. The convergence statistics for the estimated parameters should be less than 0.1. For non-estimated parameters, if the convergence statistic values are between 1 and 2, it indicates a plausible fit, and a good fit if the statistics are less than 1 (Shumate & Palazzolo, 2010).

BPNet and LPNet are the two extensions of PNet. BPNet, which allows the examination of the other two-mode network simultaneously into the model, was used to analyze the two-mode resource network (Wang, Sharpe, Robins, & Pattison, 2009). The edge (density) is commonly used as a control variable in the two-mode network (Zhu, Huang, & Contractor, 2013). H1 was tested using the estimation of the EdgeAB parameter (Figures 1 and 2). A significant and positive EdgeAB would suggest that the earlier relief network predicts the formation of the subsequent network. H3 tested the effect of potential closure with the L3 parameter. A significant and positive L3 would suggest that active relief organizations are connected to multiple resource contacts and one of those resource contacts is also tapped by other relief organizations. In other words, active relief organizations and popular resource contacts show the tendency of forming a potential cluster. H5a and H5b were tested with star parameters (Sa2, Sp2, Sa3, Sp3). Because models with higher order alternating star parameters did not converge, only the lower order degree distribution effects (two stars, three stars) were included in the model. Significant and positive Sp2 and Sp3 would suggest a certain level of concentration among a few active relief organizations, and significant and positive Sa2 and Sa3 would suggest a certain level of centralization among a few popular resource contact types. H7 examined the effects of similarities in levels of multimodal communication on sharing similar resource contact types at two time points with the Attr_match_2pP parameters. H8a and H8b suggested the effects of relief organizations’ attributes, including the headquartered location and the scope of operations, on the formation of resource ties. The Attr_rP parameters were estimated; significant and positive Attr_rP parameters suggest the tendency of relief organizations to construct resource ties if they operate in the affected country and cover a wider scope of relief areas. H9 examined the similarities in resource contact types by organizations located in the affected country, using the Attr_match_2pP parameter.

Figure 1.

Summary of network graph statistics included in BPNet estimation.

Figure 2.

Summary of network graph statistics included in LPNet estimation.

LPNet was used to analyze the longitudinal evolution of the Twitter networks because it allows the examination of the effects of endogenous factors and attributes on the evolved networks. Edge is often used as a control variable. H4 examined the closure effects with the triangle and alternating two-path parameters. The triangle and alternating two-path parameters represent the degree of closure and the pattern of multiple connectivity, respectively (Snijders, Pattison, Robins, & Handcock, 2006). H6 was tested using the estimation of both the lower order star parameters (two stars, three stars) and the higher order alternating star parameter. The former measure the level of degree distribution (centralization) and the latter captures the spread of the degree distribution (variance between degrees; Robins et al., 2007). MR-QAP was used to test the predictive effects among network relationships, specifically the effects of earlier Twitter networks on the later ones (H2). Similar to multiple-regression analyses, MR-QAP tests the effects and explanatory power of predictors (attributes, relational factors) on a dependent variable (i.e., network relationship). But it uses the random permutations method to account for the inter-dependency of network observations.

Results

The BPNet model of the resource network had a plausible fit, with the absolute values of the convergence statistics for estimated parameters below 0.1, and for additional parameters less than 2 (Table 3).⁴ The network ties with resource contacts immediately after the typhoon predicted the tie structure later on. Hence, H1 was supported. The relief organizations tended to continue their ties with the same resource contact types in a later phase of disaster relief. That also means, if these ties were not established at earlier time, it was less likely that they were able to be made later. The insignificant three-path parameter indicates lack of potential closure, as active relief organizations that were connected with diverse resource contact types did not show a tendency of connecting to the popular resource contact types. H3 was thus not supported. Several central and active relief organizations stood out with multiple links to resource contacts, as indicated by the positive and significant relief-org two-star parameter. H5a was supported. But it is worth noticing that having ties with too many resource contact types was not observed, which was shown in the negative relief-org three-star parameter. No particular resource contact type was identified as more popular than others, which were shown in the insignificant resource contact two-star and three-star parameters. H5b was not supported.

Table 3.

Results of BPNet Estimation.

	Hypothesis	Estimate	t ratio
Density (edge; L)	Control	−4.932 (1.497)*	−0.038
Relief organization’s resource network (first period; EdgeAB)^a	H1	0.762 (0.202)*	0.038
Three path (L3)	H3	−0.006 (0.006)	−0.041
Relief org two-star (Sp2)	H5a	1.186 (0.268)*	−0.032
Resource contact two-star (Sa2)	H5b	0.175 (0.189)	−0.022
Relief org three-star (Sp3)	H5a	−0.123 (0.045)*	−0.055
Resource contact three-star (Sa3)	H5b	−0.002 (0.010)	−0.000
Relief organizations’ multimodal communication two-star (first period; Attr_match_2pP)	H7	−0.181 (0.080)*	−0.037
Relief organizations’ multimodal communication two-star (second period; Attr_match_2pP)^b	H7	0.069 (0.023)*	0.001
Relief organizations’ HQ location density (Attr_rP)	H8a	−0.091 (0.136)	−0.055
Relief organizations’ scope of operation density (Attr_rP)	H8b	−0.034 (0.128)	0.060
Relief organizations’ HQ location two-star (Attr_match_2pP)	H9	0.011 (0.023)	−0.095

Note. Parameter estimates (SE). The hypotheses in bold were supported.

First period: Immediately after the typhoon.

Second period: Within three months following the typhoon.

Significant effect (p < .05).

A significant homophily effect was observed in the variable of external multimodal communication. Similarities in the level of multimodal communication with external actors at different times had an effect on organizations’ connections. Hence, H7 was supported. Interestingly, organizations with similar levels of external multimodal communication immediately after the typhoon were less likely to have connections to the same resource contact types. Yet, this similarity had a positive effect on the connections that took place weeks or months after the typhoon. Whether relief organizations were headquartered in the Philippines and the scope of relief actions covering multiple affected regions did not make a difference in terms of having external resource ties. Hence, H8a and H8b were not supported. Being headquartered in the affected country (the Philippines) did not reflect the preference of building ties to common resource contact types; thus, H9 was not supported.

Twitter Networks

The results of MR-QAP analyses suggested that in the Twitter mention network, four of seven networks at earlier points predicted the network at the last point of observation (T8; Table 4). By comparison, all the seven earlier networks of organizations built on shared hashtags significantly predicted the network at T8. In both mention and shared hashtag networks, the network structure at T8 was strongly influenced by the immediate period preceding that time point (T7). Together, H2 was supported, indicating that organizations were embedded in their prior relationships and participation in these online humanitarian networks could garner expeditious benefits in terms of being integrated into the networks.

Table 4.

Results of MR-QAP on Twitter Organizing Networks.

	Hypothesis	Model 1	Model 2
	Hypothesis	Post-Haiyan network (mention contacts)	Post-Haiyan network (shared hashtag contacts)
Intercept		.000 (−.001)***	.000 (.004)***
Organizational type similarity		−.015 (−.031)	.010 (.048)
Geographic proximity		−.008 (−.016)	−.019 (−.096)
T1	H2	.184 (.237)***	.176 (.237)***
T2	H2	−.025 (−.030)	.073 (.089)**
T3	H2	.052 (.024)*	−.048 (−.040)*
T4	H2	−.023 (−.020)	.119 (.097)**
T5	H2	.196 (.192)***	.062 (.068)*
T6	H2	.001 (.002)	.205 (314)***
T7	H2	.488 (.383)***	.412 (.437)***
Adjusted R²		.580	.778
Significance level		.000	.000
No. of observations		4,830	4,830

Note. Coefficients in parentheses are unstandardized regression coefficients. The hypothesis in bold was supported. Post-Haiyan network (T8) refers to the network ties measured from January 18 to 31, 2014. Each of the time point represents the ties captured in the following: T1: October 12, 2013, to October 25, 2013; T2: October 26, 2013, to November 8, 2013; T3: November 9, 2013, to November 22, 2013; T4: November 23, 2013, to December 6, 2013; T5: December 7, 2013, to December 20, 2013; T6: December 21, 2013, to January 3, 2014; T7: January 4, 2014, to January 17, 2014. Relief operations of Typhoon Haiyan started at T3. MR-QAP = Multiple-Regression Quadratic Assignment Procedure.

p < .05. **p < .01. ***p < .001.

The LPNet models for both mention and shared hashtag networks had a good fit, with the absolute values of the convergence statistics for estimated parameters below 0.1, and for additional parameters less than 1 (Table 5). In both mention and shared hashtag networks, the positive effects of the triangle parameter suggest that organizations tended to interact in clique forms. The negative and significant alternating two-path parameter indicates that the situation where organizations with shared partners but without direct connection was rare. Over time, organizations were less likely to form ties that only resulted in multiple short indirect paths between them. Combining positive triangle and negative alternating two-path parameters implies that organizations with multiple partners were likely to form ties with each other and exhibited a preference for forming direct ties in the Twitter networks. H4 was thus supported.

Table 5.

Results of LPNet Estimation.

	Hypothesis	Mention network		Shared hashtag network
	Hypothesis	Estimate	t ratio	Estimate	t ratio
Edge	Control	−3.219 (0.879)*	−0.074	−2.155 (1.008)*	−0.039
Triangle	H4	0.271 (0.107)*	−0.013	0.095 (0.032)*	0.039
Alternating two-path	H4	−0.076 (0.036)*	−0.013	−0.129 (0.025)*	0.023
Two-star	H6	0.036 (0.101)	0.041	0.016 (0.054)	0.008
Three-star	H6	0.0002 (0.007)	−0.010	0.001 (0.002)	−0.027
Alternating star	H6	0.016 (0.282)	0.069	−0.156 (0.509)	0.008
Organizational type homophily	RQ1	0.041 (0.199)	0.003	0.431 (0.195)*	−0.003
Geographic homophily	RQ1	−0.295 (0.172)	0.016	−0.410 (0.214)	0.004

Note. Parameter estimates (SE). The analysis compared the network structures at T1 and T8 and the hypothesis in bold was supported.

Significant effect (p < .05).

In both the mention and shared hashtag networks, insignificant two-star and three-star effects suggested a lack of highly connected relief organizations. Furthermore, there was no evidence indicating an equal distribution of degrees among relatively unconnected relief organizations, as shown by the insignificant effects of the alternating star parameter in both networks. Hence, H6 was not supported.

To answer RQ1 regarding the homophily effect in the online humanitarian network, the effect of type similarity was positively significant in the shared hashtag network, suggesting that the social selection process was in place because organizations of similar type were likely to form ties with one another. Yet in the mention network, similarity in organizational type and geographic location was not salient in the formation of network ties. In sum, the results of the longitudinal network analysis showed the consistent strength of the structural processes in the evolution of the Twitter networks, controlling for the effect of homophily factors.

In answering RQ2 regarding the network outcomes of collaborative relief networks, visualization of the network structures is used to identify different forms of resilience as manifested in the resource network and the online network. Due to space limitations, the visualization of the resource network on the ground is available upon request. First, instead of a few resource contact types, multiple resource contact types (including local and provincial government agencies, NPOs/NGOs, news media, businesses, and individuals from outside the Philippines) served as suppliers in the resource network on the ground after Haiyan. Second, a clear pattern of cross-sector and cross-geographic collaboration was discerned as relief organizations at national and international levels were involved in relief actions in different phases of disaster relief. For example, local NPOs and NGOs worked with businesses from outside the Philippines, local and provincial government agencies, and other international NPOs/NGOs.

As in the disaster response situation, a higher level of cohesion and coordination within the disaster response network was often observed in successful cases of disaster response (Kapucu & Garayev, 2013). In fact, comparing the network within three months after Haiyan with the network immediately after, relief organizations exhibited a decline of average resource ties from 4.46 to 4.02 and resource contact types received a decrease of nominating organizations from 18.3 to 16.5. Despite these decreases of activity in the resource network, it was observed that the network consisting of multiple organizational actors within the affected country working in concert with resource contacts from outside has established a resilient structure over time.

Figures 3a to 3d provide the visualization of popular hashtags and corresponding shared hashtag networks before and during Typhoon Haiyan. Figures 3e and 3f illustrate the popular hashtags and the network at the last point of observation. The typhoon appeared to serve as a force to assemble the network focusing on this particular topic. The network was relatively dense and active immediately after Haiyan. Three months after Haiyan, the network resumed its looser network structure constructed on a group of other common topics, which resembled the pre-Haiyan network. Together, this indicates that the humanitarian organizations within the Twitter network exhibited resilient capacity to respond and adapt to environmental changes in the form of self-sustaining network over time.

Figure 3.

Popular hashtags used at three specific time points and the corresponding networks.

Discussion

Following the bona fide network perspective, this study treats a natural disaster (Typhoon Haiyan) as an antecedent that facilitated organizational collaborative networks involved in relief operations, and examines the network processes and outcomes of these networks across online and offline domains. The results from the survey data supported the hypotheses about structural and positional embeddedness. The resource network immediately after the typhoon predicted the network structure at a later time point. The resource network was also concentrated in a few highly connected organizations, yet with a cap in terms of the number of possible resource contact types. Moreover, having socio-technical similarity (i.e., those having similar levels of multimodal communication with resource contacts) immediately and months after the typhoon helped the formation of common resource ties. Similar to the survey data, the analysis of the organizations’ Twitter data supported the hypothesis about structural embeddedness. As early as one month before the typhoon and as well as two weeks before the final period of observation, the network ties at those points significantly predicted the network ties at the last point of observation. As the inter-organizational networks evolved, they tended to interact with one another in the form of a closure structure. Yet in the online networks, these organizations exhibited a relatively equal distribution of connections, with a lack of popular central organizations or preferred connections with those central organizations.

Online and Offline Bona Fide Collaborative Humanitarian Networks

This study examines disaster response networks on the ground and online, which offers an important way to conceptualize bona fide networks by considering organizing and collaborative behaviors in different forms across online and offline domains (Lai, 2014). Despite the observed significance of structural embeddedness, the resource network on the ground and online humanitarian networks exhibited different network processes and outcomes. In the resource network on the ground, while positional embeddedness in the form of a core-periphery network structure mattered, structural embeddedness in the form of closure and attribute- and geography-based homophily exerted insignificant effects on the formation of collaborative networks following the disaster. In contrast, the online humanitarian networks presented a different picture because with the reduced influence of central players and elimination of bridges, organizations were likely to engage in direct connections with one another and form a closure structure.

Together, these findings enrich the literature on organizational collaborative networks, especially in the emergency response context. Specifically, the domination of a few organizations in the relief resource network is prevalent (Abbasi & Kapucu, 2012). Yet, instead of a few popular resource contact types, several entities such as government agencies, NPOs/NGOs, businesses, and individuals from outside the affected area stood out as solid resource contacts in different phases of disaster relief, which indicates the resilience of the network in response to rapid changes such as a natural disaster. Moreover, unlike previous studies where local NGOs were usually relatively peripheral in the relief network (Moore et al., 2003), this study found evidence of the active role played by local NGOs and NPOs in the cross-sector and cross-geographic collaboration in the resource network after Haiyan. These findings echo the previous research, which suggests that self-organizing networks of organizations consisting of the public, non-profit, and private sectors is deemed an optimal approach to deal with various complex challenges imposed on communities around the world, especially for disaster management (Comfort, Oh, & Ertan, 2009; Luna, 2001).

On the other hand, the lack of support for hypotheses regarding the influences of geographic proximity to the affected area and scope of geographic coverage on the formation of resource ties indicates the possibility that relief operations may require little engagement in the external resource network. For example, the well-established international humanitarian organizations covering wider affected areas might be seen as “isolates” in the external resource network because they may mostly tap into their own internal network of country offices for resources and support. As a matter of fact, half of the isolates in the resource network at different time points were international NGOs and NPOs. Another interpretation is that regardless of their headquartered or covering location, some organizations focused on a particular type of emergency response (e.g., child safety, health issues, online crowdsourcing), and were thus less involved in the external resource network.

In the online humanitarian networks, the network mechanisms of embeddedness are constructed based on communication relationships. Organizations’ use of the features of mentions and hashtags on Twitter represent communicative behaviors either through publicizing other contacts’ names or engaging in topic-focused conversations, which requires actors to know and practice the common norms of using these keywords. Inter-organizational collaboration entails positive links being initiated and maintained. As such, the findings about the structural embeddedness of the online humanitarian networks provide important empirical evidence for the research on digital organizing, which suggests that the positional embeddedness (skewed degree distribution) tends to relate to negatively connoted links (e.g., aggressive actions) whereas structural embeddedness (clustering) relates to positive links (e.g., friendship, communication; Szell, Lambiotte, & Thurner, 2010).

Emergent and sustained communicative practices

The framework of media multiplexity implies that media use is embedded into and interacts with a social collective’s existing communicative practices. The context of disaster necessitates the enactment of a set of communicative practices that might be especially suitable for post-disaster action. During the process of adapting to this situational demand, organizations build norms and learn about ways to enrich the relationship by exchanging a wide scope of resources, and also to reach others by using different modes of communication. The results of negative and positive effects of similarity in multimodal communication (immediately and months after the typhoon, respectively) are evidence of this learning process. Organizations adapted to the chaotic and uncertain environment immediately after the typhoon, and they engaged in modes at varying levels to reach out or be reached by resource contacts. This variation became part of the learning process, which helped relief organizations find common connections to certain resource contacts later on. When the triage was over, relief organizations with similar levels of multimodal external communication tended to select common resource contacts. Multimodal communication thus represents an enactment of adaptability by organizations in carrying out relief support work. This study extends the bona fide network perspective to a disaster relief context, and offers empirical evidence highlighting the importance of considering emergent and robust communicative practices adopted by organizations in evolving inter-organizational collaboration following a disaster.

Compared with the significant effects of earlier relationships and socio-technical similarity, attribute- and geography-based factors are not salient in post-disaster relief networks, as indicated by the insignificant effects of geographic- and organizational-type similarities in the evolved networks on Twitter, and by the insignificant effect of geographic proximity for operations in the resource network on the ground. Although this finding fails to support the existing research linking similarities of organizational attributes and geography to network formation, this study evidences the argument that social media has the capacity to foster collaboration and mobilization beyond geographic limitations and expand the scope of participation (Penney & Dadas, 2014). Specifically, it appears that what matters more for post-disaster organizational collaborative networks online is whether and how the relationships are built and evolve over time, rather than attribute- or geography-based enablers or constraints such as organizational type or geographic location.

The bona fide network perspective emphasizes the fuzzy boundaries of collaboration as a network is formed revolving around a specific issue (Cooper & Shumate, 2012). The survey data show the evolving and consistent cross-sector and cross-geographic resource network constructed by relief organizations revolving around Haiyan. The online data suggest that interactions and relationships on Twitter helped relief organizations explore different types of foci that sustained the loosely connected network relationships. In other words, the bona fide organizing networks on Twitter are enacted in a temporally dynamic way because the common social issues that connect those humanitarian organizations together are subject to change over time. Combining these two sources of data, it is clear that although this online network may not necessarily translate into physical humanitarian aid operations, it offers the practical insights of creating and maintaining a socio-technical system where humanitarian organizations operating on the ground can restructure their focused activity and articulate connections, which potentially can be activated when needed.

This study examines resilience as a type of network outcome of organizational collaboration, especially in the disaster relief context. Despite having no direct measures, resilience was observed through the consistent presence of particular resource contact types and the diversity of partnering organizations in the resource network, as well as the existence of topic-centered self-sustaining activities in the online network. Future research should investigate different types of resilience as network outcomes of humanitarian collaborative efforts. In countries with resource limitations such as the Philippines, it has been a growing trend that humanitarian aid organizations engage in broader community development initiatives, beyond disaster response (Luna, 2001). It is possible that organizations from outside the affected country are able to tap into the local network after a particular disaster, which helps them engage in other initiatives related to the community in the long run.

This study makes a major methodological contribution to the existing research on organizational and communication technology by using a mixed-methods approach and encompassing cross-sectional and longitudinal data in examining bona fide humanitarian organizing networks. Using social media data offers a more economical way of obtaining data on different types of organizations (IGOs, INGOs, national NGOs, and NPOs), which addresses the limitation of current research focusing on only one category of organizations per study (e.g., Atouba & Shumate, 2010; Shumate et al., 2005). It also enables longitudinal observations of the interactions among organizations of different types, which helps yield more accurate insights into the process of organizing networks.

Conclusion

This study has three limitations. First, the online network data relied on 70 organizations’ Twitter data. Undoubtedly, detailed analysis on the features and content available on Twitter and other social media systems is necessary to enrich social understandings about the role of social media in humanitarian organizing. Second, this study uses multiple sources to extract the list of humanitarian organizations involved in Haiyan relief, espeically for the collection of Twitter data. Yet, convenience and snowball sampling is a common approach in disaster-related research (Norris, 2006). Last, the online survey had a relatively low response rate and resulting small sample size. The mixed-methods approach incorporating Twitter data, however, helped mitigate the potential bias associated with only getting insight from the organizations responding to the survey. Specifically, the Twitter data were retrieved from the organizations that responded to the survey as well as those identified by the survey respondents as their network contacts. Accordingly, the Twitter data of those 15 organizations that did not respond to the survey invitation were included and analyzed.

Using a mixed-methods and longitudinal approach, this study answers two questions, the first of which is how collaborative response networks that take different forms across online and offline domains are sustained over time. Findings of this study revealed that the resource network on the ground corroborates the hypothesis about structural embeddedness and positional embeddedness, whereas the online networks exemplify the patterns of structural embeddedness in the form of triadic closure. By conceptualizing technology use as organizations’ adaptive relational behavior, this study showed that organizations with similar levels of multimodal communication were likely to engage with similar resource ties. Nonetheless, similarities in organizational type and geographic locations were not salient in both the resource network on the ground and the Twitter networks, which further reinforce the importance of investing in emergent relationships that evolve over time. This study also answers the second question about the network outcomes of these collaborative networks in terms of resilience, which is manifested through the consistent presence of multiple key players and cross-sectional and cross-geographic collaboration among organizations on the ground and the topic-centered self-sustaining network online. With more empirical research expanding the scope of inquiry, more knowledge will be obtained as to how to better implement effective adaptive methods to cope with environmental threats and changes.

Footnotes

Acknowledgements

The authors are grateful to the two anonymous reviewers and the editor Dr. Michael Roloff for their incredibly useful comments and feedback on the earlier drafts of this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project was partially supported through the first author’s Start-Up Grant at Nanyang Technological University.

Notes

Author Biographies

Chih-Hui Lai (PhD, Rutgers University, 2012) is an assistant professor in the Wee Kim Wee School of Communication and Information at Nanyang Technological University, Singapore. Her research focuses on how individuals, groups, and organizations use information and communication technologies (ICTs) to communicate and how relationships evolve or emerge through the process.

Bing She (PhD, Wuhan University, 2015) is a research assistant of the China Data Center at the University of Michigan. His research interests include spatial data analysis, CyberGIS, and geographical information retrieval.

Xinyue Ye (PhD, University of California–Santa Barbara, 2010) is an assistant professor in the Department of Geography at Kent State University with expertise in space-time analytics and big social data.

References

Abbasi

Kapucu

(2012). Structural dynamics of organizations during the evolution of interorganizational networks in disaster response. Journal of Homeland Security and Emergency Management, 9(1). doi:10.1515/1547-7355.1975

Adger

W. N.

Quinn

Lorenzoni

Murphy

Sweeney

(2012). Changing social contracts in climate-change adaptation. Nature Climate Change, 3, 330-333. doi:10.1038/nclimate1751

Atkinson

C. L.

(2014). The 2011 flood in Minot (North Dakota, USA) and the role of faith-based and nonprofit groups in hazard event response and recovery. International Journal of Disaster Risk Reduction, 8, 166-176. doi:10.1016/j.ijdrr.2014.03.003

Atouba

Shumate

(2010). Interorganizational networking patterns among development organizations. Journal of Communication, 60, 293-317. doi:10.1111/j.1460-2466.2010.01483.x

Atouba

Y. C.

Shumate

(2014). International nonprofit collaboration: Examining the role of homophily. Nonprofit and Voluntary Sector Quarterly, 44, 587-608. doi:10.1177/0899764014524991

Borgatti

S. P.

Everett

M. G.

Freeman

L. C.

(2002). UCINET for windows: Software for social network analysis. Harvard, MA: Analytic Technologies. Retrieved from https://sites.google.com/site/ucinetsoftware/home

Borgatti

S. P.

Halgin

D. S.

(2011). Analyzing affiliation networks. In Scott

Carrington

P. J.

(Eds.), The SAGE handbook of social network analysis (pp. 417-433). London, England: SAGE. doi:10.4135/9781446294413.n28

Bromideh

A. A.

(2011). The widespread challenges of NGOs in developing countries: Case studies from Iran. International NGO Journal, 6, 197-202. doi:10.5897/NGOJ11.014

Chewning

L. V.

Lai

C.-H.

Doerfel

M. L.

(2013). Organizational resilience and using information and communication technology to rebuild communication structures. Management Communication Quarterly, 27, 237-263. doi:10.1177/0893318912465815

10.

Comfort

L. K.

Ertan

(2009). The dynamics of disaster recovery: Resilience and entropy in hurricane response systems 2005-2008. Public Organization Review, 9, 309-323. doi:10.1007/s11115-009-0098-3

11.

Cooper

K. R.

Shumate

(2012). Interorganizational collaboration explored through the bona fide network perspective. Management Communication Quarterly, 26, 623-654. doi:10.1177/0893318912462014

12.

Doerfel

M. L.

Lai

C.-H.

Chewning

L. V.

(2010). The evolutionary role of interorganizational communication: Modeling social capital in disaster contexts. Human Communication Research, 36, 125-162. doi:10.1111/j.1468-2958.2010.01371.x

13.

Doerfel

M. L.

Taylor

(2004). Network dynamics of interorganizational cooperation: The Croatian civil society movement. Communication Monographs, 71, 373-394. doi:10.1080/0363452042000307470

14.

Freeman

L. C.

(1979). Centrality in social networks: Conceptual clarification. Social Networks, 1, 215-239. doi:10.1016/0378-8733(78)90021-7

15.

Gabbay

S. M.

Talmud

Raz

(2001). Corporate social capital and strategic isomorphism: The case of the Israeli software industry. Research in the Sociology of Organizations, 18, 135-150. doi:10.1016/S0733-558X(01)18006-7

16.

Gillan

(2009). The UK-antiwar movement online. Information, Communication & Society, 12, 25-43. doi:10.1080/13691180802158532

17.

Goodreau

S. M.

(2007). Advances in exponential random graph (p*) models to a large social network. Social Networks, 29, 231-248. doi:10.1016/j.socnet.2006.08.001

18.

Granovetter

(1992). Problems of explanation in economic sociology. In Nohria

Eccles

(Eds.), Networks and organizations: Structure, form, and action (pp. 25-56). Boston, MA: Harvard Business School Press.

19.

Gulati

(1995). Social structure and alliance formation patterns: A longitudinal analysis. Administrative Science Quarterly, 40, 619-652. doi:10.2307/2393756

20.

Gulati

Gargiulo

(1999). Where do interorganizational networks come from? American Journal of Sociology, 104, 1439-1493. doi:10.1086/210179

21.

Guo

Acar

(2005). Understanding collaboration among nonprofit organizations: Combining resource dependency, institutional, and network perspectives. Nonprofit and Voluntary Sector Quarterly, 34, 340-361. doi:10.1177/0899764005275411

22.

Haythornthwaite

(2002). Strong, weak and latent ties and the impact of new media. The Information Society, 18, 385-401. doi:10.1080/01972240290108195

23.

Haythornthwaite

Wellman

(1998). Work, friendship, and media use for information exchange in a networked organization. Journal of the American Society for Information Science, 49, 1101-1114.

24.

Highfield

Harrington

Bruns

(2013). Twitter as a technology for audiencing and fandom: The# Eurovision phenomenon. Information, Communication & Society, 16, 315-339. doi:10.1080/1369118X.2012.756053

25.

Kapucu

(2005). Interorganizational coordination in dynamic context: Networks in emergency response management. Connections, 26, 33-48. doi:10.1.1.126.5923

26.

Kapucu

Garayev

(2013). Designing, managing, and sustaining functionally collaborative emergency management networks. American Review of Public Administration, 43, 312-330. doi:10.1177/0275074012444719

27.

Kim

Barnett

G. A.

(2000). The structure of the international telecommunication regime in transition: A network analysis of international organizations. International Interactions, 26, 91-127. doi:10.1080/03050620008434962

28.

Lai

C.-H.

(2014). Can our groups survive? An investigation of the evolution of mixed-mode groups. Journal of Computer-Mediated Communication, 19, 839-854. doi:10.1111/jcc4.12075

29.

Lai

C.-H.

Chewning

L. V.

(2014, May). Organizing and adaptations of collective grassroots efforts for disaster response. Paper presented at the 64th annual conference of International Communication Association, Seattle, WA.

30.

Landwehr

P. M.

Carley

K. M.

(2014). Social media in disaster relief: Usage patterns, data mining tools, and current research directions. In Chu

W. W.

(Ed.), Data mining and knowledge discovery for big data (pp. 225-257). Berlin, Germany: Springer. doi:10.1007/978-3-642-40837-3_7

31.

Larsson

A. O.

Moe

(2012). Studying political microblogging: Twitter users in the 2010 Swedish election campaign. New Media & Society, 14, 729-747. doi:10.1177/1461444811422894

32.

Lee

Monge

(2011). The coevolution of multiplex communication networks in organizational communities. Journal of Communication, 61, 758-779. doi:10.1111/j.1460-2466.2011.01566.x

33.

Lin

(1999). Building a network theory of social capital. Connections, 22, 28-51. Retrieved from http://www.insna.org/PDF/Keynote/1999.pdf

34.

Luna

E. M.

(2001). Disaster mitigation and preparedness: The case of NGOs in the Philippines. Disasters, 25, 216-226. doi:10.1111/1467-7717.00173

35.

Majchrzak

Jarvenpaa

S. L.

Hollingshead

A. B.

(2007). Coordinating expertise among emergent groups responding to disasters. Organization Science, 18, 147-161. doi:10.1287/orsc.1060.0228

36.

McCarthy

J. D.

Zald

M. N.

(1977). Resource mobilization and social movements: A partial theory. American Journal of Sociology, 82, 1212-1241. doi:10.1086/226464

37.

McPherson

Smith-Lovin

Cook

J. M.

(2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27, 415-444. doi:10.1146/annurev.soc.27.1.415

38.

Moore

Eng

Daniel

(2003). International NGOs and the role of network centrality in humanitarian aid operations: A case study of coordination during the 2000 Mozambique floods. Disasters, 27, 305-318. doi:10.1111/j.0361-3666.2003.00235.x

39.

Muralidharan

Rasmussen

Patterson

Shin

J. H.

(2011). Hope for Haiti: An analysis of Facebook and Twitter usage during the earthquake relief efforts. Public Relations Review, 37, 175-177. doi:10.1016/j.pubrev.2011.01.010

40.

Nahapiet

Ghoshal

(1998). Social capital, intellectual capital, and the organizational advantage. Academy of Management Review, 23, 242-266. doi:10.5465/AMR.1998.533225

41.

National Disaster Risk Reduction and Management Council. (2014, March 6). NDRRMC situation report on the effects of Typhoon Yolanda. Retrieved from http://www.ndrrmc.gov.ph/attachments/article/1329/Effects_of_Typhoon_YOLANDA_%28HAIYAN%29_SitRep_No_106_06MAR2014.pdf

42.

Nolte

I. M.

Boenigk

(2013). A study of ad hoc network performance in disaster response. Nonprofit and Voluntary Sector Quarterly, 42, 148-173. doi:10.1177/0899764011434557

43.

Norris

F. H.

(2006). Disaster research methods: Past progress and future directions. Journal of Traumatic Stress, 19, 173-184. doi:10.1002/jts.20109

44.

Penney

Dadas

(2014). (Re)Tweeting in the service of protest: Digital composition and circulation in the Occupy Wall Street movement. New Media & Society, 16, 74-90. doi:10.1177/1461444813479593

45.

Putnam

L. L.

Stohl

(1990). Bona fide groups: A reconceptualization of groups in context. Communication Studies, 41, 248-265. doi:10.1080/10510979009368307

46.

Robins

Snijders

Wang

Handcock

Pattison

(2007). Recent developments in exponential random graph (p*) models for social networks. Social Networks, 29, 192-215. doi:10.1016/j.socnet.2006.08.003

47.

Robinson

S. E.

Berrett

Stone

(2006). The development of collaboration of response to Typhoon Katrina in the Dallas area. Public Works Management & Policy, 10, 315-327. doi:10.1177/1087724X06289053

48.

Robinson

S. E.

Eller

W. S.

Gall

Gerber

B. J.

(2013). The core and periphery of emergency management networks. Public Management Review, 15, 344-362. doi:10.1080/14719037.2013.769849

49.

Shepherd

D. A.

Williams

T. A.

(2014). Local venturing as compassion organizing in the aftermath of a natural disaster: The role of localness and community in reducing suffering. Journal of Management Studies, 51, 952-994. doi:10.1111/joms.12084

50.

Shumate

(2012). The evolution of the HIV/AIDS NGO hyperlink network. Journal of Computer-Mediated Communication, 17, 120-134. doi:10.1111/j.1083-6101.2011.01569.x

51.

Shumate

Dewitt

(2008). The north/south divide in NGO hyperlink networks. Journal of Computer-Mediated Communication, 13, 405-428. doi:10.1111/j.1083-6101.2008.00402.x

52.

Shumate

Fulk

Monge

(2005). Predictors of the international HIV/AIDS INGO network over time. Human Communication Research, 31, 482-510. doi:10.1111/j.1468-2958.2005.tb00880.x

53.

Shumate

Palazzolo

(2010). Exponential random graph (p*) models as a method for social network analysis in communication research. Communication Methods and Measures, 4, 341-371. doi:10.1080/19312458.2010.527869

54.

Simo

Bies

A. L.

(2007). The role of nonprofits in disaster response: An expanded model of cross-sector collaboration. Public Administration Review, 67(s1), 125-142. doi:10.1111/j.1540-6210.2007.00821.x

55.

Singer

H. H.

Kegler

M. C.

(2004). Assessing interorganizational networks as a dimension of community capacity: Illustrations from a community intervention to prevent lead poisoning. Health Education & Behavior, 31, 808-821. doi:10.1177/1090198104264220

56.

Snijders

T. A. B.

Pattison

P. E.

Robins

G. L.

Handcock

(2006). New specifications for exponential random graph models. Sociological Methodology, 36, 99-153. doi:10.1111/j.1467-9531.2006.00176.x

57.

Starbird

Palen

(2012). (How) will the revolution be retweeted? Information diffusion and the 2011 Egyptian uprising. In Proceedings of the ACM 2012 conference on computer supported cooperative work (pp. 7-16). ACM. doi:10.1145/2145204.2145212

58.

Szell

Lambiotte

Thurner

(2010). Multirelational organization of large-scale social networks in an online world. Proceedings of the National Academy of Sciences, 107, 13636-13641. doi:10.1073/pnas.1004008107

59.

Taylor

Doerfel

M. L.

(2011). Evolving network roles in international aid efforts: Evidence from Croatia’s post war transition. Voluntas: International Journal of Voluntary and Nonprofit Organizations, 22, 311-334. doi:10.1007/s11266-010-9155-3

60.

Treem

J. W.

Leonardi

P. M.

(2012). Social media use in organizations: Exploring the affordances of visibility, editability, persistence, and association. Communication Yearbook, 36, 143-189. doi:10.2139/ssrn.2129853

61.

Wang

Sharpe

Robins

G. L.

Pattison

P. E.

(2009). Exponential random graph (p*) models for affiliation networks. Social Networks, 31, 12-25. doi:10.1016/j.socnet.2008.08.002

62.

Wells

Lee

S.-Y. D.

Alexander

J. A.

(2001). Institutionalized ties and corporate social capital: The case of hospital mergers and closures. Research in the Sociology of Organizations, 18, 59-82. doi:10.1016/S0733-558X(01)18003-1

63.

Yang

(2013). When transnational civil network meets local context: An exploratory hyperlink network analysis of Northern/Southern NGOs’ virtual network in China. Journal of International and Intercultural Communication, 6, 40-60. doi:10.1080/17513057.2012.719632

64.

Zhu

Huang

Contractor

N. S.

(2013). Motivations for self-assembling into project teams. Social Networks, 35, 251-264. doi:10.1016/j.socnet.2013.03.001