Abstract

Collaborative forums involving multiple stakeholders responding to natural hazards are prevalent, yet there is little conclusive evidence of how stakeholders exchange information across such forums and how different patterns of information exchange influence forum goals. This study analyzes information exchange among representatives of 51 organizations across 50 collaborative forums in response to weather warnings in Sweden, 2011–15. Using coded transcripts from forum meetings, the study estimates exponential random graph models to document the prevalence of network configurations of organizations across these forums. The results show that actors avoid engaging in information exchanges within closed subgroups and that no specific type or organization was particularly active in exchanging information. The study suggests that the forum structures are consistent with short-term operational goals as well as the long-term objective of these forums to sustain collaboration over time. The study discusses potential explanations for these patterns and implications for forum performance in relation to natural hazards management.

1. Introduction

Forums that seek to facilitate collaboration among diverse sets of stakeholders have become increasingly prevalent in response to complex environmental problems. Among these problems are natural hazards that require swift collaboration to mitigate societal impacts. Collaboration between multiple public and private actors is hence an essential feature of sustainable communities and a way to build trust and enable collective decision-making and information flows across organizations (Turner et al. 2003; Tobin 1999). Studies in crisis management suggest, however, that these arrangements may underperform due to narrow participation, tribal identities, and communication problems, which can lead to misallocation of resources and biases in collective sense-making (Boin and ’t Hart 2010). These potential pitfalls turn attention to how stakeholders actually exchange information across collaborative forums and what relationships emerge from these exchanges.

A “collaborative forum” here refers to an institutionalized arena involving representatives of government agencies, regional authorities, and business organizations that exchange information and achieve coordination in response to natural hazards as well as to sustain long-term collaboration. These forums aims at several goals at different scales: establishing a shared national-level “joint operating picture” based on different actors’ information and perceptions, enhancing situation awareness for each organization, and sustaining collaboration among forum participants over time. Efforts to achieve such goals are generally constrained by uncertainty whether participation will generate useful information and if other actors will actually contribute to joint efforts. Under these circumstances, actors typically resort to different strategies for reducing uncertainty: by forming connections with a few close and trusted colleagues to promote trust (bonding network social capital, or simply “bonding”) or by connecting with a broader set of actors to ensure access to information from many different sources (bridging network social capital, or simply “bridging”) (Andrew and Carr 2013; Berardo and Scholz 2010). Since these strategies can facilitate or hinder the solution of specific collective-action problems, it is important to disentangle what structures characterize relationships between stakeholders across forums (Berardo 2014).

This study is situated in relation to policy network research on social capital where bonding and bridging are depicted as different strategies for coping with uncertainty and risk in multistakeholder arrangements (Berardo and Scholz 2010). Bonding entails relatively closed network structures that support cooperation through dense relationships by which it is relatively easy to detect and punish defective behavior by noncooperators. These relationships improve the quality of information available to the actors and also increase the predictability of how other actors will behave in certain situations. Thus, bonding is more likely in situations where there is a high risk of defection (Berardo and Scholz 2010). Bonding furthermore increases the probability of sustaining cooperation over time.

In contrast, bridging involves ties that extend close acquaintances and connect actors with other actors with whom they have little or no prior contact. While such bridging structures do not foster close and dense relationships, they open opportunities to disseminate non-overlapping information (i.e., unique information provided by different actors) that can be useful to develop innovative solutions to complex problems (Berardo 2014). These relationships are likely to emerge in low-risk situations where assurance against defection is not a priority but where actors face different challenges, such as coordination dilemmas where they share similar goals but disagree on the means (Berardo and Scholz 2010). It should be pointed out that bonding and bridging relationships are not mutually exclusive and can appear simultaneously in any given network depending on whether multiple problems coexist and introduce actors to varying degrees of risk (Berardo and Lubell 2016).

Some studies refer to a third type of relationship based on “linking” social capital, acknowledging that actors may seek relationships with actors and organizations that are more powerful than themselves as a means to leverage new ideas, resources, and information (Woolcock 2001). However, linking relationships are not explicitly included in this analysis since it would require measures of the power of actors and forums.

While most previous work on social capital in natural hazards has focused on residents (see, e.g., Aldrich and Meyer 2014), recent studies of interorganizational networks provide evidence of both bridging and bonding strategies. For instance, in an analysis of the 2011 Thailand floods Andrew et al. (2016) show that organizations formed bridging relationships that enhanced organizational-level resilience. In contrast, studies of response systems during Hurricane Katrina in 2005 and earthquakes in Haiti (2010), Indonesia (2009), and Japan (2011) demonstrate the prevalence of bonding relationships based on close working relationships among organizations (Siciliano and Wukich 2015, 2017). These studies, however, analyze interactions during single major events and do not investigate collaboration in repeated interactions through time in multiple situations. This study investigates collaborative forums as two-mode networks defined by actors (mode 1) engaging in information exchanges in forums (mode 2). Actors are thus not directly linked to other actors in a social network but to the forums in which they exchange information.

Expanding previous work on single natural hazard response networks, this study explores how collaborative efforts unfold across multiple collaborative forums through time (Nohrstedt and Bodin 2014; Nohrstedt 2016). Governance systems, or networks of forums, have become an increasingly common study object in policy network research (e.g., Berardo 2014; Berardo and Scholz 2010) but have not yet been properly examined in natural hazards management. By examining networks of forums it is possible to draw lessons about how actors connect across forums to enhance information sharing and coordination (Fischer et al. 2018). Ultimately, this can also advance our understanding of the ability of collaboration forums to reach their goals and address the problems they are set to solve (Berardo 2014).

The aim of the current study is to examine what relationships emerge among actors who participate in multiple collaborative forums responding to natural hazards. In examining these relationships the article answers the following research questions: What is the overall structure of interactions across collaborative forums and how do these interactions support different forum goals? Data for the study were retrieved from 50 collaborative forums responding to weather warnings in Sweden between 2011 and 2015.

Linking bonding and bridging to forum objectives

The collaborative forums of interest here seek to combine the short-term operational goal of reaching a joint understanding of hazards (shared situation awareness, or a “common operating picture”) and the long-term strategic goal of sustaining collaboration among actors in the Swedish crisis management system. Given that actors represent different types of organizations, efforts to combine these goals bring challenges related to collaboration. What relationships among actors can be expected to emerge given these different objectives?

Collaborative forums are generally designed to provide a venue to facilitate information sharing that supports organizational missions and responsibilities (Fischer and Leifeld 2015). In natural hazards management, these goals are set on minimizing disturbance and maintaining or restoring societal functions (e.g., electricity, water supply, and transportation) and public services in the wake of severe weather. Collaborative forums hereby provide a venue for actors to collect and share information about weather forecasts, damage assessments, and response activities of other organizations. To achieve these goals, organizations can be expected to benefit from engaging in broad information searches across organization types.

In addition, these collaborative forums convene repeatedly over time to address hazards of different types and scales, which is likely to counteract the emergence of closed clusters of organizations of similar type that interact regularly within similar forums. Hence, it is assumed here that information exchange in pursuit of organization-level operational goals would preface bridging structures involving organizations of different types (Fig. 1a). In two-mode networks bridging is represented by centralization of activity around popular actors, as shown in Fig. 1d (Berardo 2014).

Fig. 1.

Configurations showing (a) heterophily, (b) homophily, (c) bonding social capital, and (d) bridging social capital.

Fig. 1.

Configurations showing (a) heterophily, (b) homophily, (c) bonding social capital, and (d) bridging social capital.

Another, more long-term objective of the collaborative forums of interest here is to sustain and develop collaboration by repeated interactions over time, which is also likely to preface bridging relationships. Engaging in information exchange for the purpose of sustaining collaboration over time is in essence an expression of altruistic behavior. For these exchanges, the objective is not primarily to share operational information but rather to demonstrate “good faith” and shared commitment by engaging actively in the forum despite low individual gains (cf. Ansell and Gash 2008). If this is the goal, one would expect information exchange across forums in combination with a high frequency of interactions across organizational types. It is thus assumed here that information exchange supporting the long-term development of collaboration is likely to foster open network structures and a higher frequency of brokerage activities across the network, akin to a bridging structure.

Forum interactions could, however, foster a completely different set of relationships. As noted above, some studies of interorganizational collaboration in response to natural hazards find that organizations avoid bridging relationships and forge close ties with other similar organizations (Fig. 1b). These relationships could emerge partly because the rapid escalation of some hazards brings immediate urgency and uncertainty, which reduce the time for actors to assess the relevance, reliability, and trustworthiness of potential collaborators. Attribute similarity (i.e., “homophily”) (Fig. 1b), therefore becomes an important heuristic for sorting out relevant partners and reducing transaction costs (Siciliano and Wukich 2015). In addition, although collaborative forums address a variety of events, some natural hazards occur with greater regularity than others. This pattern of repeated hazards is illustrated in Fig. 2, which visualizes the number of times two or more counties have been covered by the same weather warning (thicker links indicate greater overlap, i.e., regularity).

Fig. 2.

Spatial network of counties covered by the same weather warnings.

Fig. 2.

Spatial network of counties covered by the same weather warnings.

What this means is that several forums have convened to address relatively similar hazards in terms of hazard type and geographical and functional scope, which would logically sustain a similar pattern of information exchange among the same set of actors across these forums. In this perspective, the need to reduce transaction costs and the possibility that several forums address the same type of hazard would preface bonding relationships. Figure 1c illustrates a strong bonding configuration linking two organizations to each other through their joint participation in multiple forums (Berardo 2014; Berardo and Lubell 2016).

The analysis provides a nuanced depiction of how information exchanges are shaped in the context of collaborative forums. Following the discussion above, the expectation is that the types of relationships observed are contingent upon what objectives are pursued by actors within forums (cf. McAllister et al. 2014). Thus, it is assumed that depending on forum objectives, bonding, bridging, or a combination of the two would be most prominent.

2. Data collection and methods

a. Network data

Weather warnings in Sweden are issued by the Swedish Meteorological and Hydrological Institute (SMHI) according to a three-level scale (ranging from 1, indicating less severe, to 3, indicating most severe). The scale applies to any hazard type (wind, temperature, precipitation, etc.) and the level of the warning is set by forecasted variables (e.g., depth of water on floodplains) combined with the risk posed to public safety and disruption of public service. Fluvial warnings, for example, are determined by return period river flows with generic impacts: a 2–10-yr return period that can cause limited flooding is classified as a level-1 warning, a 10–50-yr return period with possible flooding as level 2, and a >50-yr return period with possible severe flooding as level 3. Collaborative forums (so-called collaboration conferences) are routinely summoned after a level-2 warning or higher, when there is an impending level-2 warning, or when any organization calls for it.

Network data for this study were collected by coding minutes from forum meetings (transcripts from audio conference format). These minutes are stored in a nationwide web-based information system (WIS), which is not publicly available. In preparation of this study, minutes from WIS were retrieved with the assistance from the Swedish Civil Contingencies Agency (MSB), which supplied transcripts from all forums in the study period. An initial selection was made to only include forums responding to weather warnings in 2011–15. Hence, the data exclude forums responding to other types of contingencies as well as forums that convened regularly in the absence of hazards. This selection procedure resulted in a total of 50 forums in the 5-yr period.

Each forum is populated by a unique set of actors; the organizations that participate vary from one forum to another and so do the individuals representing these organizations (organizations have several duty officers participating in forums, yet the data did not unveil what individual representatives participated in what forum). Some forums convened during several consecutive days in a row. These were treated here as separate forums since the problems addressed by the forums and the engagement in information exchange generally shifted from one day to another. A total of 51 organizations were represented in these forums (identified through participation lists included in the transcripts), including 8 business organizations (mainly electricity production companies), 22 national government agencies, and all of Sweden’s 21 County Administrative Boards (regional government agencies). All organizations are listed in  appendix B.

Minutes were coded to capture statements by all organizations engaging in information exchanges. A “statement” included any time an actor engaged in the forum, either to share or to request information from other actors. Each organization’s engagement in information exchange was coded as a dichotomy (engagement yes = 1, no = 0) and so the frequency of engagement (i.e., the number of times an organization made a statement) or the direction of ties (i.e., whether actors were a sender sharing information or receiver asking for information from others) was not considered. Thus, as opposed to network studies that examine collaboration based on participation (actors’ attendance in forums), this study provides evidence of individual actors’ active engagement in information exchange.

Most forum meetings followed a common structure where SMHI initially provided an updated weather forecast, which was followed by questions from the other participants regarding the forecast concerning, for example, the projected duration, intensity, and location of wind and precipitation. Finally, participants engaged in bilateral exchanges within the forums. These exchanges focused, for instance, on projections about electricity outages and what actions different organizations were conducting or planning, such as regional-level collaboration meetings or interruptions in public services (e.g., transportation systems) due to the hazard.

Organization type was coded according to three categories—national government agency, regional (county administrative boards), and business organizations—to assess whether information exchanges took place among similar organizations (homophily; Fig. 1b) or across organization types (heterophily; Fig. 1a) and whether specific types of organizations were particularly active. Data about forums, actors, and statements were combined into one two-mode rectangular matrix of information exchange with 51 rows (actors) and 50 columns (forums). Cells were assigned a value of 1 if an organization engaged in information exchange within a particular forum and 0 otherwise. The maximum number of statements per forum is 25 (28 October 2013) and the minimum is 4 (4 November 2014), with a mean of 12.2 across all forums. Figure 3 visualizes the actual network where squares represent forums (labels indicating date of the forum) and circles are organizations (see Table B1 in  appendix B for acronyms). Organization sizes are scaled using betweenness centrality, indicating the extent to which other organizations will have to go through an organization to reach other organizations. Betweenness centrality scores per organization are detailed in  appendix C.

Fig. 3.

Graphical representation of network data.

Fig. 3.

Graphical representation of network data.

b. Network configurations

There are essentially three network processes that structure interactions in networks of collaborative forums, including network activity (actors engaging in information exchange within forums), network centralization (forums connected to multiple actors; actors connected to multiple forums), and network closure (actors engaging in multiple forums together) (Lubell et al. 2014). Each one of these processes is associated with observable network configurations, which include smaller patterns of ties within a graph. Configurations that are likely outcomes of a social process within the network occur at a higher frequency than chance when controlling for other relevant processes. Documenting the frequency of different configurations is thus a way to empirically identify what relationships created and sustained the network (Lusher et al. 2013; McAllister et al. 2014). The interest here is primarily directed at configurations associated with bonding and bridging relationships respectively (Berardo 2014).

The analysis employs exponential random graph modeling (ERGM) to identify and describe the prevalence of configurations within networks, which is the empirical basis for identifying what relationships characterize collaboration in collaborative forums. ERGMs are conceptually similar to logistic regression analysis where a link between two nodes is predicted by configurations in the network. The observed network structure—in this case, the two-mode forums–actors network—is treated as a possible outcome of interdependent network processes. Hence, localized network configurations can be viewed as independent variables in the model explaining the network structure (Lubell et al. 2014). Formally, ERGM takes the form

 
formula

where Q is a given configuration, is the network statistic of the configuration included in the model, and is its parameter. The normalizing constant is generated over the graph space and forces the probability of all graphs to add to 1. Next, given the argument above that the network can also be represented by specific node attributes (in this case organization type), ERGM also includes a term to represent the probability of observing particular configurations in the graph x given node attribute y:

 
formula

Here, this addition to the equation represents the basic activity configurations for organizations, that is, whether organizations of any particular type are more likely to be active or not [equations derived from Berardo (2014)]. The computer package MPNet was used to conduct the analyses (Wang et al. 2014).

3. Results

Given the exploratory nature of this analysis, the effort to identify the best-fitting model followed common practice in some prior ERGM studies by starting with a set of simple configurations and gradually expanding the model by adding other configurations to find the model with the best fit (Bodin et al. 2016; Robins et al. 2011). After exploring different models combining density, network effects (forum and actor centralization and actor closure), and actor attributes (information exchange activities of different organization types), results indicated that the network was best represented by a few network attributes. Figure 4 presents results from the two-mode network ERGM, including parameter estimates for the configurations of theoretical interest. Estimation results and goodness-of-fit test results are detailed in  appendix A (see Table A1).

Fig. 4.

Results from the fitted multilevel exponential random graph model (ERGM).

Fig. 4.

Results from the fitted multilevel exponential random graph model (ERGM).

Parameter estimates summarized in Fig. 4 show that five configurations are statistically significant: density, actor centralization, forum centralization, three-path configuration, and forum-alternating two-path configuration. Actor closure was not significant and was removed from the final model. Two attribute configurations—dummy variables for national agency and business organizations—are not significant but contribute to improved model fit (see  appendix A). Estimated parameters are evidence indicating whether any given configuration is enhanced or suppressed. In this context, “enhanced” refers to a positive tendency among actors and/or forums to configure themselves according to any given configuration whereas “suppressed” suggests the opposite (Bodin et al. 2016).

Results reported in Fig. 4 indicate that two of the configurations were enhanced (actor and forum centralization) while the other three (density, three paths, and forum closure) were suppressed. The ERGM shows that the three-path configuration is suppressed (negative), indicating that the proportion of this configuration is significantly lower than expected by chance. The implication of this effect is a bit elusive but may imply that actors participating in forums attended by many other actors tend to avoid engaging in information exchange in multiple forums (cf. Wang et al. 2015). If so, one may speculate that some actors avoid engaging in information exchange simply because their information needs are fulfilled by listening in to exchanges taking place among other actors. An alternative interpretation of suppressed three-path configurations is that actors’ engagement in information exchange is not influenced by other actors’ engagement in other forums. In other words, when engaging in information exchange in any given forum actors do not take other actors’ engagement in other forums into account (cf. Hazir and Autant-Bernard 2012). This may suggest that the “reputation” of other actors, in terms of their experience or record of involvement in other forums, did not influence actors’ engagement in information exchange.

The forum-alternating two-path configuration is also suppressed in the model. This means that there is a significantly lower number of this configuration than what could be expected given its representation in the network. In general, interpreting enhanced or suppressed alternating two-star configurations should be done with some caution because these configurations depend on other model effects (Robins et al. 2009). According to Lubell et al. (2014), lower than expected levels of closure in bipartite networks is an indicator of brokerage activities. Conversely, in the case of collaborative forums, a positive parameter would indicate a pattern of recurrent information exchanges among the same set of actors, resulting in denser regions of the network and hence a more cohesive network structure.

The ERGM results show significantly low numbers of alternating two-path configurations in the model. In substance, this suggests that actors avoid engaging in information exchange within the same set of forums. In the opposite scenario of an enhanced effect (a positive effect of alternating two-stars), one would conclude that organizations engage in information exchange in the same set of forums, which in turn would be evidence of bonding. In two-mode networks, there is no direct connection between organizations—they are connected to each other only through their joint engagement in information exchange in the same forum. Hence, engaging in information exchange in the same forums can be a way to build and sustain relationships. However, in this case the effect is suppressed and thus actors seem to actively reduce network closure by engaging in information exchange through more open relationships.

The density parameter (a measure of general network activity predicting the presence or absence of an actor–forum connection, equivalent to the intercept in a linear regression model; see Lusher and Robins 2013) is negative and significant. This is evidence that the formation of links in the network can be attributed to certain specific nonrandom processes that are driven by configurations included in the model (Berardo 2014). The positive parameter of the forum centralization configuration indicates a greater prevalence of forum–actor connections than what could be expected by chance. This shows that there are some highly central forums in the network (i.e., forums with more forum–actor ties). A more detailed examination of the most central forums (with the highest number of actors engaging in information exchange) reveals that the most central forums are those that convene in relation to storms in the winter or late fall. Out of the 10 most central forums (with 15 or more stakeholders engaging in information exchange), 8 convened in the fall or winter (October–March) in response to storms and/or intense snowfall (see Fig. 5). This pattern may be explained by a combination of broad geographical dispersion and uncertainty regarding the location and magnitude of hazard impacts.

Fig. 5.

Collaborative forums by date and number of stakeholders engaging in information exchange.

Fig. 5.

Collaborative forums by date and number of stakeholders engaging in information exchange.

Finally, the actor centralization configuration is also enhanced, which suggests that certain actors have a higher level of engagement in information exchange across forums than others. Here, this can be explained by the fact that some actors occupy more central positions in the hazard management system than others and also that some forums are more important than others given hazard magnitude and severity. Figure 3 visualizes these patterns. If interpreted in conjunction with centrality measures ( appendix C), these data show that three actors hold particularly central positions in the network, including SMHI, MSB, and the Swedish Transportation Agency (STA). This is expected since MSB is the main organizer of the forums and SMHI has a standing item on the agenda by presenting updated weather forecasts. The fact that STA is a central actor is also rather obvious since most major natural hazards bring challenges to the transportation infrastructure (roads and railways), which is the responsibility of STA.

4. Discussion

The ERGM results showed that network closure configurations were suppressed, indicating significantly fewer closure configurations than expected by chance. This is illustrated by the lower than expected number of alternating two-stars. In addition, centralization of activity around popular actors has been used as a proxy to indicate bridging relationships, based on the assumption that certain central actors will take a coordinating role to facilitate information exchanges among individuals who would otherwise not be connected (Berardo 2014). In this case, the results showed that the effect of actor centralization was enhanced, providing additional evidence of a more open network structure (indicating that some actors engage in information exchanges in multiple forums and thereby possibly acting as brokers). Taken together, these observations suggest that there is a tendency among the actors to actually avoid engaging in information exchanges within closed subgroups. That is, there are certain constellations of actors that do not tend to engage in information exchange repeatedly within the same forums but rather in different forums, which in aggregate reduces the level of closure within the forum network. Several implications emerge from these observations.

First, the tendency of bridging in this case can partially be attributed to the design of collaborative forums. Natural hazards are variable phenomena and each event represents a unique combination of severity, geographical, and functional scope. Information exchanges are thus fitted to the attributes of each individual hazard; it involves actors with relevant resources and competencies as well as actors whose functional responsibilities are at risk. At the same time, some hazards occur with greater regularity than others. In Sweden storms are more common in the south compared to other parts of the country (Bärring and von Storch 2004) and thus one would have expected bonding relationships to be more prominent given that several hazards affected roughly the same set of actors (specifically counties in the same region) (Fig. 2). The fact that these bonding relationships were suppressed can partially be explained by network brokerage, that is, the fact that some central actors participate in multiple forums and thereby connect otherwise disjoined subgroups (Berardo 2014; Lubell et al. 2014).

Second, here network closure is linked to bonding and bridging relationships. It was theorized that closed configurations associated with bonding would facilitate short-term hazard response, which generally depends on smaller type of subnetworks that are adapted to the situation (in terms of including the relevant set of organizations to pool relevant resources and competencies that are “fit” to the specific hazard) and with high levels of trust among the participants (Bodin and Nohrstedt 2016). Conversely, in theory bonding is less likely to support the more long-term objective to sustain collaboration, which would require active engagement of a broader set of actors. The implication of bridging relationships for hazard management might thus be the opposite from bonding; while bridging may be helpful to sustain broad collaboration over time, it may be less effective for short-term hazard response.

Yet, at closer inspection the relationship between bridging social capital and hazard response performance is not as straightforward. In theory, bridging relationships may in fact support hazard response performance while bonding can be more problematic. This assumption goes back to the insight that effective hazards management typically involves balancing a generic “all-hazards” approach with more specific contingency plans targeted at certain hazard scenarios. Thus, in order to enhance effective collaborative responses actors can benefit from engaging in more closed network structures fitted to specific known risks while also facilitating and maintaining broader networks for extraordinary and partially unprecedented contingencies.

Such “synchronous” networking, which entails elements of both bonding and bridging in a set of nested collaborative forums, should ideally be accompanied by efforts to build interpersonal trust and to include the “right” actors in the network. Studies have noted that the boundaries of crisis responder networks are often drawn too narrowly, thereby excluding important organizations from collaboration (Boin and ’t Hart 2010). In this perspective, bonding relationships can provide a useful resource in response to known events while bridging relationships might foster readiness to respond to more complex and uncertain events. The latter is particularly important following the importance of innovation, improvisation, and adaptation in response to more complex events and situations (Mendonça et al. 2007; Farazmand 2007).

Third, any effort to draw inferences about network structure must take the specific governance setting into account (Berardo and Lubell 2016; Siciliano and Wukich 2017). This study covers information exchange within 50 collaborative forums that convened in response to weather warnings in Sweden. In addition to these forums, MSB also organizes (since 2013) weekly collaboration conferences gathering about 30 participants on average. These conferences are assembled regardless of any contingency to share information as a basis for developing collective risk assessments and also to give the participants an opportunity to regularly practice their collaborative skills. Given some overlap in attendance between the weekly conferences and the collaborative forums (both are attended by a duty officer from each organization), it is possible that the weekly collaboration conferences support bonding relationships over time in a broader circle of organizations, which in turn may mitigate potentially negative effects of bridging relationships in collaborative forums responding to hazards. Regular collaboration conferences may thus provide preexisting relationships that facilitate collaboration in response to natural hazards, which is also an explicit objective of these meetings. Actors can furthermore be expected to engage in direct personal contacts with each other outside collaborative forums, which also need to be taken into account when drawing conclusions about the possible consequences about bonding and bridging relationships.

Finally, in this case actor attributes were neither enhanced or suppressed, which is evidence that configurations incorporating organizational type do not have to be included in the ERGM to adequately represent the network. This is a findings that has been reported elsewhere as well (Berardo and Lubell 2016). Here, actor attributes were included as dummy variables extending the simple model to test whether network activity was higher or lower for any type of organization ( appendix A; see Table A1). Given that neither of these attribute configurations was significant, it can be concluded that organization type was not an important attribute explaining stakeholder interactions. One implication is that actors did not engage in information exchange to a greater or lesser extent with representatives of similar or dissimilar organizations, suggesting that tendencies for heterophily (Fig. 1a) or homophily (Fig. 1b) were not evident in this specific setting. The absence of relationships based on organizational similarity or dissimilarity may also speak against linking social capital (i.e., exchanges based driven by power asymmetry). Yet, it is not possible to draw any certain conclusions regarding linking social capital in this case, which would require measures of power relationships among organizations.

This finding is at odds with some other studies of collaboration in natural hazards management, which show that organizations seek collaboration with similar organizations (Siciliano and Wukich 2015, 2017). One plausible explanation for these differences is that bigger and more complex hazard events impose immense pressure on organizations to reduce transaction costs, forging ties among similar organizations (Jung et al. 2018). Alternatively, these differences may also stem from differences in how studies measure collaborative ties as well as what specific organizational attributes are accounted for in the analysis (Berardo and Lubell 2016). These alternative explanations remain speculations that warrant further examination in future work.

5. Conclusions

There is growing interest in collaboration networks as strategy of coping effectively with natural hazards at different scales, yet prior research gives little insight into what relationships emerge among multiple stakeholders participating in collaboration forums and how these relationships affect attainment of forum goals. This study empirically examined interactions among stakeholders in 50 collaborative forums in Sweden from 2011 to 2015 to specify what relationships characterized the network, focusing specifically on bonding and bridging social capital.

Documenting network configurations to identify stakeholder interactions constitutes a novel approach with potential for advancing network theory (Berardo 2014; Bodin 2017). Studies demonstrate, for instance, tendencies among policy actors to engage in bridging relationships across organizational types to advocate for organizational agendas, resulting in biased and selective use of information (McAllister et al. 2014). While these findings apply in stable policy environments, this study has examined forums responding to natural hazards, which present a different set of external conditions for stakeholder interaction including uncertainty, urgency, and a need for swift collective action. Given these conditions, the study linked specific network configurations to forum objectives, including operational organizational goals and the long-term sustainability and development of collaboration.

The study finds that in this case forum participants avoided engaging in closed, close-knit relationships. Rather, they engaged in bridging relationships mediated by network brokers that coordinated information exchange among actors that would otherwise be disconnected. Why did this open network structure emerge?

Forum interactions may be seen as manifestations of the institutional system in which forums are embedded as well as the rules that structure information exchange among stakeholders. The forums studied here are one type of arrangement that supports collaboration in hazards management and how stakeholders interact in these forums is likely affected by their engagement in other forums. While these forums are utilized to exchange information within a broader set of actors to achieve joint situation awareness at the national level, other forums might foster more closed structures among actors sharing similar short-term interests. In Swedish natural hazards management, collaborative forums are accompanied by regular weekly forums at the national level, subregional forums for interorganizational coordination, and forums coping with other types of contingencies. In this complex system of overlapping and interconnected forums it can be hypothesized that different forums support different types of interactions; whereas national-level forums may feed bridging social capital, it is likely that subregional forums are more conducive to bonding relationships.

Relationships between actors participating in these forums are also likely affected by the rules that prescribe what actors participate in forum meetings, how information is communicated, and what outcomes are desired (cf. Carter et al. 2016). These rules provide a standardized procedure for information exchange that places certain actors in central positions in the network. For example, the SMHI is given a central role as it shares weather forecasts with the other forum participants who in turn request specific information from SMHI about the expected severity and trajectory of hazards. Hence, in this case broker positions are partially predetermined by institutional rules, which in turn influence what relationships emerge among actors as they exchange information. In this regard, this study corroborates results reported elsewhere, which suggests that these broker positions are likely to be occupied by public agency organizations that connect other actors from different collaborative forums (Fischer et al. 2018).

One obvious limitation of this analysis is that it only partially links stakeholder interactions to the effectiveness of collaborative forums. One common approach to evaluate the effectiveness of networks is to assess whether they “actually deliver what they are supposed to deliver” (Sørensen and Torfing 2009, p. 242). This study has engaged in a discussion about how different patterns of information exchange may influence the ability to meet forum objectives. Nevertheless, whether and how certain relationships actually influence forum effectiveness is beyond the scope of this study. The analytical approach applied here can, however, inform future research on forum outcomes.

One issue for future work on natural hazards management is to investigate how patterns of information exchange influence the capacity to achieve “shared situation awareness,” which is one goal of these forums. Some scholars argue that shared situation awareness is crucial for effective hazards management, yet others point to the difficulties involved when attempting to consolidate information in complex situations involving heterogeneous groups of stakeholders. Under these conditions all information will not be relevant to all parties and specific information that is relevant to some parties will be left out (Mendonça et al. 2007). In this regard, an important next step would be to examine whether and how forum interactions influence the perceived relevance as well as the practical importance of shared situation awareness as a tool for collective natural hazard management.

Situation awareness, however, is merely one collective outcome of collaborative forums. It has been argued here that bridging relationships also support the long-term objective of enhancing and developing collaboration in a broader circle of stakeholders. In theory, such interactions can ensure access to new information and foster development of innovative solutions (Berardo 2014). In contrast, bridging relationships might be less effective for short-term hazard response, which generally relies on more closed subgroups conducive to cooperation and trust to be effective. These assumptions remain theoretical speculations and more research is warranted to assess the effects of different types of configurations in relation to different forum goals. As suggested by this study, one particularly interesting issue involves the ability to maintain “synchronous” networking—nested collaborative forums with elements of both bonding and bridging—in complex systems of multiple forums.

The findings of the study offer a few reminders for practitioners in natural hazard planning and management. Collaborative forums exemplify a collaborative arrangement that has been put in place to facilitate information exchange and coordination of joint activities between organizations across sectors, jurisdictions, and levels of authority. Yet, as demonstrated in this study collaboration can unfold in many different ways, which gives rise to network structures that bring different opportunities and constraints for collective natural hazards management. The analytical approach adopted in this study (and elaborated further elsewhere) can thus serve as a diagnostic tool for disentangling various substructures within collaboration networks and how these affect information sharing and, ultimately, goal attainment of collaborative forums.

Specifically, the study has shown that information exchanges in forums span organizations of different types, levels of authority, and the public–private divide. These relationships have been enabled by actors that engage in multiple forums and thereby link otherwise disconnected organizations. These brokering roles are important as they coordinate, and thereby “control,” information flows between other actors. Therefore, an important practical concern is whether actors that occupy these brokering roles actually transmit relevant information and if this information is exploited in ways that enhance natural hazard management. Taking the step from merely observing these relationships to developing “best practice” strategies for improved collaboration and collective natural hazards management is no easy task, however. This would require a more ambitious study design to connect forms of network social capital with associated capacity to solve specific collective-action problems through forums and, ultimately, with goal attainment.

Acknowledgments

This work was supported by funding from the Swedish Research Council Formas, under project 2012-627 entitled “Collaborative Governance and Local Community Resilience to Environmental Shocks in Sweden.” The author is grateful to Örjan Bodin for steadfast assistance on the network analysis, to Karin Jonsson at the Swedish Meteorological and Hydrological Institute (SMHI) for supplying information on weather warnings, and to Halldor Stolt at the Swedish Civil Contingencies Agency (MSB) for providing access to forum meeting transcripts. The author would also like to thank Susan Cutter and three anonymous reviewers of Weather, Climate, and Society for insightful and constructive comments on a previous version of the article.

APPENDIX A

Fitting the Two-Mode Exponential Random Graph Model

 Appendix A summarizes the steps taken to fit the exponential random graph model using the package MPNet (Wang et al. 2014). In this case, the final model was fitted in two steps. In the first step, multiple models were tested to estimate a simple model including density, centralization, and closure for actors and forums respectively (see McAllister et al. 2014). These configurations capture the level of activity around actors and forums (centralization configurations) and closure among actors and forums and are recommended starting points for models for undirected networks (Robins and Lusher 2013). Results from this simple model returned significant parameter estimates for several network configurations except actor closure, which was then excluded from the model.

The goodness of fit (GOF) was assessed by comparing the configurations from the simple model (observed network) with configurations resulting from a number of simulations using the coefficients from the converged simple model plus a range of other graph statistics. The GOF tests showed that the resulting simple model (Table A1) had a relatively good fit when each network statistic was assessed independently, with t ratios less than 0.1 for statistics modeled in the ERGM and t ratios less than 2.0 for other network statistics, which is generally considered a good model fit. However, based on an assessment of the overall fit of the model (taking correlations among network statistics into account) the simple model had a relatively poor fit. This is indicated by a relatively big Mahalanobis distance measure (6363), which is a measure of how far away the observed network is from the center of the distribution. Accordingly, lower Mahalanobis values indicate better fit (Lubell et al. 2014; Lusher et al. 2013).

Table A1.

Comparison of estimation and goodness-of-fit results for simple and extended ERGM models.

Comparison of estimation and goodness-of-fit results for simple and extended ERGM models.
Comparison of estimation and goodness-of-fit results for simple and extended ERGM models.

In the second step, I ran multiple extended models to test if the inclusion of attribute configurations improved the model further (see Berardo 2014; McAllister et al. 2014). In this step, estimation and goodness-of-fit results for the initial simple model were compared with several extended models including the network configurations in the simple model plus combinations of activity configurations for the three organization types (national agencies, regional, and business) engaging in forums. Specifically, these activity configurations measure the level of engagement for each organization type respectively and adding them is a way to examine the level of activity (in this case engagement in information exchange in forums) for specific types of organizations compared to other types of organizations. Thus, six extended models were tested, each using the simple model plus one, two, or three actor attribute configurations in different combinations. These models were examined in steps by comparing the results of the estimations and GOF tests with the results for the simple model. In cases where the initial model did not converge, it was reexamined with a higher multiplication factor (up to 50) before concluding the model was degenerate (see Robins and Lusher 2013).

Results from these tests showed that only one specific model (extended model; Table A1)—extending the simple model with attribute configurations for national agencies and business organizations—converged with t values less than 0.1 for all coefficients. The goodness of fit test showed that the extended model provided a better fit compared to the simple model as indicated by low t ratios for individual statistics and also a considerably smaller Mahalanobis distance measure (423). It should be noted here however that the GOF t ratio for national agencies was −0.159, thus slightly above the 0.1 threshold for t ratios indicating a good fit. However, as some ERGM studies (Bodin et al. 2016) apply a 0.2 threshold for GOF t ratios, this is considered to be within the range of acceptable goodness of fit. However, while the extended model indicated a statistically significant effect of all network configurations, the two-attribute configuration coefficients were not statistically significant and so the effects of these configurations are thus indistinguishable from 0. In summary, based on the comparison between the simple model and the extended model, the conclusion is that the extended model is the best-fitting model in this case (i.e., the network is best represented by purely structural effects that do not depend on network node characteristics such as homophily or heterophily; see, e.g., Lusher and Robins 2013).

APPENDIX B

List of Actors by Organization Type

The stakeholders are listed by organization type; business, national government agencies, and regional [County Administrative Boards (CABs)] with their respective roles and responsibilities (Table B1).

Table B1.

Forum participants.

Forum participants.
Forum participants.

APPENDIX C

Centrality Scores for Forum Actors

Actors’ centrality scores were calculated to identify what actors held the most central positions in the actor–forum network. Betweenness centrality measures were used to indicate the uniqueness of the links by an organization, which is a way to test implications for across-network connectivity if any given actor is removed from the network. For instance, if the shortest path between actors n2 and n3 is through actors n1 and n4 (given a geodesic of n2n1n4n3), then actors n1 and n4 may “control” interaction between n2 and n3 (Wasserman and Faust 1994, p. 188). Betweenness centrality thus measures the “power” or “influence” of an actor by controlling interactions of pairs of other network actors.

Eigenvector centrality considers the degree of a node’s alters measuring an actor’s connection with more popular nodes (i.e., actors connected to many other actors) (Bonacich 1987). McAllister et al. (2014) suggest that betweenness centrality can be combined with eigenvector centrality measures to indicate bridging roles of actors. Concretely, they suggest that actors with high betweenness scores compared to their eigenvector scores have important bridging roles (Fig. C1). This is central to the current analysis, given indications of bridging relationships in the forum network. In this case, the County Administrative Board (CAB) of Västerbotten (CAB-VB) and the Swedish Armed Forces (SAF) stand out as actors with potentially important bridging roles. By contrast, several CABs (Gotland, Kronoberg, Kalmar, Östergötland, and Örebro) had higher eigenvector scores relative to their betweenness scores, indicating that they do not play bridging roles.

Fig. C1.

Histograms of actor centrality scores.

Fig. C1.

Histograms of actor centrality scores.

REFERENCES

REFERENCES
Aldrich
,
D. P.
, and
M. A.
Meyer
,
2014
:
Social capital and community resilience
.
Amer. Behav. Sci.
,
59
,
254
269
, https://doi.org/10.1177/0002764214550299.
Andrew
,
S.
, and
J.
Carr
,
2013
:
Mitigating uncertainty and risk in planning for regional preparedness: The role of bonding and bridging relationships
.
Urban Stud.
,
50
,
709
724
, https://doi.org/10.1177/0042098012455718.
Andrew
,
S.
,
S.
Arlikatti
,
L.
Siebeneck
,
K.
Pongponrat
, and
K.
Jaikampan
,
2016
:
Sources of organizational resiliency during the Thailand floods of 2011: A test of the bonding and bridging hypotheses
.
Disasters
,
40
,
65
84
, https://doi.org/10.1111/disa.12136.
Ansell
,
C.
, and
A.
Gash
,
2008
:
Collaborative governance in theory and practice
.
J. Public Adm. Res. Theory
,
18
,
543
571
, https://doi.org/10.1093/jopart/mum032.
Bärring
,
H.
, and
H.
von Storch
,
2004
:
Scandinavian storminess since about 1800
.
Geophys. Res. Lett.
,
31
,
L20202
, https://doi.org/10.1029/2004GL020441.
Berardo
,
R.
,
2014
:
Bridging and bonding capital in two-mode collaboration networks
.
Policy Stud. J.
,
42
,
197
225
, https://doi.org/10.1111/psj.12056.
Berardo
,
R.
, and
J.
Scholz
,
2010
:
Self-organizing policy networks: Risk, partner selection, and cooperation in estuaries
.
Amer. J. Pol. Sci.
,
54
,
632
649
, https://doi.org/10.1111/j.1540-5907.2010.00451.x.
Berardo
,
R.
, and
M.
Lubell
,
2016
:
Understanding what shapes a polycentric governance system
.
Public Adm. Rev.
,
76
,
738
751
, https://doi.org/10.1111/puar.12532.
Bodin
,
Ö.
,
2017
:
Collaborative environmental governance: Achieving collective action in social-ecological systems
.
Science
,
357
,
659
, https://doi.org/10.1126/science.aan1114.
Bodin
,
Ö.
, and
D.
Nohrstedt
,
2016
:
Formation and performance of collaborative disaster management networks: Evidence from a Swedish wildfire response
.
Global Environ. Change
,
41
,
183
194
, https://doi.org/10.1016/j.gloenvcha.2016.10.004.
Bodin
,
Ö.
,
G.
Robins
,
R.
McAllister
,
A.
Guerrero
,
B.
Crona
,
M.
Tengö
, and
M.
Lubell
,
2016
:
Theorizing benefits and constrains in collaborative environmental governance: A transdisciplinary social-ecological network approach for empirical investigations
.
Ecol. Soc.
,
21
,
40
49
, https://doi.org/10.5751/ES-08368-210140.
Boin
,
A.
, and
P.
’t Hart
,
2010
:
Organising for effective emergency management: Lessons from research
.
Aust. J. Public Adm.
,
69
,
357
371
, https://doi.org/10.1111/j.1467-8500.2010.00694.x.
Bonacich
,
P.
,
1987
:
Power and centrality: A family of measures
.
Amer. J. Sociol.
,
92
,
1170
1182
, https://doi.org/10.1086/228631.
Carter
,
D.
,
C.
Weible
,
S.
Siddiki
, and
X.
Basurto
,
2016
:
Integrating core concepts from the institutional analysis and development framework for the systematic analysis of policy designs: An illustration from the US National Organic Program regulation
.
J. Theor. Polit.
,
28
,
159
185
, https://doi.org/10.1177/0951629815603494.
Farazmand
,
A.
,
2007
:
Learning from the Katrina crisis: A global and international perspective with implications for future crisis management
.
Public Adm. Rev.
,
67
,
149
159
, https://doi.org/10.1111/j.1540-6210.2007.00824.x.
Fischer
,
M.
, and
P.
Leifeld
,
2015
:
Policy forums: Why do they exist and what are they used for?
Policy Sci.
,
48
,
363
382
, https://doi.org/10.1007/s11077-015-9224-y.
Fischer
,
M.
,
M.
Angst
, and
S.
Maag
,
2018
: Co-participation in the Swiss water forum network. Int. J. Water Resour. Dev., https://doi.org/10.1080/07900627.2017.1374929,
in press
.
Hazir
,
C.
, and
C.
Autant-Bernard
,
2012
: Using affiliation networks to study the determinants of multilateral research cooperation: Some empirical evidence from EU framework programs in biotechnology. GATE Working Paper 1212, 31 pp., https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2060275.
Jung
,
K.
,
M.
Song
, and
R.
Feiock
,
2018
: Isolated and broken bridges from interorganizational emergency management networks: An institutional collective action perspective. Urban Affairs Rev., https://doi.org/10.1177/1078087417690257,
in press
.
Lubell
,
M.
,
G.
Robins
, and
P.
Wang
,
2014
:
Network structure and institutional complexity in an ecology of water management games
.
Ecol. Soc.
,
19
,
23
37
, https://doi.org/10.5751/ES-06880-190423.
Lusher
,
D.
, and
G.
Robins
,
2013
: Personal attitudes, perceived attitudes, and social structures: A social selection model. Exponential Random Graph Models for Social Networks, D. Lusher, J. Koskinen, and G. Robins, Eds., Cambridge University Press, 189–201.
Lusher
,
D.
,
J.
Koskinen
, and
G.
Robins
,
2013
: Exponential Random Graph Models for Social Networks. Cambridge University Press, 336 pp.
McAllister
,
R.
,
R.
McCrea
, and
M.
Lubell
,
2014
:
Policy networks, stakeholder interactions and climate adaptation in the region of South East Queensland, Australia
.
Reg. Environ. Change
,
14
,
527
539
, https://doi.org/10.1007/s10113-013-0489-4.
Mendonça
,
D.
,
T.
Jefferson
, and
J.
Harrald
,
2007
:
Collaborative adhocracies and mix-and-match technologies in emergency management
.
Commun. ACM
,
50
,
44
49
, https://doi.org/10.1145/1226736.1226764.
Nohrstedt
,
D.
,
2016
:
Explain mobilization and performance of collaborations in routine emergency management
.
Admin. Soc.
,
48
,
135
162
, https://doi.org/10.1177/0095399712473983.
Nohrstedt
,
D.
and
Ö.
Bodin
,
2014
:
Evolutionary of crisis preparedness collaboration: Resources, turbulence and network change in Swedish municipalities
.
Risk Hazards Crisis Public Policy
,
5
,
134
155
, https://doi.org/10.1002/rhc3.12055.
Robins
,
G.
, and
D.
Lusher
,
2013
: Illustrations: Simulation, estimation, and goodness of fit. Exponential Random Graph Models for Social Networks, D. Lusher, J. Koskinen, and G. Robins, Eds., Cambridge University Press, 167–185.
Robins
,
G.
,
P.
Pattison
, and
P.
Wang
,
2009
:
Closure, connectivity and degree distributions: Exponential random graph (p*) models for directed social networks
.
Soc. Networks
,
31
,
105
117
, https://doi.org/10.1016/j.socnet.2008.10.006.
Robins
,
G.
,
L.
Bates
, and
P.
Pattison
,
2011
:
Network governance and environmental management: Conflict and cooperation
.
Public Adm.
,
89
,
1293
1313
, https://doi.org/10.1111/j.1467-9299.2010.01884.x.
Siciliano
,
M.
, and
C.
Wukich
,
2015
:
Network features and processes as determinants of organizational interaction during extreme events
.
Complexity Governance Networks
,
2
,
23
44
, https://doi.org/10.7564/14-CGN10.
Siciliano
,
M.
, and
C.
Wukich
,
2017
:
Network formation during disasters: Exploring micro-level interorganizational processes and the role of national capacity
.
Int. J. Public Adm.
,
40
,
490
503
, https://doi.org/10.1080/01900692.2016.1140200.
Sørensen
,
E.
, and
J.
Torfing
,
2009
:
Making governance networks effective and democratic through metagovernance
.
Public Adm.
,
87
,
234
258
, https://doi.org/10.1111/j.1467-9299.2009.01753.x.
Tobin
,
G. A.
,
1999
:
Sustainability and community resilience: The holy grail of hazards planning?
Global Environ. Change
,
1
,
13
25
, https://doi.org/10.1016/S1464-2867(99)00002-9.
Turner
,
B. L.
, and Coauthors
,
2003
:
A framework for vulnerability analysis in sustainability science
.
Proc. Natl. Acad. Sci. USA
,
100
,
8074
8079
, https://doi.org/10.1073/pnas.1231335100.
Wang
,
P.
,
G.
Robins
,
P.
Pattison
, and
J.
Koskinen
,
2014
: MPNet: Program for the simulation and estimation of (p*) exponential random graph models for multi-level networks. http://www.melnet.org.au/pnet/.
Wang
,
P.
,
G.
Robins
, and
P.
Matous
,
2015
: Multilevel network analysis using ERGM and its extensions. Multilevel Network Analysis, E. Lazega and T. Snijders, Eds., Springer, 125–143.
Wasserman
,
S.
, and
K.
Faust
,
1994
: Social Network Analysis. Cambridge University Press, 825 pp.
Woolcock
,
M.
,
2001
:
The place of social capital in understanding social and economic outcomes
.
Isuma Can. J. Policy Res.
,
2
,
11
17
.

Footnotes

© 2018 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).