Abstract
Recent interorganizational communication research has taken up the question: why are networks structured the way they are? This line of inquiry has advanced communication network research by helping explain how and why networks take on certain structures or why certain organizations become positioned advantageously (or not). Yet, those studies assume relationships among organizations are either present or absent. This study considers how the strength of ties and multiplex relationships among organizations may reveal a more complex explanation for why networks take on certain structures. Our results challenge some long held assumptions about the mechanisms that influence network formation. For instance, our results offer important insights into the consequences of closure mechanisms, the applicability of preferential attachment to real-world networks, and the nuances of homophily in network formation on multidimensional relationships in a communication network. Implications for interorganizational networks are discussed.
Keywords
Research on interorganizational relationships using network theories and methods has grown across various communication subfields, noticeably in studies of civil society and international development organizations (e.g., O’Brien et al., 2019). Interorganizational networks in civil society are thought to generate “synergistic effects; that is, more and/or better outcomes are attained than if the network partners acted independently” (Brinkerhoff, 1999, p. 126). Researchers in this area initially sought to describe the structural characteristics of relationships to assess a network’s capacity for collective action and identify organizations’ positions in networks to better understand their abilities to access or influence available resources (e.g., Doerfel & Taylor, 2004; Flanagin et al., 2001; Taylor & Doerfel, 2003).
More recently, however, scholars have directed their attention to the antecedents of network structures by asking: why are civil society networks structured the way they are? Instead of examining strategic outcomes connected to network structures or an organization’s position in the network, this newer line of research focuses on developing what we might call a theory of civil society networks—concentrating on the factors that influence why organizations form relationships with certain others and why organizations occupy the positions they do (Atouba & Shumate, 2015; Lai et al., 2019). Knowing the antecedents and mechanisms that underlies relationship formation and dissolution can help us better understand how and why civil society networks take on certain structures or how to leverage such networks.
To that end, studies examining the predictors of interorganizational relationships have used analytic frameworks that examine the patterns of relations among civil society and development actors like NGOs, particularly ones that remain faithful to network assumptions of interdependence (Atouba & Shumate, 2010, 2015; Yang & Liu, 2018). Such research attempts to capture the relational properties (i.e., endogenous factors like reciprocity and centralization) as well as organizations’ attributes (i.e., exogenous factors like age and organizational type) to investigate the aspects that influence interorganizational relationship formation using exponential random graph modeling (ERGM) (Shumate & Palazzolo, 2010). While such developments have been fruitful, the theorizing rooted in these works assume relations are either present or absent.
For instance, Diani (2015) theorizes social networks as the cement of civil society, what he calls the relational view. The relational view contrasts and even compliments the aggregate view, which views social change as emerging from the modification of various input variables (e.g., increases education, organizational resources, etc.). The relational view looks at “how actors carrying different traits and orientations link to each other in distinctive structural patterns” (p. 4). Thus, to understand social change is to understand how various actors are coherently connected together. Just like if one was to install modifications to a cement building, they must know how various features are structurally arranged together. Otherwise, alterations to the structure could create unintended consequences, hence the phrase, built on a house of cards.
We consider the qualities of cement in a civil society network by examining tie strength for multiple types of ties. As with any construction project, big or small, it is vital to understand the grades (e.g., Megapascals) and types (e.g., Type I, Type II, etc.) of cement. Similarly, network ties can be primarily characterized by strength and multiplexity (Kilduff & Tsai, 2003, pp. 32–33). Varying levels of strength and different types of ties are often related to different goals, purposes, and are even interdependent themselves (Lee & Monge, 2011). Incorporating these factors builds and expands upon the relational view by further describing the strength and types of relations used to hold together civil society.
The purpose of this study is to expand upon the relational view of civil society by considering the role tie strength and multiplexity among organizations. We believe that for a theory of civil society networks to advance, it is critical for research to account for the fact that interorganizational relationships are multiplex and vary in strength. Moreover, we employ valued exponential random graph modeling (VERGM) with valued multiplex relationships rather than binary uniplex relationships (Pilny & Atouba, 2018). This allows us to theorize how tie strength impacts relationships in a civil society network.
Global Civil Society and Interorganizational Networks
Global civil society is “the sphere of ideas, values, institutions, organizations, networks, and individuals located between the family, the state, and the market and operating beyond the confines of national societies, polities, and economies” (Anheier et al., 2001, p. 21). Research on global civil society networks has sought to understand the various players in those networks, the shape of such networks, and what influences relationship formation. NGOs, as institutional manifestations of civil society, have assumed increased visibility at the international level to tackle global problems (Taylor, 2010). As such, NGOs, and the relationships among other civil society organizations, have been the focus of many prior works.
The structure of interorganizational relationship networks in civil society and development contexts has been a subject of interest among communication scholars for nearly two decades (e.g., Doerfel & Taylor, 2004; Taylor & Doerfel, 2003). Communication scholars interested in civil society have therefore sought to understand how network structures impact the efficacy of individual organizations as well as collective efforts. Some have specifically studied the role of network composition and structure on organizational efficacy. A series of studies examining the development of Croatian civil society hypothesized that organizations with certain network positions would have certain relational outcomes (Doerfel & Taylor, 2004, 2017; Taylor & Doerfel, 2003, 2011). Taken together, these studies and others that followed provide insights into how organizations can build relationships with others to position themselves in networks to more effectively achieve organizational and collective goals (Kent et al., 2016).
Another area in the literature has sought to reveal the mechanisms that influence tie formation at multiple network levels. At the dyadic level, Lee and Monge (2011) used the theory of embeddedness (Granovetter, 1985) to explain multiplex relations when two firms with one type of tie (e.g., knowledge sharing tie) had a greater likelihood of having another type of tie (e.g., project implementation tie). And, at the triadic level, they used transitivity to predict and explain their findings that firms with common partners had a greater likelihood of forming relationships. Shifting toward the attributes of the organizations themselves, scholars have also drawn on homophily theory to examine the degree to which shared attributes, geography, and institutional factors influences relationship formation (Atouba & Shumate, 2015).
Previous interorganizational civil society research has been instrumental in enhancing our understanding of network development; yet, there are two important limitations that need addressed. First, many of the aforementioned and related studies conceptualized and analyzed relationships as either present or absent (Atouba & Shumate, 2010; Shumate et al., 2005). As such, existing theory rests on the understanding of relationships as a binary—present or absent—and does not fully capture the nuances in interorganizational relationships (Pilny & Atouba, 2018). Indeed, this is due in part to the analytical tools available. Table 1 1 reports the prevalence of binary data in many ERGM studies on interorganizational relationships. This study explores if the local mechanisms driving relationship formation among organizations can be explained by tie strength. Table 1 also indicates the endogenous and exogenous factors that previous ERGM studies found to be significant, varied, or null. It is possible that certain network structures might change if tie strength is taken into account, as we do in this study.
Summary of Selected ERGM Studies in the Context of Interorganizational Relationships.
Note. + = positive; − = negative; −/+ = varied; ∅ = null; Triads also included transitivity. For k-In-Stars, GWI was interpreted as anti-preferential attachment so, a negative value is centralization. Resource space also included funding source or issue space. Additional exogenous factors were also considered but not included in the table (i.e., centrality Lai et al., 2019; Lee & Monge, 2011). The shaded columns indicate different categories of information gathered from each study.
Multi data source included: aSurvey & Co-occurrences & Social Media.
Survey & Hyperlink & Twitter.
Second, many previous studies have only examined a single tie. Rather than assuming that mechanisms influencing relationships work the same for each type of relationship, we consider whether mechanisms studied in prior NGO research (Atouba & Shumate, 2010, 2015) like reciprocity, closure, centralization, or homophily have the same magnitude of influence across a multiplex of relations used in previous studies (Doerfel & Taylor, 2017). Specifically, drawing from Shumate and Contractor (2013), we consider a flow relationship (i.e., information exchange) and three affinity relations (i.e., cooperation, trust, and communication value). Flow relationships describe how various resources are transmitted from node to node and mesh well with interactional models of communication that emphasize exchange and retrieval. Affinity relationships describe socially constructed states between nodes and mesh more with more meaning-making models of communication that emphasize social construction. Finally, we use recently developed analytical tools (i.e., VERGM) for analyzing such valued multiplex relations.
Tie Strength in Interorganizational Relationships
Tie strength refers to the general intensity and intimacy of a network relationship (Granovetter, 1973). Hence, the common tenet: weak ties are good if you need help moving a couch, strong ties are good if you need help moving a body (Pilny & Atouba, 2018). Yet, much of the theorizing on communication network structures assumes that a network tie is either present or absent; thus, either downplaying or ignoring the concept of tie strength. If researchers only see relationships between organizations as a binary, it limits the mechanisms that can be theorized to influence communication networks because the idea of tie strength is null.
For instance, sometimes characterizing complex relationships between organizations as binary is difficult, if not impossible. Take an important concept in interorganizational communication, and one highly relevant to civil society networks: cooperation. Civil society can be understood as the “evolution of cooperation and trust” among citizens and organizations (Hadenius & Uggla, 1996, p. 1622). Key communicative processes within interorganizational cooperation, like defining problems and generating solutions (Poole, 2007) are incredibly hard to define in binary terms. For instance, what does it mean to say two organizations either cooperated or did not when we know that cooperation is often thought of in terms of levels and degrees and even in a dialectical tension with competition (Doerfel & Taylor, 2004). Indeed, in more traditional methods of analysis like ordinary least squares regression, concepts like interorganizational cooperation, collaboration, trust, etc. are almost always measured with Likert-scales that capture tie strength (e.g., Atouba, 2019). When organizations communicate with one another to do civil society work, all of these components may be present to some degree but are difficult to grasp if researchers are operating under an “either or” logic.
Likewise, consider how theoretical mechanisms that invoke tie strength can go unnoticed given methodological limitations of viewing networks as binary. Our imagination about self-organizing structures can be limited. For instance, Easley and Kleinberg (2010) theorize about a structure called the strong triadic closure property, which states that if A has a strong tie with B and C, at least a weak tie is more likely to exist between B and C. If the ties are measured as binary, it is reasonable to assume that many weak ties will be absent, leading the researcher characterize the network as full of structural holes when it may actually be more clustered via the strong triadic closure property. Valued networks capture quantity and tie strength in a more nuanced way. We explore this by considering endogenous and exogenous factors.
Endogenous Factors in Valued Interorganizational Relationships
Endogenous factors are “characteristics of the relations within the network that are themselves used to explain the structural tendencies of that relation” (Monge & Contractor, 2003, p. 55). In binary networks, this means the presence or absence of a communication tie in one area has consequences for the presence or absence of a tie in another. Atouba and Shumate (2010) wrote how relational patterns like reciprocity, transitivity, and centralization are interdependent on entities and relationships throughout civil society networks. To extend their work, we discuss each theoretical explanation and its relevance below, emphasizing how the meaning of these mechanisms are amended through valued logic, not binary logic.
Reciprocity
Starting at the dyadic level, researchers have turned to exchange and dependency theories to explain the most basic network interdependency—reciprocity (Blau, 1964; Homans, 1958, 1974; Monge & Contractor, 2003). These theories have more explanatory power when there is some sort of resource flowing through the network like information (Shumate & Contractor, 2013). In this case, reciprocity focuses on network members’ abilities to offer information to others in exchange for their information (Atouba & Shumate, 2010). For civil society organizations, reciprocal flow relations offer a more efficient means for exchanging needed information rather than having multiple ties to access information from many sources.
With binary and directed data, reciprocity is measured by merely assessing the symmetry of relationships. Valued and directed data, on the other hand, assesses the similarity of the relationship value between two entities. Rather than a direct symmetry effect, there is a “pulling up” effect as described by Krivitsky (2012). For instance, if two entities rate how much they trust the other on a seven-point scale, valued reciprocity measures how similar their scores are to each other. While Atouba and Shumate (2010) relied on binary data, they suggested that reciprocity “increases the depth of relations between actors and may act as a prerequisite to more relational orientation” (p. 296). Likewise, Gulati and Stych’s (2007) posited that the relational orientation associated with dependency may increase trust and information exchange in a dyad.
Understanding the depth of relations among civil society organizations is crucial. Relationships in civil society networks can face a number of inhibitors like members’ diverse organizational interests in an issue, varying levels of involvement in the issue, and geographic distance to name a few. Yet, previous studies have mostly found reciprocity to be a positive predictor of ties among organizations (see Table 1). Building from these prior studies but adapting for the strength of ties, we posit that reciprocity will be associated with stronger relationships of multiple types. That is, when two organizations have a reciprocal tie, we predict the stronger ties reported from organization B to organization A will have similar strength as the tie organization A will report to organization B. We specifically study cooperation, information exchange, trust, and communication importance ties among civil society actors given that each has been used in prior studies and also have theoretical variety in the form of flow relations and affinity relations (Shumate & Contractor, 2013). Based on prior works, be hypothesize that:
Closure: Transitive and cyclical
Network closure is the tendency for organizations to form relationships with their own partner’s partners, resulting in “closed” or “cliquey” local configurations. When networks are directed, there can be a variety of different closure mechanisms present depending on the patterns of in-coming, out-going, and reciprocal ties. Perhaps two of the most prevalent forms of closure are (1) transitive closure and (2) cyclical closure. Transitivity occurs when an organization’s partner’s outgoing partner is more likely to be connected with that focal organization: (A→ B is more likely if A→C and C→B). Cyclical closure considers a more recursive version that looks like the common “recycling” symbol: (B→ A is more likely if A→C and C→B). Pallotti et al. (2013) discuss some of the theoretical motivations behind transitive and cyclical closure, which are inherently tied to network type. For instance, in flow relationships transitivity may be more linked to rational motivations like cost-benefit risks, reducing uncertainty, and establishing “rollover” trust (e.g., Uzzi, 1997). On the other hand, transitivity in more socially constructed networks like affinity and representational networks, transitivity is more commonly linked to Coleman’s (1990) ideas of establishing common norms, reputation, and solidarity. Cyclical closure has received less theoretical treatment but is often linked with what is called generalized exchange, meaning that the focal organization will eventually receive a link to their initial partner. The logic resembles a “pay it forward” or “belief in karma faith” because the ties must cycle all around the triad.
Both versions of closure are not immune to be influenced by values in ties. Indeed, in Simmel’s (1950) original treatise on triadic closure, he defined relationships in valued term of triviality: “Triviality connotes a certain measure of frequency, of the consciousness that a content of life is repeated, while the value of this content depends on its very opposite—a certain measure of rarity" (p. 125, emphasis our own). The power of a third actor was also defined by valued terminology, what Simmel called a Continuum of Discretion, ranging from full power to no power. Other explanations of closure stem from balance theory (Heider, 1958), which suggests that humans have a tendency toward cognitive consistency and goes back to Georg Simmel’s work on groups in the early 20th century. Heider (1958, pp. 201–202) was actually quite explicit about the type of network ties most prevalent for balance theory. They can either be (1) cognitive in nature and have a valence associated with it (e.g., like/dislike) or (2) semantic, when the nodes are linked by being “perceived as belonging together in a specially close way” (p. 201). As Heider noted, the ties must have some sort of affective logic to them.
In this sense, closure can emerge in such communication networks because of the psychological discomfort and tension generated through imbalanced triads, sometimes referred as the forbidden triad (i.e., where two organizations have a common relationship with a third, but not each other). For instance, Lee and Monge (2011) found that civil society organizations that shared multiple common third parties across types of relationships were more likely to have dyadic relations. Likewise, Kim et al. (2017) posited and found some preliminary evidence that transitivity helps organizations “ensure the reliability of information exchanged among actors and builds trust” (p. 152).
Like reciprocal relations, network closure in a civil society network may assuage some relational inhibitors. Yet, closure as a mechanism for tie formation among organizations may also be influenced by the magnitude of relationships with that common third partner: the odds of a tie between A and B may be influenced by tie strength between A and B’s relationship with the third partner C. Extending the logic of embedded transitivity to this study, and building on past civil society research that has examined closure (i.e., Atouba & Shumate, 2010), we reason that civil society organizations will have strong relationships when they share many of the same strong transitive and cyclical partners across relationships with others.
Centralization
Network centralization refers to a global or network-level metric (rather than dyadic or triadic level) because it attempts to see how concentrated and unevenly distributed ties are within a entire network. Centralization has been prominent in the collective action and public goods literature as a way to create a structure to mobilize when the tie of interest represents contributions toward a public good (e.g., Marwell et al., 1988). Crossley and Ibrahim (2012) describe network centralization as an effective way to ensure coordination, mobilize resources, and reduce costs to participate because a small number of actors are the ones acting as leaders. In this sense, networks that are too decentralized may lack a sense of leadership and may fall victim to what Freeman (1972) described as the tyranny of structurelessness.
One of the most cited processes that leads to network centralization is called the Matthew effect (Merton, 1968), later rebranded as preferential attachment (Barabási & Albert, 1999). The idea here is a “rich get richer” effect where when new nodes enter any given network, they are likely to attach themselves to already popular nodes with high link counts themselves. In civil society networks, this would suggest that central organizations are likely to continue to receive new links (i.e., indegree counts) at an exponential rate as the network continues to evolve. Yet, as Shumate et al. (2013, p. 109) noted, the odds of a centralized network emerging largely depends on the cost associated with the type of tie. For example, in online representational networks like hyperlinks or retweets, the receiver does not really infer any additional costs (e.g., resources it takes to process a tie) when receiving more and more of these types of ties. On the other hand, the resources required to maintain flow (e.g., email inbox) and affinity ties (e.g., cooperative activities) may be greater, resulting in a less likelihood of observing a centralized network.
In ERGMs, a simple way to acquire evidence of network centralization is to look at what are called k-stars, which describe the propensity of configurations with hubs that have lots of connections with other nodes who are not connected to each other (i.e., the network looks more like a star rather than a flock). In other words, stronger ties should be more likely with highly central nodes rather than nodes with low centrality:
Exogenous Factors on Interorganizational Relationships
Relationship formation can also be influenced by attributes of the organizations or the environment within which a relationship exists. Exogenous factors shift the focus from the relational attributes (i.e., endogenous) to the attributes of nodes or, in the case of this study, organizations. One of the most common categories of exogenous factors is homophily (Lazarsfeld & Merton, 1954). Homophily is the tendency of social actors (individuals and organizations) to have connections with similar others at a higher rate than dissimilar others (Bryne, 1971). Following previous scholars’ considerations of different types of homophily in civil society organizations (e.g., Atouba & Shumate, 2015), we focus on attribute-based homophily like organizational type and geography-based like the global region or economic development zones organizations are headquartered.
Organizational type
Organizational type refers to the categorization of network members. For instance, organizations in a civil society network may include but are not limited to NGOs, iNGOs, government agencies, or private firms. The logic of homophily suggests that organizations of the same type typically share “similar foci, goals, and types of operations, offering the potential for commensalism cooperative relations” (Shumate et al., 2005, p. 491). Homophily among civil society organizations is believed to make collaboration easier, reduce risks of failures, facilitate interactions, and manage conflict within networks (Atouba & Shumate, 2015). Following previous studies, we surmise that:
Geographic and economic attributes
Atouba and Shumate (2015) wrote that the operating mechanism of geographic homophily is the “operation in and identification with common space, polity, environment” that “creates familiarity and common interests due to confronting similar issues and dealing with similar stakeholders” (p. 589). Studies have typically considered the geographic homophily by looking at the connections among those in the Global North and the Global South. For instance, Atouba and Shumate (2015) found that “being located in the same global hemisphere” increased the likelihood of being partners by 172%” (p. 599). We expect to find similar results when considering the magnitude of their relations.
To extend the literature on homophily, we offer a more nuanced way of looking at the geographic and economic divisions among organizations. Provided that the logic of geographic homophily is that actors who operate in and identify with others who share a common space or environment will face similar issues and/or similar stakeholders, we expect to find the same results when looking at the economic zones an organization is situated within. For instance, the International Monetary Fund has categorized countries into one of three economic zones: advanced, developing, or low-income developing. Based on prior research, we believe that civil society actors from the same economic zones will exhibit similar magnitudes of relationship strength because of their similar experiences and types of stakeholders. However, absent of any known studies that has studied this, we pose the following research question:
Method
The primary data used in this study were gathered through a network analysis survey of members of the Sustainable Sanitation Alliance (SuSanA). SuSanA is an international network of NGOs (local and international), private firms, research institutions, and government entities that formed in 2007 to address the stalled progress toward sanitation improvements in the 1990 Millennium Development Goals (MDGs). SuSanA’s mission, as listed on their website, “is to contribute to the achievement of the MDGs by promoting sanitation systems which take into consideration all aspects of sustainability.” Its founding also coincided with the United Nations’ decision in 2006 to designate 2008 as the International Year of Sanitation. This focused political and media attention on sanitation needs led the founding partners to form SuSanA.
Procedures and Participants
SuSanA had 217 partners at data collection. Some partners were inactive or passively a part of the alliance. A refined roster of active partners was needed. Active partners were defined following SuSanA Secretariat’s guidelines for membership: current contact person, working website, and SuSanA logo on website. The Secretariat reviewed and confirmed the final list.
Representatives who were listed as an organization’s contact for SuSanA were sent an invitation to the social network analysis survey. Representatives were asked to complete the survey if they were the person most familiar with their organization’s interactions with other SuSanA partners. 2 In the cases that the representative was not the appropriate respondent, an email invitation was sent to the person who was most familiar with the organization’s relations to other SuSanA partners. Representatives from 107 of the 137 active partners completed a usable portion of the network survey. The survey response rate was 78%, exceeding minimum recommended response rates for network research (Doerfel & Taylor, 2004).
Measures
To begin, the survey asked respondents to identify who their organization had worked with in the past year from a roster of all the SuSanA partners. On average, respondents indicated their organizations had 15 partners. Respondents’ selection from this roster filtered the remaining questions to reduce respondent burden (Borgatti et al., 2013). Following previous network surveys (e.g., Saffer, 2019), respondents were then asked questions to measure communication value, cooperation, information exchange, and trust. These relations represent flow relations (i.e., information exchange) and affinity relations (i.e., trust, communication value, cooperation).
Communication value
To measure the value partners placed on their communication with others, respondents rated the value of their organization’s communication relationship with each organization selected from the interaction roster (Doerfel & Taylor, 2004, 2017). The scale ranged from zero (not at all important) to ten (very important) (M = 5.86, SD = 3.04).
Cooperation
A 14-item scale adapted from Doerfel and Taylor (2004, 2017) was used to measure cooperation (α = .91, M = 5.38, SD = 1.25). Respondents rated other organizations on a seven-point scale for items such as: helping their organization “accomplish our goals,” “collaborates with my organization,” and the reverse-coded item “this organization is deceptive.”
Information exchange
Information exchange was measured with four items and assessed the quality, aptness, and rate of information exchanged among SuSanA partners. Information exchange items were adapted from Haythornthwaite’s (1996) scale. Again, a seven-point Likert scale was used. The internal consistency of the measure for information exchange (α = .90, M = 5.27, SD = 1.28) met acceptable levels of reliability.
Trust
The final relational indicator measure was interorganizational trust. Respondents rated other partners’ trust behaviors on an eight-item trust scale adapted from Zaheer et al. (1998). They defined interorganizational trust as “the expectation that an actor (1) can be relied on to fulfill obligations, (2) will behave in a predictable manner, and (3) will act and negotiate fairly when the possibility of opportunism is present” (p. 143). The scale met acceptable levels of internal consistency (α = .86, M = 4.98, SD = .42).
Endogenous Factors: Relational Attributes
The network structures reciprocity, closure and centralization were examined (see graphics in Table 2). All of these effects use the presence of some ties to predict the odds of another tie (i.e., which is why they are also referred to endogenous mechanisms). Reciprocity is the extent to which a relationship between two entities are symmetrical (Monge & Contractor, 2003). With binary and directed network data, reciprocity measures whether the two entities reciprocate and “match” their relationships to each other (e.g., if A claims they are friends with B, B will claim they are also friends with A). As such, the presence of organization A having ties to B is used to predict if B will have a tie to A. A positive estimate in a binary ERGM model would indicate that reciprocity is, in general, more likely than chance alone.
VERGMs Predicting Information Exchange, Communication Value, Interorganizational Trust and Organizational Cooperation.
Note. VERGM = valued exponential random graph modeling; Info exchange = information exchange; Comm value = communication value; Interorg trust = interorganizational trust; Org cooperation = organizational cooperation. Numbers reported are standardized z-values for easier comparison across networks.
Organizational type was operationalized in two ways: core group membership and organization classification into seven categories. The former is a nontraditional operationalization while the latter follows traditional operationalization of organizational type.
p < .01. **p < .05.
However, with valued relational data, reciprocity can be assessed in several different ways (Krivitsky, 2012, pp. 18–22). A commonly used and easily interpretable reciprocity effect is to use the “minimum” value between two nodes (see also Squartini et al., 2013) as predictive information. For instance, if A has a value of 2 to B and B has a value of 7 to A, the minimum value would be 2. In a VERGM, the overall reciprocity effect would decrease if many cases fall into the said example, where ties from A to B are less than from B to A. (i.e., A does not reciprocate as much as B in this dyad). However, consider if A had a value of 6 rather than 2. Here, the value of 6 which would be the new minimum value. In this case, a positive effect on reciprocity would indicate that B “pulls up” (Krivitsky, 2012, p. 20) node A to its level.
Closure
There are several ways to operationalize closure patterns. We used transitivity (A > C, C > B, A > B) and cyclicity (A > C, C > B, B > A). A relatively straightforward strategy is a similar “minimum” formation, instead of paying attention to common ties to a third node (e.g., node C above). Like reciprocity, the “minimum” setting simply uses the minimum value between A to C and from C to B to predict ties from A to B for transitivity. For instance, if A has a value of 7 to C and C has a value of 8 to B, a positive estimate would indicate a similar “pulling up” effect, predicting high values from A to B (e.g., 7) that resemble the observed valued network. Cyclical closure would be similar except that ties from B to A are being predicted.
Centralization
In binary ERGM, measures of centralization have evolved over the years (Hunter, 2007). The most stable configurations to test rich-get-richer mechanisms is to use different forms of k-stars. For valued networks, one similar mathematical solution proposed by Krivitsky (2012) was to implement a statistic that models tie heterogeneity. That is, the tendency of some nodes to send and receive more links than others. The opposite would be tie homogeneity, meaning that the degree distribution is relatively equal (i.e., decentralization). This was done by including a statistic accounting for “positive within-actor correlation among the dyad values” (p. 22), which should increase the more centralized the network becomes.
Exogenous Factors: Organizational Attributes
Organizational type
We operationalized organizational type in two mutually exclusive ways. First, we considered organizational type as the role a partner takes within the network. For this, we used SuSanA partners’ membership in the alliance’s formal “core group.” The SuSanA core group was made up of nine partners who provided strategic direction and advice, planned meetings and events, proposed strategies, and made operational decisions. Second, organizations were categorized into one of seven types: local NGO (n = 28), iNGO (n = 28), private sector (n = 21), education/research (n = 15), government/state-owned organization (n = 7), multilateral organization (n = 3), or network (n = 5). Organizations self-selected from these types when they joined SuSanA, and the secretariat confirmed their classification.
North/South divide
SuSanA partners are located in Africa, Asia, Europe, North America, and South America. Using the organization’s headquarters was classified as being either in the Global North (n = 94) or Global South (n = 12).
Economic zone
Similar to the previous section, all partners were classified as being either advanced (n = 59), developing (n = 28), or low-income developing (n = 19) economic zones. The economic zone classification was based on a partner’s headquarters location and the economic zones were defined by the International Monetary Fund (2014).
Analysis
To model the network, we used VERGM to examine which processes may be responsible for why the networks look the way they do. For a more technical introduction to VERGM, we refer the reader to Krivitsky (2012) and Krivitsky and Butts (2013) but will provide a more digestible breakdown below. Before explaining VERGM, it is important to understand the original binary ERGM. We will use regression as a comparison as it is a common method used in communication research. Each parameter entered into the ERGM is like an independent variable in a logistic regression model. They offer competing explanations for the dependent variable: the odds of a relationship existing between any two organizations. For instance, reciprocity would predict that a relationship exists from organization A to B because a relationship exists from B to A. Preferential attachment would predict that a relationship exists from organization A to B because B has lots of relationships with other organizations.
Under the hood of an ERGM is a bit more complicated than logistic regression, which typically relies on maximum likelihood estimation (MLE). MLE, as it is used in logistic regressions, is simply impossible in ERGM because of the large sample space inherent in social networks: the amount of possible ways a network can be configured. 3 Instead, ERGM uses a Markov Chain Monte Carlo (MCMC) procedure to simulate random networks used to approximate MLE values (Geyer & Thompson, 1992).
Results
Table 2 presents the results of the multiplex VERGMs with endogenous and exogenous parameters. The goodness of fit statistics are reported in Table 3. The endogenous factors included sum, reciprocity, transitive closure, cyclical closure, and centralization. Rather than reporting the MLEs, we opt for the standardized z-values to make comparison across the four networks easier. For instance, the sum parameter, a measure of network density, calculates the sum values from the relationships in the observed network (Scott, 2016). It essentially indicates if the odds of reporting high tie strength is rare, somewhat random, or quite likely. Across the type of relations reported in Table 2, the significant and negative sum parameters vary depending on network type. Organizational cooperation (z = −2.16, p < .01) and interorganizational trust (z = −21.78, p < .01) were the strongest of the negative sum parameters, which suggests that organizations were quite selective, especially with trust. Information exchange had no significant baseline (z = −.01, p = .98) and communication value (z = 11.87, p < .01) was more prevalent among organizations.
Goodness of Fit Statistics.
Note. Obs. = Observed value in the data; Sim. = Average value from 10,000 simulations. p = significance values. Non-statistically significant values indicate there is no difference between observed and simulated networks, indicating a good fit. The higher the p-value, the better.
The mutual parameter, also referred to as reciprocity, examines whether the observed reciprocated ties are higher in their edge value than expected (Woldense, 2018). The positive and significant coefficients across the four types of relationships lends support to H1: information exchange (z = 7.19, p < .01), communication value (z = 10.89, p < .01), interorganizational trust (z = 9.81, p < .01), and organizational cooperation (z = 11.22, p < .01). As hypothesized, we found that network members with mutual ties exhibited similar magnitude in their relationships. More specifically, the strength of relations in terms of information exchange, communication value, interorganizational trust, and cooperation increased when two partners had reciprocal ties and was fairly similar in magnitude across each of the four networks.
The second hypothesis directed attention to transitive and cyclical closure. Similar to reciprocity, the general pattern was consistent across each of the four networks: when valued ties became similar to a common third party, transitive closure was more likely and cyclical closure was less likely. This was apparent with positive transitive estimates on information exchange (z = 16.14, p < .01), communication value (z = 118.05, p < .01), interorganizational trust (z = 17.40, p < .01), and cooperation (z = 16.88, p < .01) and negative cyclical estimates on information exchange (z = −4.89, p < .01), communication value (z = −11.98, p < .01), interorganizational trust (z = −6.04, p < .01), and cooperation (z = −6.12, p < .01).
The final hypothesis concerning the endogenous factors suggested that centralization (i.e., a relatively small number of highly social organizations) would have stronger magnitudes of relationship strength with others. However, across the four relations, we found no support for the hypothesis. Instead, our data indicate that the prominence of an organization’s degree centrality had the opposite effect; the magnitude of relationships was weakened as network members’ degree centrality increased. Whether for information exchange (z = −36.15, p < .01), communication value (z = −69.63, p < .01), trust (z = −15.76, p < .01), or cooperation (z = −38.29, p < .01), organizations were likely to have strong relationships with others that were not very central. That is, there was not tie heterogeneity, but instead tie homogeneity, meaning that the network was rather decentralized and equally distributed.
Next, we turn to the exogenous factors to examine the impact of homophily on the magnitude of organizations’ relationships. Here we have for parameters: core group membership, organizational type, location of headquarters in the Global North or Global South as well as their location in the IMF’s economic development zones.
H4 asserted that organizations of the same type would exhibit stronger magnitudes in their relationships with others. We operationalized organizational type in two ways. First, we assumed that organizations that were members of the SuSanA “core group” would likely have similar foci and goals for the network. However, we found that for core group members the magnitude of information exchange (z = −29.18, p < .01), communication value (z = −23.73, p < .01), interorganizational trust (z = −10.75, p < .01), and cooperation (z = −21.75, p < .01) were not strengthened by having the same core group status (both core group or both non-core group). Rather it was the opposite: heterophily. Core group members had stronger ties to non-core groups than those of ties within groups (i.e., core to core).
The second way we operationalized organizational type, which is more in line with previous studies, produced mixed results. With this approach, organizations were categorized into one of seven types to indicate the industry or resource space they operate in. The results revealed that organizations with relationships to other organizations in the same category had a modest increase in the magnitude of the interorganizational trust (z = 2.63, p < .01), organizational cooperation (z = 2.34, p < .05) and communication value (z = 3.82, p < .01). However, organizations were neither more nor less likely to share stronger levels of information with similar organizational types (z = .02, p = .86). Nevertheless, H4 was mostly supported.
Last, H5 posited that organizations headquartered in the same global hemisphere would have similar magnitudes of relationship strength. The results indicate that the magnitudes of information exchange (z = 5.78, p < .01), communication value (z = 5.96, p < .01), trust (z = 4.49, p < .01), and organizational cooperation (z = 6.59, p < .01) among organizations were significantly and positively strengthened by the geographic homophily. Still, we sought to further investigate geographic homophily with more a precise measure that moves toward a local measure of geographic homophily rather than the binary North/South divide.
The logic of geographic homophily posits that organizations from the same hemisphere are more likely to operate in and identify with others who share a common resource space and have similar issues and/or stakeholders. RQ1 sought to extend this logic with a more precise measure of geographic homophily and used the IMF’s three economic development zones. Specifically, organizations with relationships to others in the same economic zones indicated having higher levels of information exchange (z = 3.74, p < .01), communication value (z = 4.06, p < .01), interorganizational trust (z = 4.98, p < .01), and cooperation (z = 2.44, p < .01) with others. To address RQ1, our results indicate that network members from the same economic zones tend to exhibit similar magnitudes of relationship strength but to a somewhat lesser extent than the binary parameters.
Discussion
The purpose of this study was to expand upon the relational view of civil society by considering the role tie strength and multiplexity among organizations. We did this by reexamining prior studies’ assertions about what influences interorganizational relationship formation in civil society by considering the strength of ties across multiple relations. Our results offer important insights into the consequences of closure mechanisms, the applicability of preferential attachment to real-world networks, and the nuances of homophily on multiple types of relationships in a communication network.
Multidimensionality and Strength Matter, But How So?
Communication researchers have long conceptualized relationships as multidimensional and recognized that relationships among communicators can vary in strength (Contractor et al., 2011; Monge & Contractor, 2003). While recent work indicates that communication research is making such advances, researchers have been challenged in two ways from addressing tie strength and multidimensional relations. First, researchers have been challenged with the assumption that relationships are either present or absent. When hypotheses are based on this questionable binary, researchers are forced to interpret results in a relatively simplistic manner. Indeed, scholars have long theorized about relationships in terms of their strength. Until recently, the availability of appropriate analytical frameworks limited researchers to use binary data. In a binary form, the relational data represent the presence or absence of ties instead of tie strength.
Second, studies have tended to examine single types of relations; thus, researchers have been dependent on the type of relationship examined. Scholars have asserted that civil society interorganizational relationships are inherently multidimensional (Doerfel & Taylor, 2004), identified the patterns and determinants of multiplex relations (Lee & Monge, 2011), and more recent work reveals how multiplexity influences cooperation among civil society organizations (Liu et al., 2019). Yet, these have been based on an either or logic that limits the true nature of network ties. Our results demonstrate that multidimensionality and strength matter.
Before discussing the hypotheses, consider the sum parameter, which models network density (i.e., are they sparse and selective or thick and dense?). Information exchange, the only flow relation, was neither sparse nor dense, just average. This differs from the three affinity networks, which were either quite selective (trust and cooperation) or dense (communicative value). This suggests that although there are differences in density between communication network types, there are also differences within each type. Indeed, there are a lot of different ways to characterize socially constructed affinity relations, such constructions matter.
Take for instance the results of H1. Prior works, rooted in resource- and exchange- theories, have asserted that reciprocity between two civil society groups allows them to offer information and resources in exchange for other information and resources (Atouba & Shumate, 2010). This has led researchers to suggest that reciprocity offers organizations an efficient way to access and exchange needed information and resources rather than having multiple relationships to access resources and information from many sources. Yet, our results found that organizations had stronger relationships in terms of trust, cooperation, and communication value when the relationships were reciprocal (i.e., a “pulling up” effect). Indeed, information exchange was significant but had the weakest in terms of magnitude, which may suggest that in affinity-like relationships, reciprocity may matter a bit more.
The results also investigated two closure mechanisms: transitivity and cyclicity. Despite some different variations in magnitude, the story was quite similar: organizations tended to engage in transitive closure and avoid cyclical closure, regardless if the tie represented a flow or affinity type of relationship. For transitivity, this was not surprising as previous research has found it to operate at various types of interorganizational ties. For instance, Kim et al. (2017) reasoned that the multiple transitive ties in a network of emergency organizations contributed to the “flow of timely and reliable information” (p. 154). Likewise, for more affinity-like ties like cooperation, Atouba and Shumate (2010) found similar results and posited that transitivity could reduce the risk and cost of a civil society organization establishing a new relationship with an unfamiliar other by giving it “cues about potential partners from their current partners” (p. 297).
However, why was cyclical closure less likely to appear across all of the ties? Cyclical closure is commonly interpreted as generalized exchange stemming from the foundational work on social exchange by Ekeh (1974). Specific to interorganizational networks, Pallotti et al. (2013) link cyclical closure with diffusion and delay, writing that cyclicity implies that “participants have to be ‘givers’ before they may become ‘receivers’” (p. 205). If patience and “good-faith” is required for cyclical triads to develop, it is not too surprising that the model produced negative estimates. Indeed, some previous interorganizational research have found either negative or nonsignificant estimates on cyclical closure, from flow ties like patient transfers (Pallotti et al., 2013) to affinity ties like partnerships (Laven et al., 2010). On the other hand, Fu and Shumate (2017) report a positive estimate of cyclical closure on representational networks like hyperlinks, which may make sense because the timing of a tie is more ambiguous (i.e., organizations often are not aware of when somebody hyperlinks to them). Nevertheless, our results, compared to the context of previous research, suggests that more theorizing is needed to explicate generalized exchange mechanisms operating in cyclical closure and under what conditions (e.g., network type) are we more or less likely to observe it.
Questioning the Applicability of Preferential Attachment in Interorganizational Networks
Since the boom of network science in the 1990s, much has been made about preferential attachment and centralization being a primary mechanism influencing networks (Barabási & Albert, 1999). However, how well does preferential attachment explain real-life networks like interorganizational communication? Recent research is seeming to question this mechanism. For instance, Broido and Clauset (2019) analyzed 928 different types of networks and found that preferential attachment (i.e., a scale-free distribution) rarely explained why the network looked the way it did; this was especially true for social networks involving humans. Our results go further to suggest the opposite: decentralization may actually be playing a role.
Why do we see decentralization instead of centralization then? It is our contention that there are two reasons. The first is a simple statistical consequence of dichotomizing networks. On the other hand, the second is more theoretical and conceptual. Each is reviewed below.
First, when communication ties are dichotomized into binary form, information about the ties, especially ones with lower than average values are simply lost. However, just because nodes have lower than average values, does not make them meaningless. And although there are best practices for and very good reasons to dichotomize ties (Borgatti & Quintane, 2018), there has been almost no research done considering what dichotomization does to overall network degree distribution. As a simple test, consider a normal degree distribution of 10 nodes on a typical 4-point Likert scale: (1, 2, 2, 2, 2, 3, 3, 3, 3, 4). The Gini coefficient is a modest .19 (higher Gini coefficients indicate higher skewed inequality). If we dichotomize this degree distribution to only keep the 3’s and 4’s, the Gini coefficient jumps to .56, indicating much higher centralization. This phenomenon was also reported by Pilny and Atouba (2018), who found that when dichotomizing a collaborative scientific network, decentralization was severely underestimated. Indeed, although largely defending dichotomization techniques, Borgatti and Quintane (2018) do admit that “successive dichotomization confirms a core/periphery structure” (p. 3), which would point toward a more centralized network topology. The key takeaway is that when only looking at a binary network, researchers may be underestimating levels of decentralization. As Thomas and Blitzstein (2011) titled their study on similar detrimental effects of dichotomization: valued ties tell fewer lies.
The second reason is more theoretical and may be explained by something as simple as communication overload (see Stephens et al., 2018). Both flow networks like information exchange and affinity networks like trust take work. Though the magnitude differed a bit, with communication value being the most decentralized and trust being less so, all four networks had significant estimates pointing toward decentralization. For centralized nodes with these types of ties, there is a cost of effort to sustain and maintain these types of network ties. Organizations, like people, tend to have what Monge and Contractor (2003) call a relational carrying capacity, meaning there is a certain threshold of how many network ties actors can hold at one time before overload takes place. Of course, things like organizational capacity, the amount of resources an organization has at its disposal (Shumate et al., 2018), can influence this: the more capacity, the more likely an organization can maintain such ties. However, other types of ties may not depend on organizational capacity at all. Take representational ties like hyperlinks. Such connections are less affected by relational carrying capacity because there is no real direct cost associated with garnering more ties like this. In fact, most websites are completely unaware of how many other websites are linking to them. They simply are different from regularly exchanging information or co-constructing meaning together, which may be influenced by an upper bound limit.
Birds of a Feather May Flock Together, But Which Feather Decides?
The adage of homophily is that birds of a feather flock together, but our results lead us to ask which feature decides? Using two different attributes for each type of homophily (organization type and geography), the results of this study suggest there are contingencies of birds flocking together. To understand exactly why some feathers might be more important than others, it is important to step back and understand some of the general mechanisms working behind theories of homophily. For example, Monge and Contractor (2003) point to two explanations: (1) Byrne’s (1971) similarity-attraction hypothesis and (2) Turner et al. (1987) theory of self-categorization. The similarity-attraction hypothesis posits that homophily emerges because such interactions are more predictable and decreases the potential stress or psychological discomfort associated with encountering diversity or experiencing cognitive and emotional inconsistency. As such, this could also make interorganizational cooperation less complex, less risky, and less likely to create conflict (Atouba & Shumate, 2015). The theory of self-categorization, on the other hand, suggests that actors define their social identity through a cognitive process of self-categorization during which they classify themselves and others into various categories (e.g., age, gender, profession, ethnicity, race, etc.), and that they use these categories to further differentiate or discriminate between similar and dissimilar others (i.e., create the in-groups and out-groups that help solidify a particular social identity).
For the current results, both types of attribute-based homophily may be true and it is hard to disentangle exactly which mechanism is more likely. Take for example the results for the homophily of organization type (core group and organizational type). We found that organizations that were part of the core group were significantly less likely to report stronger relationships with other core group members. But for organizational type (i.e., seven different sector types), evidence of homophily was positive for trust, cooperation relationships, and communication value, the three affinity relationships. Organizational type was not significant for the sole flow relationship: information exchange. The implication here may be that civil society organizations socially construct relationships (but not share more information) with others in the same sector (e.g., local NGO, international NGO, private sector, etc.). Moreover, organizations were less likely to have any relationships with similar core groups. Although this may seem like speculation, the implication for future civil society research is clear: researchers should strive to measure attribute-based homophily in terms of potentially relevant both (a) attributes that may make interorganizational communication less complex and risky, and (b) may trigger in-group and out-group identities that may lend themselves to self-categorization processes.
Finally, geographic homophily is another important factor on relationship formation among civil society groups because operating and identifying with similar spaces, polities, and environments can create “familiarity and common interests due to confronting similar issues and dealing with similar stakeholders” (Atouba & Shumate, 2015, p. 589). Our results support hypotheses regarding geographical-homophily because every MLE coefficient was significant and positive regardless if the relationship represented a flow or affinity tie. This was true even when geographic locations were measured by global headquarters and (b) economic development zones. As a result, even in times of increasing use of information and communication technologies, we would urge researchers to resist being tempted to ignore the role geography and proximity play in explaining civil society communication networks.
Limitations
This study makes important conceptual and analytical advances for interorganizational civil society network research in communication, but there are some limitations. First, surveys come with their own limitations. Efforts were made to solicit responses from individuals who were most familiar with their organization’s relationships to others in SuSanA. The results of this study are based on those individuals’ perceptions of their organization’s connections. Second, while prior research has established that communication networks matter, we believe the literature now needs to know how networks matter. Future research should consider exploring the role of organization’s agency in relation to the endogenous and exogenous factors studied here to better understand how organizations use their relationships to achieve individual and collective goals. Finally, and somewhat related to the previous point, this study is a cross-sectional view of one network. While our results challenge some of the assumptions in previous studies, we recognize that SuSanA is a unique civil society network. Based on fieldwork with this network, we know that many of the partners were focused on sustaining the network when the data were collected. Because of this, partners may have been more concerned with their affinity with other partners rather than about exchanging resources.
Conclusion
Networks of global civil society organizations are critical spaces where communicators can address pressing social and political issues, ideas, and values. Much of the literature details how these network structures can influence a collective’s efficacy or identify organizations that may have varying types of influences depending on their network position. We join those who have sought to explain why these networks take the structures they do or why certain actors have certain influences based on their position. We did so with a unique dataset and novel analytical framework that suggest the mechanisms influencing relationship formation are more complex than previously understood.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
