Interorganizational homophily and social capital network positions in Malaysian civil society

ABSTRACT The interorganizational relationship communication literature has identified homophily – the tendency for actors to form ties with similar others – as a mechanism predictive of tie formation among organizations in civil society networks. This study examined the connection between homophily and network structures equated with different types of social capital and perceptions of influence. Using survey data gathered from a network of Malaysian civil society organizations (n = 90), exponential random graph models and autologistic actor attribute models were used to test the association between homophily characteristics and the networked social capital positions of bridging, bonding, and gatekeeping. Results showed that bonders and brokers tended to be influenced by homophily, whereas gatekeepers were influenced by heterophily and homophily. Homophily was also associated with the likelihood of CSOs rating each other as more influential on government reform.

Perhaps because homophily inherently focuses on questions of similarity, past studies have not thoroughly considered the extent to which homophilous ties are, in reality, an advantage for CSOs. Indeed, in the management literature scholars are only recently beginning to make sense of the outcomes associated with creating homophilous intraand interorganizational relationships (Ertug et al., 2022). There may be organizational benefits to heterophilous as well as homophilous relationships with others (Atouba & Shumate, 2015;Doerfel & Taylor, 2017). The question of managing homophilous versus heterophilous ties is important to CSOs if they are to maintain agency and have a diversity of contacts, which are essential to problem solving (Newman & Dale, 2007). Yet, as Doerfel and Taylor (2017) found, homophilous interorganizational relationships "are one way to build the capacity of specific sub sectors of civil society" (p. 17). Whether or not homophilous tie formation is associated with beneficial outcomes for CSOs is a key question for interorganizational network research. In other words, it may be time to look past whether homophilous processes merely exist and ask whether such processes are associated with positive outcomes for CSOs.
With this study, we intended to make two contributions to advance the theory of homophily as it pertains to the study of interorganizational networks in civil society. First, we aimed to replicate previous interorganizational network research by examining the extent to which common characteristics like an organization's age, type, and issue focus predict tie formation among government reform CSOs in Malaysia following the dramatic 2018 election in which the opposition party won power for the first time in the nation's history. Although most prior homophily research has focused on geographic homophily at the level of a global region (Atouba & Shumate, 2010, 2015 we turned our focus here to a national network. Few studies have examined civil society networks at the national level (for exceptions see Doerfel & Taylor, 2004;Fu & Shumate, 2017;Taylor & Doerfel, 2011). During times of intense transition, such as after a democratic election, there is "a natural impetus for organizations to cooperate" (Doerfel & Taylor, 2004, p. 390). Thus, the post-election context in Malaysia provided a unique opportunity in which to study how homophily impacts relationships among CSOs.
Our primary contribution connected network positionality and outcomes with theories of homophily. To do so, we assessed the degree to which CSOs' homophilous ties are associated with network positions believed to be advantageous for accessing a network's social capital (Borgatti et al., 1998;Burt, 2005;Lin, 1999). We considered two classic views on how social capital network arrangements occur. One view maintains that bridging social capital results from spanning structural holes, affording an actor brokerage over information and resources. The other view holds that social capital is a function of network closure, wherein bonding, dense ties provide easy access to information and facilitate trust and group norms (Burt, 2005). We also worked to problematize the somewhat reductive bridging/bonding dichotomy by including the role of gatekeepers in social capital assessments (Gould & Fernandez, 1989). Finally, beyond network positions, we looked at how homophily influences whether or not organizations perceive each other as influential in government reform, as influence is often a key network outcome for well-positioned CSOs (Doerfel & Taylor, 2004).
The influence of homophily on tie formation in civil society networks First introduced to the network literature by Lazarsfeld and Merton (1954), the logic behind the theory of homophily is simple: "similarity breeds connections" (McPherson et al., 2001, p. 415) and "birds of a feather flock together" (p. 417). Theories of homophily argue that connections are more likely to occur between similar others than between dissimilar others. According to Monge and Contractor (2003), homophily is generally thought to reduce discomfort (i.e., similarity attraction view) and/or valorize in-group identities (i.e., self-categorization view). Scholars of interorganizational communication have turned their attention to how organizational characteristics predict tie formation within networks, often within the context of civil society.
Emerging literature in this area has suggested that homophily plays a significant role in the formation of CSO networks in civil society. Atouba and Shumate (2015) reasoned that "when organizations share certain key attributes (e.g., age, mission, interests, culture, operating systems), the similarity provides inducement and opportunities to form collaborative ties with particular partners" (p. 589). Recent research on interorganizational homophily has pointed to four factors, three of which were comprehensively explicated by Atouba and Shumate: (1) age, (2) institutional type, (3), geographic location, and (4) common issue priorities (Atouba & Shumate, 2010;2019;O'Brien et al., 2019;Saffer et al., 2021). Each is described below.
First is organizational age. Organizations with similar founding dates are thought to have a cohort effect where organizations have greater compatibility based on shared experiences, market conditions, and the perceived needs at the time of founding. Such compatibility may create perceptions that collaboration is easier, reduces risks of failure, and can serve as a springboard to initiate contact between organizations (Atouba & Shumate, 2015;Shumate et al., 2005). Second, organizations often cooperate with others of the same institutional type or sector. Similar logics of compatibility may also apply here but being of the same type may also introduce institutional forces that incentivize CSOs to conform to similar standards and adopt comparable structures (Atouba & Shumate, 2015;Doerfel & Taylor, 2017;Fu & Shumate, 2016;Shumate et al., 2005). Third, similar geographic locations and socio-economic conditions shape relations among CSOs due to similar environmental conditions and issues (Atouba & Shumate, 2015;Lai et al., 2019;Lee & Monge, 2011). Indeed, theories of proximity predict that close location increases the probability for interaction, which allows individuals and organizations to explore common interests (Monge & Contractor, 2003). Finally, similarity in issue prioritiesthe homophily characteristic that has received the least attentionhas led to tie formation in some cases (O'Brien et al., 2019;Sommerfeldt et al., 2022), as ties with similarly-oriented others help build specific sub-sectors of civil society (Doerfel & Taylor, 2004). Forging relationships with others who work on the same issue may pool resources and expertise to address the issue or may help reduce uncertainty or risks of working with others whose priorities do not align with a CSO (Sommerfeldt et al., 2022). As an illustrative aid, Online Supplement File 1 summarizes these homophily types and provides example studies. Based on the current literature, we propose a simple replication hypothesis to test these concepts in a national-level setting.
H 1 : CSOs will be more likely to have relationships with others that are similar in (a) age, (b) organizational type, (c) issue priority, or (d) region (geography).

What does homophily do for CSOs?
In the title of their recent article, Ertug and colleagues (2022) asked, "What does homophily do?" Their review primarily centered around firms and other likeminded capital-accumulating organizations, reviewing such outcomes related to performance, firm evaluation, diffusion of ideas, and organizational learning. In explaining what homophily might do for some organizations, Ertug et al. pointed to two key general mechanisms: (1) a comforting effect, and (2) a siloing effect. The comforting effect is rooted in the similarity attraction hypothesis in social psychology (Byrne et al., 1971). Applied to social networks, Monge and Contractor (2001) posited that similarity-attraction generally reduces psychological discomfort that may be lurking within heterogeneous groups seeking to form new ties. With respect to organizations, homophilous network ties can thus result in "smoother coordination, better communication, and enhanced trust between an actor and contacts" (Ertug et al.,p. 3). The siloing effect can reduce "diversity in knowledge, perspectives, and other resources that an actor can access through contacts" (Ertug et al.,p. 3). Indeed, much has been made about concepts like fragmentation and echo chambers linked to homophilous tie formations (Riles et al., 2018). In such cases, more heterophilous tiesconnections to others different than oneselfcan provide benefits in terms of exposure to new information and resources, thereby reducing echo and improving the spread of diverse ideas and promoting network dynamism (Granovetter, 1973). However, as noted by Barranco, Lozares, and Muntanyola-Saura (2019), while the literature has primarily treated homophily and heterophily as opposites, the reality is heterophily is a complement to homophily as actors are likely to possess both homophilous and heterophilous relationships, and sometimes for strategic reasons (Sommerfeldt & Yang, 2017).
For CSOs, how these mechanisms might relate to resources associated with homophilous tie formations remains understudied, primarily because most interorganizational communication research tends to focus on homophily as a predictor of communicative tie formation, as described earlier. Nevertheless, some research has shown that homophily can increase a CSOs' tactical repertoire (Wang & Soule, 2012), help find desired organizational partners (Knoben et al., 2019), and reduce uncertainty (Hulbert et al., 2000). We focus here on the outcomes that may be afforded to CSOs through homophily such as perceptions of influence as well as social capital outcomes connected to network positionality.

Homophily and organizational influence in civil society
One of the few studies that looked at a similar national civil society network was Doerfel and Taylor (2017), who have also published several pieces related to the Croatian civil society movement in the 1990s and early 2000s (e.g., Doerfel & Taylor, 2004;Taylor & Doerfel, 2003). They found that a question measuring importance (i.e., which organizations are most important in Croatian civil society?) was related to homophily regarding similar organizational sectors (i.e., democracy, human rights, media, and funders). Thus, some homophilous organizations may be more likely to perceive each other as more influential.
Including factors like the comforting and siloing effects, insights from research regarding social identity and self-categorization are useful to consider here as well (Hogg & Reid, 2006). For instance, many attributes used to measure homophily (e.g., age, type, issue priority, location) can serve as indicators of in-groups and out-groups when such identities are made salient (i.e., old versus new, activist versus reform, local versus national, etc.). A common theme in much of the social identity literature is that in-group members tend to have more positive evaluations within rather than between group identities because such connections valorize the in-group (Dragojevic & Giles, 2014). As such, we might call a third homophily mechanism the valorization effect. If this is true, then it is reasonable to extend such mechanisms to CSOs if such attributes can represent distinct social/organizational identities. For instance, if homophilous ties can engender perceptions of comfort, self-esteem, and silo alternative or even competitive perceptions, we would expect organizations that are homophilous to perceive each other as more influential. As such, we hypothesized: H 2 : CSOs will have stronger perceptions of influence with others that are similar in (a) age, (b) organizational type, (c) issue priority, or (d) state (geography).

Positionality and social capital
A central aim of interorganizational relationships research is to "explain why and how organizations connect effectively, work cooperatively, and coordinate their activities to achieve superior performance" (Nahapiet, 2008, p. 1). Theories of social capital attempt to address this, however, there are two broad understandings of social capital. One understanding conceives of social capital as a quality of groups and societies and speaks to ideals like generalized social trust and civic participation (Fukuyama, 1995;Putnam, 2000). The other conceives of social capital as the value of an actor's social relationships and their structure (Burt, 1992;Lin, 1999). From the latter perspectiveon which this study reliedsocial capital is seen as the structures actors use to achieve their interests through networks (Coleman, 1990). Social networks as a proxy to social capital, called "structural social capital," refers to the advantages that come through network positions (Borgatti & Foster, 2003).
Indeed, structural social capital is "a metaphor about advantage" (Burt, 2001, p. 31), and most definitions of the concept recognize that "social capital constitutes both the structure and some value produced by the structure" (Shen et al., 2014, p. 461). Some social actors do better, and their performance is considered, in part, to be a function of their network size and position. Network position simply refers to where a node is located within the context of the larger network structure and how such a position puts that node in advantageous and/or disadvantageous predicaments (Leonardi & Contractor, 2018). We considered two popular perspectives on local network arrangements that have gained prominence in social capital network research: bridging and bonding. While popularized by Putnam (2000), the originators of the terms, Gittell and Vidal (1998), described bridging social capital as "the type that brings together people or groups who previously did not know each other" (p. 15) and bonding social capital as "the type that brings closer together people who already know each other" (p. 15). Related parlance for bridging and bonding social capital are the network terms brokerage and closure (Shen et al., 2014). These concepts have similarly structured debate on network inquiries into the theoretically bifurcated nature of social capital and strategies for CSO tie formation (O'Brien et al., 2019). As such, there are differing views on which type of social capital arrangement provides the most benefit (Doerfel et al., 2013;O'Brien et al., 2019).
Bonding. It is worth going back to the basic premise of homophily: when two organizations have a shared characteristic, they will be more likely to forge a relationship with each other. If the logic of homophily holds true beyond a dyad, then there should be clusters of ties among organizations with similar characteristics because the attribute serves as a key attraction, bringing organizations together (i.e., a siloing effect). Evidence from network simulation models that use homophily as a key predictor of ties indeed confirms such network-based clustering (Liu et al., 2018). Such clustering parallels contemporary understandings of bonding social capital. Bonding social capital is developed through strong ties of repeated interaction, where many social actors are connected such that no one can escape the notice of another (Norbutas & Corten, 2018). Coleman (1990) similarly viewed social capital in terms of closurea deeply interconnected network that provides access to information quickly and more easily creates feelings of trust among network members (i.e., a comforting effect).
As such, it should be no surprise that homophilous ties seem to be beneficial in times of environmental turbulence (Doerfel et al., 2013;Hulbert et al., 2000). Highly embedded actors tend to form relationships with other like-minded highly embedded actors, which works to reduce environmental uncertainty and enhances organizational prestige (Ahuja et al., 2009). Bonding capital builds trust within groups and undergirds network reciprocity and solidarity (Shen et al., 2014). Atouba and Shumate (2010) found evidence of closure among development CSOs that worked to facilitate activity coordination, and O'Brien et al. (2019) concluded that CSOs will pursue network closure in early stages of CSO network development as closure enables resource exchange in local network communities (Astley, 1985). Moreover, O'Brien et al. found closure to be associated with network homophily, as evidenced by collaborations with partners in the same issue domain.
Bridging. Despite the advantages that may be derived from network closure, at some point homophily may become an obstacle for organizations. Excessive reliance on the same partners limits opportunity for action outside of the interconnected group (Ahuja et al., 2009). Thus, in contrast to the view that homophily results in bonding social capital is the argument that heterophily (the tendency to form ties with different others) and brokerage provide another form of social capital. This view requires a broadening of scope. Homophily research has focused on similarities within a dyad. For example, if organization A has a characteristic in common with organization B, the theory of homophily suggests the two have a greater likelihood of forming a tie. But that does not consider the network holistically. If we take the premise that homophily influences ties based on an attribute, it is likely that at some point there may be fragmented clusters and cliques (e.g., echo chambers) revolving around such attributes. As such, it is necessary to zoom out from the dyad to consider the larger network of relationships.
Logically, homophily can divide those that have fewer or "not the right" similarities; instead, some organizations form relationships with others who share differences. These organizations may also have advantageous network positions but not because of the bonds they form. Rather, organizations can be positioned as brokers to different parts of the network. To that point, bridging social capital is accrued through diverse contacts and weak ties (Granovetter, 1973). Burt (2005) described this view of social capital as a function of brokerage and control, relying extensively on weak ties (Granovetter, 1973) and betweenness centrality (Freeman, 1977). Weak or nonexistent connections among groups in networks allow for structural holes or gaps in the network, creating an opportunity for those who bridge holes between groups to broker the flow of information and resources (Stohl & Stohl, 2005). A node who bridges structural holes is "the third who benefits" from brokering the connections between diverse others.
H 4 : CSOs exhibiting bridging social capital are less likely to be influenced by homophily when forming ties.
Gatekeeping. Despite the differences in structural emphases, bridging and bonding perspectives on social capital are not entirely mutually exclusive. Burt (2001) proposed that although cohesive ties provide social support, it is also necessary to exploit structural holes outside dense arrangements of ties to gain access to unique and diverse information. Newman and Dale (2007) similarly noted that a mix of ties will create "a resilient blend of local support and dedication and links to external resources" (p. 82). Likewise, these positions are not static and can change over time. For instance, Burt and Merluzzi's (2016) longitudinal study of bankers shows some of them "oscillate" between brokering and bonding positions to maximize beneficial outcomes. Even more, it is possible for an organization to simultaneously play both bridging and bonding roles by occupying positions in dense cliques and linking to otherwise disconnected nodes (Murray et al., 2020). These nodes have been called gatekeepers.
As explicated by Barzilai-Nahon (2008), the concept of gatekeeping was coined by Lewin (1947) to "explain the focal points of social changes in communities" (p. 1493) and was mostly applied to the news industry (cf. DeIuliis, 2015). The concept was further positioned within social network analysis by Gould and Fernandez (1989). Here, gatekeepers had such influence because they largely set the agenda for their groups as the ones who had access to outside information from other groups. Much like the two-step flow model of communication (Katz & Lazarsfeld, 1955), actors can be "gatekeepers of a subgroup by controlling outsiders' access to members within the group" (Contractor & Forbush, 2017, p. 16, emphasis in original). As such, gatekeepers are often both embedded in dense subgroups (i.e., bonding) and contain ties across different subgroups (i.e., brokering). For Gould (1993), network brokering concepts like gatekeeping not only were important because of such information control, but also because they help explain collective action mobilization beyond psychological attributes like class consciousness. However, little research has examined the network behavior of gatekeepers (cf., Foster et al., 2011). In terms of homophily, it stands to reason that they may be influenced by both homophily and heterophily given that gatekeepers tend to be embedded in both structures (i.e., bridging gaps and in dense cliques). As noted by Barranco et al. (2018), some actors will often have relationships with others from their own groups as well as with members of other groups. Given the paucity of research on gatekeepers and homophilous/heterophilous ties, we pursued this line of inquiry as a research question: To what extent are CSOs that exhibit gatekeeping social capital influenced by homophily when forming ties?

Method
The data for this study come from a survey of Malaysian CSOs dedicated to government reform. Malaysia has endured a particularly turbulent political environment in the last few years, with a political sea change in 2018 that saw the first democratic shift in power to the opposition party in the nation's history. After decades of one-party rule, rampant corruption, and resistance to reform, the new government in 2018 signaled a welcome of civil society back to the decision-making table and talks of substantive social and political reform. Many Malaysians described the election outcome as "surreal" (Hooi, 2018). This opportunity prompted the late 2019 study of networks among Malaysian CSOs, just prior to the unfortunate collapse of the new opposition government in early 2020 and the outbreak of the COVID-19 pandemic.

Sample and procedure
At the outset of this project a Kuala Lumpur-based research firm created a list of 140 CSOs in Malaysia aimed at government reform. Interviews with leaders from 10 prominent CSOs were conducted, asking them to review the list of CSOs for its completenessadding missing CSOs and removing inactive groups. This process yielded a roster of 125 CSOs. Leaders from each CSO on this roster were asked to participate in a survey. Of the 125 organizations on the roster, leaders from 90 CSOs participated in a face-to-face survey for a response rate of 72%, meeting minimum response rates for network research (Doerfel & Taylor, 2004).

Measures Dependent variables
Collaboration network. The collaboration network was created using a full roster method (Borgatti et al., 2018) where participants were given the roster of 125 CSOs and asked to identify the ones they had worked with in the past six months. The phrase "worked with" was defined for participants as any formal or informal collaboration, partnership, or linkage. CSOs had from zero to 32 ties with an average of nine ties (SD = 6.72). There were 334 ties in the network. This produced a network density of 0.064, clustering coefficient of 0.322, and edgewise reciprocity of 0.258.
Bridgers, bonders, and gatekeepers. There are many network metrics, usually measured continuously, that can be used to assess the importance of nodes (e.g., betweenness centrality). When trying to identify if these metrics capture the theoretical ideas they were supposed to embody, several issues may arise. First, many network metrics, especially centrality metrics, are highly correlated (Oldham et al., 2019), raising questions about whether or not they are measuring the same thing. This also raises multicollinearity issues when these metrics are put into a linear model. Most importantly, such network metrics are reduced to continuous attributes without regard for the implications they have on the overall network structure. For instance, if one has a betweenness score of 37 and another of 89, does that mean the latter is a broker, but the former is not? If we removed one of these nodes, would it impact the network structure at all?
One solution to this issue is the use of what Borgatti (2006) called key player analysis (KPA). KPA is a method to nominally find relevant nodes in the network based on what impact they would have on the network if they were removed. Here an algorithm chooses a node for removal based on its impact on the network, recalculates network metrics, and then the process starts all over again until saturation is reached. For instance, Pilny and Proulx (2022) used KPA to identify relevant terrorist organizations regarding four different types of centralities, allowing them to simultaneously enter in a predictive model with no concern for multicollinearity.
Consider the idea of brokerage. Without brokers, a network should theoretically be fragmented into components of unconnected cliques. We used KPA to identify the minimum number of nodes that, if removed, would create an extremely fragmented network based on a network level metric of noncohesion. Saturation is based on scree plots, such as those used for an exploratory factor analysis to identify key points of inflection (i.e., an elbow). In using this procedure on our data, the baseline measure of Borgatti's (2006) noncohesion was 0.57. An inflection point was found at 0.90, when 21 key players were removed from the network. Similarly, we used an algorithm using the overall clustering coefficient to be minimalized to identify bonders. The baseline was quite high at 0.22, so it took the removal of 58 nodes to reduce the clustering coefficient to 0.01 using Watts and Strogatz's (1998) metric of node level clustering as information to nominate organizations to remove. Finally, because gatekeepers are nodes that serve both roles (connect distant cliques while having dense ties in at least one clique), we consider the 14 overlaps to be gatekeepers. In the end, using KPA, we identified 7 brokers, 14 gatekeepers, and 39 bonders in the network. More detailed information is included in Online Supplement File 2, including validity tests with traditional node-level metrics to measure these concepts (e.g., constraint, betweenness, local clustering) and network visualizations.
Influence. The measure for influence was adapted from Doerfel and Taylor (2004). If the organization was perceived to have any influence, respondents were asked to "rate the influence this actor has on government reform" on a scale from one (little influence) to five (great influence). This measure indicates the degree to which organizations perceived their partners to be influential on reform (M = 3.64, SD = 1.04).

Independent variables
Organization age. Participants were asked the year in which their organization was founded. The difference between the founding year and 2019 (year of data collection) was used as the age variable. CSOs' age ranged from 1 to 72 years (M = 19.86, SD = 17.07).
Organization type. Several kinds of actors may play a role in civil society (Taylor, 2011). Participants self-categorized their CSO as follows: NGOs (n = 67), government agencies (n = 6), activist groups (n = 5), academic institutions (n = 5), international NGOs (n = 4), donors (n = 1), or other (n = 2). Given that the number of some organization types was quite small, a new variable was created where CSOs were categorized as NGOs (n = 67) or non-NGOs (n = 23).
Issue priority. Many issues require the attention of Malaysian civil society aimed at government reform. The research team and local research firm implementing the survey iteratively discussed and refined those issues into eight categories and verified the face validity and exhaustiveness in the key informant interviews with 10 leaders of prominent Malaysian CSOs. Respondents selected their organization's primary issue focus from this list. These issues included: good governance (n = 6), Indigenous people and minority rights (n = 29), security reform (n = 7), media reform (n = 4), institutional reforms (n = 21), civil service reforms (n = 2), labor rights (n = 8), and reform of government-link companies (n = 2). Our initial model building found those issues with fewer than four selections were being overestimated in the models; thus, only Indigenous rights, institutional reforms, labor rights, and security had acceptable frequencies for analysis.
Geography. More than half of the organizations on the roster were located in the Klang Valley (n = 90), which includes the capital city Kuala Lumpur. The remaining CSOs were in the peninsular states of Johor (n = 3), Kelantan (n = 3), Penang (n = 5), Perak (n = 2), and in East Malaysian states of Sabah (n = 14) and Sarawak (n = 9). Of the CSOs that completed the survey, 52 were in the capital region and 21 were in the provinces.

Analysis
Each analysis was chosen to most appropriately test which association is being looked at based on how they were measured. For instance, varying homophilous attributes are predicted to be associated with binary collaborative ties for H 1 . As such, binary exponential graph models (ERGMs) are used. On the other hand, H 2 posited a relationship between a valued homophily matrix and a valued network of general influence. As a result, valued ERGMs (VERGMs) were used instead. For H 3 , H 4 , and RQ 1 , we used binary indicators of structural positions from the KPA (i.e., bonders, brokers, and gatekeepers) and their association with the likelihood of forming homophilous ties. In this case, newly developed autologistic actor attribute models (ALAAMs) were used. Each type of analysis is briefly described below.
ERGM/VERGM. ERGM was used for the binary data that indicates the presence of ties among CSOs. This procedure provides the statistical likelihood of a tie forming between two CSOs based on endogenous (i.e., ties with other nodes) or exogenous (i.e., node attributes) parameters (Robins et al., 2007). VERGM was used on the valued relations data for influence among CSOs. VERGM is similar to ERGM but is capable of modeling valued ties (for detailed explanation see Krivitsky, 2012;Pilny & Atouba, 2018). Both techniques use a Markov Chain Monte Carlo algorithm to explore parameter values, simulate networks based on those values, and refine those parameter values until they maximize the likelihood of returning a network that looks like the observed one (Shumate & Palazzolo, 2010). This produces Maximum Likelihood Estimates (MLE) that identify significant endogenous and exogenous parameters influencing tie formation (Robins et al., 2007) or increases in tie strength. The "ERGM" and "ERGM.count" packages in R were used for our analyses (Goodreau et al., 2008;Hunter et al., 2008).
ALAAM. In ALAAMs, the key dependent variable is a binary attribute. 1 What is being used to predict such binary attributes are statistical network configurations that one would use to assemble an ERGM (cf. Parker et al., 2021). In a sense, ALAAMs are similar to logistic regression models, but rather than assuming independence, ALAAMs assume interdependence via network configuration and use such possible forms of interdependence as predictors (i.e., a model with zero network terms takes the form of a logistic regression). Here, we use network statistics regarding outgoing and incoming homophilous ties as predictors of whether or not organizations were classified as key players (i.e., brokers, bonders, and gatekeepers). MPNet (Wang et al., 2009) was used to fit ALAAMs to answer H 3 , H 4 , and RQ 1 . 2

Results
Below are the results of three different inferential network models (ERGMs, VERGMs, and ALAAMs). Though they all predicted different things, they have similar modelbuilding procedures. First, each model needed enough simulations to estimate MLEs that do not produce degenerate networks, where the average value of a parameter is as it should be, but distribution is skewed (e.g., bimodal) such that any one network does not contain the expected parameter value. Often this is a sign of a poorly specified model or too few runs to properly estimate the MLEs. Second, each final model should consistently produce networks that look very similar to the observed network, often known as goodness of fit tests. These metrics are included in Online Supplement File 3, and they demonstrated non-degenerate and well-fitting final models. Table 1 contains the results from the binary ERGMs. Common endogenous network self-structuring mechanisms were included as control variables. These included an edges parameter, which is typically interpreted as a naive probability of a tie between any two random nodes (i.e., by itself, it is a measure of network density). Results suggested that CSOs were quite selective with ties and that they were nonrandom (MLE = −2.888, SE = 0.15, p < .001). The parameter for when CSOs return ties to each other, reciprocity, indicated that the observed network has a greater chance for CSOs having reciprocal relationships than would be expected by chance alone (MLE = 1.164, SE = 0.25, p < .001). The triad closure parameter (Goodreau et al., 2009) suggested CSOs in the observed network were more likely to form triadic relationships than by chance alone (MLE = 1.296, SE = 0.11, p < .001). The final endogenous parameter in the baseline model assessed the "extent to which a network shows a tendency of nodes not directly linked to each other being at least indirectly linked" (Broekel & Hartog, 2013, p. 59), or referred to as transitive shared partners. The significant and negative coefficient suggests that CSOs in the network had generally fewer unconnected dyads with shared partners than predicted (MLE = −0.249, SE = 0.02, p < .001). CSOs in the observed network tend to directly connect. Previous studies suggested that CSOs would be more likely to have relationships with other CSOs that were similar in (a) age, (b) type, (c) issue priority, or (d) region (geography). Here, the results were somewhat mixed, but for the most part they lend support for homophilous explanations of tie formation. The model for H1 first indicated that the age of CSOs produced degenerate results in the model; therefore, it was removed from the model. CSOs' organizational type had no significant impact on tie formation among NGOs (MLE = −0.034, SE = 0.11, p = n.s.) and non-NGOs (MLE = 0.261, SE = 0.18, p = n.s.). Turning to the issue priorities, there is some evidence that having a common issue priority had a significant positive effect on tie formation. There was a greater likelihood of ties forming among CSOs that shared the issue priorities of Indigenous rights (MLE = 0.307, SE = 0.13, p < .01) and labor rights (MLE = 0.593, SE = 0.30, p < .05). Insignificant results were found for issue priorities regarding institutional reforms (MLE = −0.182, SE = 0.26, p = n.s.) and security (MLE = −0.826, SE = 0.95, p = n.s.). The final parameter, geographic proximity based on the region in which a CSO was located, was a significant positive predictor of ties forming among CSOs located in the capital region (MLE = 0.464, SE = 0.08, p < .001). Yet, geographic proximity did not appear to impact tie formation among CSOs located in the provincial regions (MLE = 0.240, SE = 0.17, p = n.s.). Capital CSOs were more likely to form ties with those who were also located in the capital area. Because the key dependent variable was a valued measure of influence with respect to government reform, we used VERGM to answer H2 (see Table 2). Structurally, the perceived influence network was characterized by high reciprocity, low transitivity, and high decentralization. We also controlled for positionality in the cooperation network. Of the three, brokers were seen as the most influential on government reform (MLE = 4.45, SE = 0.25, p < 0.01) while gatekeepers were not far behind (MLE = 3.39, SE = 0.21, p < 0.01). Bonders, too, were rated as somewhat influential on government reform (MLE = 1.39, SE = 0.15, p < 0.01).
H 3 : Are bonders likely to be influenced by homophily when forming ties?
In essence, H3 tested for whether or not organizations in dense clusters (i.e., bonders) were more likely to form homophilous ties. Bonders are representative of bonding social capital, wherein dense clusters of ties provide support and resources. This hypothesis was partially supported. Bonders were more likely to receive homophilous ties from those with similar ages (MLE = −0.04, SE = 0.02, p < 0.01) and states (MLE = 0.42, SE = Interestingly, gatekeepers seemed to be influenced by both homophily and heterophily (see Table 3). Although gatekeepers were more likely to receive homophilous ties with organizations focused on of the same type (MLE = 0.26, SE = 0.14, p = 0.06), they were less likely to receive ties with organizations working in the same issue domain (MLE = −0.51, SE = 0.27, p = 0.06). Put another way, gatekeepers tended to collaborate with CSOs with different issue priorities (opposite of brokers), but also collaborated with other CSOs of the same organizational type. Interestingly, gatekeepers were the only type of organizations that tended to have reciprocal ties with other gatekeepers (MLE = 1.09, SE = 0.60, p = 0.09). Table 3. Autologistic actor attribute models.

Discussion
Overall, we aimed to (1) replicate studies of homophily theory as a source of tie influence in a national civil society network and (2) hypothesize upon the relationship between homophily and other features of network embeddedness like social capital positions. Overall, our contributions to homophily and civil society network research can be distilled into three key points. First, previous homophilous attributes replicate well in a national CSO network (i.e., type, issue priority, and location), except for age. Second, homophily seems to be pervasive. Both bonders and brokers tend to be influenced by homophily, albeit in different ways (i.e., age and state for bonders, and issue priority for brokers). Third, gatekeepers tend to be influenced by heterophily (i.e., issue priority) and homophily (i.e., organizational type). We elaborate on these findings in the following sections.

The homophilous attributes that influence tie formation
Our research confirms the findings of past studies that some homophily characteristics like common issues and geographic location lead to tie formation among CSOs (e.g., Atouba & Shumate, 2010, 2015O'Brien et al., 2019). Given that these factors seem present in a variety of interorganizational settings (see Online Supplement File 1), we would suggest that they be treated as "must-have" control attributes for future research on homophily as a predictor of tie formation. But what do these findings mean with respect to existing theories of homophily?
For instance, Malaysian CSOs have a likelihood to form ties with others when located in the capital region. This finding builds on past research that found geographic homophily was a mechanism relevant to CSO relationship formation at the international level (Atouba & Shumate, 2015). Our research adds more nuance by considering the impact of a specific region and found that geographic homophily was particularly important for influencing tie formation at the national level among capital-region CSOs. Conversely, the same did not occur among CSOs in the provincial regions. CSOs outside of the capital region have a greater distance among each other, and potentially less reason to connect with each other as capital-based CSOs tend to be better resourced (Douglass, 2005). Finally, issue priority was a significant predictor as similar priority on two issues were good predictors of ties (i.e., Indigenous rights and labor rights). Future theorizing on specific issue priorities and homophily seems ripe for understanding the causes that bind. That is, it seems entirely plausible that certain social, political, or economic issues might attract communication network ties more than others.
Finally, although age did not seem to influence tie formation overall, our results do not suggest that age played no role. For instance, bonders were more likely to be embedded in clusters with similarly aged CSOs, suggesting that age may be more of a local-level, rather than global-level network general mechanism. Local-level network mechanisms are a significant feature of Poole and Contractor's (2011) network ecosystem theory. Mechanisms operating at some local levels but not others simply mean that "different generative mechanisms may be in operation in different parts of the network" (p. 207). That is, if large networks are indeed composed of different local cliques, each of those cliques may be characterized by similar or different network general mechanisms. For the use of homophily as a theory of network tie formation, we suggest that teasing out local-level versus global-level network general mechanisms could be a promising area of future research. In other words, sources of homophily that may influence one clique may or may not be a source of homophily that influences another. Analyzing exactly why that might be true in some contexts could be a way to advance theories of homophily as a general mechanism influencing network tie formation.

The pervasiveness of homophily and network social capital
Although we hypothesized that homophily would be associated with bonders, those who are embedded in dense, clique-like structures, we somewhat unexpectedly found that homophily was also associated with CSOs that occupied brokering and gatekeeping positions as well. Moreover, homophily across all measurements was related with the likelihood of CSOs rating each other as more influential on government reform, even after controlling for key brokerage positions. This would suggest that homophily might be quite pervasive not only with respect to tie formation, but important civil society resources like perceptions of influence. However, the devil is in the details with respect to what sources of homophily matter the most.
Consider the nature of the four homophily attributes measured. If, as described earlier, some of these attributes can be linked to distinctive social identities and produce valorization effects, it might be worth asking if such homophilous attributes differ by dimension. For instance, Hajek and Giles (2002) describe how actors can engage in various identity management strategies, with one important process being mobility, the ability to change or leave current social identities. However, some identities can be more or less mobile. For instance, things like age and ethnicity are ostensibly stable while other identities like profession or political affiliation can be relatively more capable to change or leave.
Of the four current attributes, we would argue that age and state seem less mobile than organizational type and issue priority. After all, it might be harder to change a CSO's headquarters than to change its issue preferences, never mind it being impossible to change the age of one's organization. Viewed through this lens of mobility, we see that bonders tend to have an association with homophilous attributes that are less mobile, such as a CSO's age and state. Could this mean that dense, clique-like structures are more likely to form with attributes that are less mobile than others? Brokers, in contrast, tend to have an association with perhaps the most mobile attribute, issue priority. For instance, Rao et al. (2000) theorize a link between social identity and social network theory, demonstrating how some firms engage in social mobility (i.e., exiting an in-group to join an out-group). Their study focused on firms leaving NASDAQ stock exchange to join the rival NYSE, a mobile identity indeed. Nevertheless, the larger point is that it might be time to start classifying homophilous attributes across relevant dimensions (e.g., mobility) to more fully understand why they are associated with tie formations and why they might be associated with other relevant forms of social capital (e.g., Atouba & Shumate, 2015).

The curious case of gatekeepers
Gatekeepers represent somewhat of a peculiarity in both how they are classified and their association with homophilous ties. With respect to the former, they are embedded in both dense clique-like structures and have ties bridging other cliques in the network. Likewise, their ties are associated with both homophily and heterophily. More specifically, gatekeepers tended to have more ties associated with similar organizational types, but also tended to have ties with organizations focusing on different issue priorities. Why might this be the case?
Heterophilous patterns associated with gatekeeping are intuitive to explain as this represents part of the gatekeeping job: selecting what information enters the group and what information does not. Indeed, Lewin (1947) theorized that such selection processes are at the heart of gatekeeping and the very idea of selection implies that there are heterogenous alternatives available (cf. Shoemaker, 1991). This was, of course, the opposite of the homophilous issue patterns associated with brokers. As such, it makes sense that gatekeepers, if they do perform selection functions, are more likely to have ties with others that have different issue priorities.
Moreover, gatekeepers were also more likely to have homophilous ties with other CSOs of the same type (NGOs or non-NGOs). Indeed, of the 14 CSOs classified as gatekeepers, 12 (85.71%) were NGOs. This seems to support Stohl and Stohl's (2005) idea of NGOs' unique ability to fill structural holes and become embedded in dense cliques influenced by network closure. Moreover, gatekeepers were the only ones who were more likely than by chance alone to have reciprocal relationships with other gatekeepers. All of these findings point to the idea that gatekeepers might be performing a wide portfolio of network tie strategies as compared to brokers and bonders. Future research might consider more analysis of gatekeepers to extend and further refine theories of gatekeeping (e.g., Barzilai-Nahon, 2008). For instance, under what conditions do gatekeepers seek homophilous and heterophilous strategies? Why do they tend to establish ties with other gatekeepers? Which strategies are related to various outcomes?

Limitations
There are several limitations to acknowledge. First, future research would benefit from considering how ties in networks evolve with time and affect the social capital benefits offered to bridgers, bonders, and gatekeepers (cf. O'Brien et al., 2019). The cross-sectional nature of the current data points to a chicken-and-egg dilemma about the types of network positions associated with homophily. For instance, it could be possible that central brokering positions enable CSOs to find other like-minded CSOs to connect with. And for others, homophilous ties could be the network strategy to gain advantageous network positions. Without longitudinal comparison it is difficult to determine if homophily is an antecedent or an outcome to network structures like social capital positioning.
Second, as our data relied on self-reports from CSO leaders, there is the possibility of biased or incomplete information provided by the participants. For instance, Corman et al. (2020) show how the correlations between perceived and observed networks evolve and change over time. Third, like most studies of interorganizational networks our study is limited by the fact that it is a case study, and while few survey-based network studies achieve full participation (cf. Doerfel & Taylor, 2004;, full participation by all actors in a network is desirable.