A Social Network Analysis of Collaborative Rangelands Governance: The case of Borana Rangelands, Southern Ethiopia

Effective natural resources governance plays a crucial role in enhancing the resilience of socio-ecological systems (SES) in the face of environmental changes. It is recognized that the ability to adaptively respond to complex environmental change and manage SES resilience resides in actors’ networks. Network forms of governance facilitate both horizontal and vertical interconnection of actors, bring different perspectives and sources of knowledge, and develop shared values and innovative solutions to problems. However, the structural pattern of actors’ collaborative linkages within the network strongly influences actors’ behavior and, hence, delivery and impacts of effective governance. We analyze social networks (SNA) among pastoralists in the Borana rangelands of Ethiopia to identify the structural gaps that result in misfits. Our quantitative SNA revealed a low level of network density with very few horizontal and vertical interactions and linkages among actors in the governance system, which considerably limits flows of knowledge, experiences, and other resources, leading to a failure to establish shared values and undertake joint action. Rangelands governance in Borana is further hampered by the absence of adequate network heterogeneity and solidarity that in turn blocks the building of collaborative planning and efficient use of available resources to address the complex problems of rangeland pastoralism. Our results suggest that a policy environment that can create conditions for greater collaboration, the strengthening of actors’ ties, and the development of trust and social capital enabling the design of effective collective governance should be developed.


Introduction
Borana pastoralists have over the years developed a set of rules and regulations on how to use their rangeland resources and evolved robust customary institutions to support their systematic use and sustainable management of these resources under harsh environments without government Teferi Tolera ttolera2009@gmail.com (Ayana & Oba, 2007;Li and Li, 2012;Reid et al., 2014;Kelemework, 2016).
Natural resource management has long been dominated by a centralized and hierarchical decision-making approach that has attempted to simplify the complexity of such rangeland governance systems by focusing on a few variables, but this has inevitably resulted in the erosion of SES resilience (Adger et al., 2005;Folke et al., 2005;Walker et al., 2006;Reid et al., 2014). In the face of apparent failures to address complex environmental problems by centralized management and reductionist approaches, network forms of governance have emerged that engage with the concept of collaborative governance (Newig et al., 2010), which recognizes that actors' engagement is the basis for maintaining SES resilience and sustainability in rangelands as elsewhere (Hruska et al., 2017). A collaborative governance system that involves social networks comprised of various actors including government, non-government, and user groups across geographical and jurisdictional governance scales, levels, and units is more likely to be effective in addressing the evolving complexity and unpredictability of environmental problems (Ostrom et al., 2007;Bodin and Crona, 2009;Armitage et al., 2010). A collaborative governance system plays an important role in facilitating collective learning, producing common understanding or shared goals, mobilization of key resources, deliberation of commitment and collective actions, and resolution of conflicts (Carlson and Berkes, 2005;Scholz and Wang, 2006;Bodin and Crona, 2009;Newig et al., 2010).
However, the presence and engagement of multiple actors alone are not sufficient to build an effective natural resource governance system (Bodin and Crona, 2009). The types and features of individual actors and networks among actors fundamentally determine actors responses and behavior in governance regimes even more than the existence of formal institutions (Scholz and Wang, 2006;Walker et al., 2006;Bodin and Crona, 2009;Larson et al., 2013) reflecting the critical importance of investigating the characteristics and influences of each actor in the network and the overall features of relational ties (Bodin and Crona, 2009).
Understanding the existence and structure of social networks of the Borana rangelands governance system is important to search for how "collaborative barriers" can be overcome in addressing the complexities in the rangeland resource governance (Bodin and Crona, 2009;Larson et al., 2013). However, to the best of our knowledge, empirical evidence on the effect of social networks on natural resource governance regimes in general and rangelands resources, in particular, is still lacking in the Borana setting.
Drawing on the collaborative governance and SES frameworks, we examine the Borana rangeland network governance structures affecting the capacity and effectiveness of the governance system in responding to current challenges by applying social network analysis (SNA). Specifically, we address two interrelated research questions: (1) which actors are involved, and in which ways do they engage in collaboration with others? (2) which actors are key players in coordinating or bridging roles in the rangelands network governance system?
We first present our methodology employed and define the case study and network boundaries, followed by our findings. Finally, we conclude with a discussion of the major research questions.

Study Location
We conducted this study in the Borana zone rangelands of southern Ethiopia (Fig. 1). The Borana zone is one of 13 administrative zones in Oromia National Regional State. It is located between 3°36-6°38'North latitude and 3°43'-39°30' East longitude and borders Kenya in the south, the Somali Regional Government in the east, and the Ethiopian highland districts in the north. The Borana rangelands cover an area of about 50,000 km 2 of which 75% consists of lowland, and are frequently exposed to droughts. Based on CSA (2017) population census, the Borana administrative zone is currently inhabited by an estimated population of about 1.2 million people.
The Borana rangelands are characterized by an arid and semi-arid climate, with pockets of sub-humid zones. The average annual rainfall varies between 350 and 900 mm with a considerable variability of 21-68% (Lasage et al., 2010). Rainfall is bimodal, with 60% of annual rainfall occurring between March and May (main rainy season) followed by a minor peak between September and November (small rainy season). The long-term variability in the quantity and the distribution of the rainfall results in recurrent droughts in the area. In the past, droughts lasting several years occurred approximately once in 20 years and isolated dry years (< 400 mm) once in five years (Coppock, 1994). Recently, the period between droughts has decreased to 5 years, and the most recent drought occurred only 3 years after the previous one (Bekele, 2013).
The topography of the Borana rangelands is distinguished by plain rangelands, intersected with occasional mountain ranges, volcanic cones, and depressions, and an altitude between 750 and 1700 m.a.s.l. They are dominated by tropical savannah vegetation, with varying proportions of open grasslands and perennial herbaceous and woody vegetation (Alemayehu, 2002). Cossins and Upton (1988) classified the Borana rangelands into four ecological zones of different potential grazing capacity; high potential savannah in the northern part, bushlands with high shrub cover in the central area, medium potential grassland in the east, and volcanic areas in the west.
Our research area, Yabello district, is representative of the core of the Borana pastoralism system and the cultural sites of the Gada system (Indigenous governance system) that govern overall rangelands resource use (Ayana & Oba, 2007). Recently, development interventions such as the establishment of ranches, intensification of cropping farms, curtailment of mobility, etc., have intensified in this district, which has led to diversified interests and interdependence of actors. Yabello district encompasses a total area of 5556.7 square kilometers, characterized mainly by lowlands and hills. It lies between 1350 and 1800 m.a.s.l. Based on current demarcation, the district comprises 18 kebele, the lowest administrative unit, which eight are mainly dependent on pastoralism, and the remaining ten are characterized as agro-pastoralist.

Social Network Boundaries Identification
Identification of the network boundary is the first step in conducting social network analysis. However, the network boundaries are largely arbitrary and subject to the nature of research questions. The unit we used for our analysis was the rangelands governance system in Yabello district ( Fig. 1), defined as a group of actors interacting on regular basis and influencing approaches to SES resilience. Toward this end, we surveyed various stakeholders involved directly or indirectly in rangeland management in Yabello district across all governance scales.

Actors' Identification Process and Data Collection
Before discussing the process involved in the identification of stakeholders/actors considered in network analysis, it is important to highlight the formal institutional setup related to rangeland resource governance in Ethiopia. The political administration has five tiers: national or federal, regional states, zones, districts, and kebeles. Generally, the institutional setup for natural resource management and rangeland management issues, we consulted available secondary data, specifically policy documents for qualitative data to complement the initial process. Ultimately, a total of 53 actors were included in our SNA.
We gathered quantitative data through in person interviews with individuals in key positions such as heads and deputy heads of the organizations, planning officers and senior experts, and department heads who have an understanding of the functions of the organizations. To assess the characteristics of actors' interactions, and how various organizations cooperate in governing rangeland resources, we asked our respondents to identify organizations with which their organization collaborated on the issues of rangeland management within the last two years. In addition, we asked them to explain the mandate and roles of their organizations. We asked representatives of some organizations, particularly at the federal level, to complete the questionnaires by themselves. All interviews were guided by pre-tested checklists.

Network Data Analysis
Analysis of the nature of actors' interactions in a network is a prerequisite to assessing the effectiveness of the rangeland resource governance network (Bodin and Crona, 2009). The network structures (relational patterns) influence the behavior and actions of actors and the overall effectiveness of the governance system (Sandström & Rova, 2010). These network structural characteristics, which have strong functional implications for the resilience of the network, broadly include number of social ties, degree of cohesion, subgroup inter-linkages network centralization, and actor centrality (Bodin and Crona, 2009).
In shedding light on how each of the network structural characteristics affect the overall performance of natural resource governance, the number of social ties, captured by network density, greatly affects the outcomes of the network governance as the more social ties tend to increase the possibilities of collaboration, mutual trust development and joint action (Sandström & Rova, 2010). The existence of higher network density also facilitates the co-production of knowledge that is useful in SES resilience building (Bodin and Crona, 2009).
The other important structural characteristic, the level of network cohesiveness, measures to what extent the network is separated into distinguishable subgroups. Looking at its effect on the outcome of the network governance, the existence of subgroups, and low cohesion, can pose challenges to the collaborative process among subgroups (Hannemann and Riddle, 2011). Generally, less cohesive networks exhibit high density in the overall network and produce positive governance outcomes as it facilitates sharing of important resources such as knowledge and information (Bodin and governance follows the same jurisdictions. Rangeland use is governed by the Federal Democratic Republic of Ethiopia Rural Land Administration and Land Use Proclamation No. 456/2005, which provides the general framework and power for national and regional governments to set hierarchical organizational structures down to the district level to regulate sustainable land-use. However, there are no formal state organizations specifically dealing with rangeland resources management. At the national level, policy issues and development interventions related to rangeland management are coordinated by the Ministry of Environment, Forests, and Climate Change (recently named as a commission), the Ministry of Agriculture and Natural Resources, the Ministry of Water, Minerals, and Energy, the Ministry of Federal Affairs, and the Ministry of Livestock and Fishery Development (currently merged with Ministry of Agriculture and Natural Resources). At the regional level, rangeland resource governance and management are shared responsibilities of the Bureau of Environment, Forest, and Climate Change Authority, the Bureau of Agriculture and Natural Resources, the Bureau of Rural Land Administration and Use, the Bureau of Water, Minerals, and Energy, and the Bureau of Livestock and Fishery Development. The institutional structures at the Zonal and district levels follow the same lines. Alongside these state institutions, many organizations such as Research Institutes, Universities, Non-government organizations, State Enterprises, and Community Based Organizations and/or customary authorities have an interest in rangeland management practices.
The Ministry of Federal Affairs is mandated to coordinate the pastoral development programs and projects in pastoral areas. However, in practice, there have been poor coordination efforts. In connection with this, many argue that the intention of government is largely to control pastoral areas (Abdulahi, 2007). More importantly, as indicated by our key informants, the linkages between stakeholders' institutions are quite weak at all levels of management, especially among different lines of management.
To identify active stakeholders/actors involved in rangeland resources management, we employed a combination of purposive and snowball sampling techniques. Initially, we engaged in successive consultations with major organizations having direct roles in rangeland governance including the Borana Zone Pastoral Development Office, the Pastoral Commission, and the Land Use and Environmental Protection Office requesting recommendation for contacts among other stakeholders actively involved in the rangeland governance system, until we reached saturation and no new organizations were mentioned. We interviewed two local actors. To avoid biases that may arise from the snowball sampling method in overlooking other stakeholders involved in governance scales and the density of subgroup structures was examined.
To identify actors' structural position, or coordinating actors that would otherwise have limited or no connections, we analyzed parameters of the network centrality. Network centrality measures how central or well-connected an actor is in a network (Sandström & Rova, 2010). It also describes the patterns of power relations and how much an actor has access to the resources in the network (Dkamela et al., 2014;Angst et al., 2018). For the present study, from various versions of centrality parameters, we considered degree centrality and betweenness centrality.
Degree centrality measures the number of direct links to and from an actor (Sandström & Rova, 2010). The degree centrality parameter helps to visualize how tightly the network is organized around its most central point or how 'star-like' the network structure is (Bodin and Crona, 2009;Sandström and Rova, 2010;Scott, 2015). A high degree of centrality is interpreted as well-connected network structures (Sandström & Rova, 2010). On a related front, to identify potential bridging actors within identified clusters, we calculated betweenness centrality scores for each actor in the network. Betweenness is the number of shortest paths from all nodes (actors) to all others that pass through one specific node (actor) (Dkamela et al., 2014).
Betweenness centrality helps to quantify how much each actor contributes to minimizing the distance between actors in the network (Bodin et al., 2006). It measures an actor's bridging position concerning the other two actors in the network. It is the probability of an actor being on the shortest path between two points, reflecting how often a node lies on the shortest path between any two nodes in the network (Angst et al., 2018). To put it another way, betweenness centrality refers to the number of times an actor is located between two other actors who are disconnected. An actor with high betweenness centrality is considered to have a great influence throughout interactions, hence affecting how the rangeland resources governance system delivers governance assets (Hanneman & Riddle, 2005).
To identify actors playing an influential role in connecting actors out of the clusters, following Vance-Borland and Holley (2011), we calculated brokerage scores. All network metrics used to describe the characteristics of the governance networks were calculated using UCINET version 6.591 (Borgatti et al., 2002). To analyze the network structure, the data were transformed into binary network measures (1 for the presence of collaboration and 0 for the absence of collaborative relationships). To depict actors' relationships visually, NetDraw software was employed. Crona, 2009). It is important to note that if connectivity exists between different subgroups (bridging ties) there would be a high possibility of using external resources, which in turn improves the capacity of the network governance (Crona & Hubacek, 2010). In Borana rangeland governance, for instance, the connections beyond the subgroups potentially promote and create a collaborative partnership between various types of actors including NGOs, government line organizations, higher education and research institutes, community-based organizations, and customary authorities.
The relative position and influence of individual actors, covered in the concepts of network centralization and actor's centrality, significantly affect the capacity of the natural resource governance regime in many ways (Bodin and Crona, 2009). The level of centrality deals with the distribution of linkage among actors (nodes) in the network and their structural importance (Bodin et al., 2006). Centrally, the network centrality helps to understand which actors and how they can use their structural position and able to influence the collaborative process and how they are advantageous over the others in accessing important resources (Crona & Hubacek, 2010). Though there are multiple ways to measure network centrality, for this study we attempted to measure two metrics: degree and betweenness centrality.
Following the relevant literature (Hanneman & Riddle, 2005;Bodin and Crona, 2009;Sandström and Rova, 2010;Prell, 2012;Scott, 2015) we used methods of social network analysis to map, quantify, and analyze the relational patterns or connection between actors in rangeland governance. We measured the structural properties of networks described above to analyze the effectiveness of network governance (see below).
The network density measures the proportion of all possible ties present in a network and is used as a proxy for social cohesiveness (Bodin and Crona, 2009). Density is calculated by dividing the number of existing connections in the structure by the maximum number of possible connections (Scott, 2015). D = l n(n−1)/2 Where: D denotes density. l = maximum number of links. n = number of actors. The network structures in which all actors are completely connected have a density of 1 reflecting strong cohesiveness in the network structures and effective governance. However, the existence of subgroups in the network may mislead the interpretation of the whole network and the existence of various subsets of actors should be considered (Sandström & Rova, 2010). In our case, the data were clustered into six subgroups based on the main organizational goals and regional level, and 0.14. Likewise, the density of networks among non-government organizations was 0.26. The density of the overall government actors (D = 0.25) at all levels resembled that of NGOs (0.26). It is notable that there was no collaboration between community-based actors, indigenous Gada institutions, and community-based organizations. Interestingly, we observed an increasing the tendency for collaborating with customary institutions.
Network governance for resilient SES demands collaboration among all actors (higher density), which improves the potential for joint actions through facilitating knowledge, resource and experience sharing, fostering mutual trust, solidifying shared goals, and helping to manage potential conflicts (Bodin et al., 2006;Bodin and Crona, 2009;Sandström and Rova, 2010). Network governance with higher density enables actors with differing knowledge, values, and interests to understand complex problems, agree on solutions and integrate their actions, hence increasing the adaptive capacity of SES (Alexander Steven et al., 2016). As it can be seen from the findings, there were very few overall and horizontal collaborations in the Borana rangeland governance, and the absence of such interactions potentially undermines the collective action, block communications of knowledge and experience, hinders resource sharing, triggers conflicts of interests, and ultimately derives SES vulnerabilities which currently prevail in the Borana pastoral system. In addition, low-density networks reflect the divergence and competition of perception, goals, and interests that lead to the absence of a common problem definition, which hinders the potential to respond adaptively to uncertainties (Crona & Hubacek, 2010). Sandström and Rova (2010) further elucidate that a wellconnected network is assumed to increase the capacity of a co-management network to craft and maintain the rules and the compliance of actors to the set rules. Contrary to the recommendations of existing literature on SES, the Borana rangeland governance system critically lacks sufficient collaborations among actors and hence lacks a well-defined common priority process. Even worse, the low level of network cohesiveness appears to have hindered the development of a common view on the future of the pastoral system, let alone setting common goals.
On the other hand, it has been argued that cross-boundary connections (heterogeneity) increase the effectiveness of adaptive governance through mobilizing diversified resources (Alexander Steven et al., 2016). Effective governance, thus, requires multilevel coordination of actors' interactions across all administrative units involving both policy-making bodies and policy implementing actors (Fliervoet et al., 2016). Such cross-scale and vertical interactions potentially facilitate the sharing of diversified resources. In our case, however, the network has low

Results and Discussion
We first present the whole network properties of the rangeland governance system. Subsequently, we discuss the actor (node)-specific characteristics of the network structures and their likely effects on the deliverance of governance assets.

The Structure of Rangelands Governance Network
Rangeland governance involves a diverse set of actors both NGOs and GOs representing various scales, sectors, and divergent interests (Fig. 2). We identified 53 network nodes or relevant actors in the rangeland governance system of which NGOs accounts for 28% (n = 15), whereas government actors made up 68% (n = 36), and the remaining 4% (n = 2) of actors belong to community-based organizations, customary institutions, and private companies. As indicated by our key informants, the most common activities that NGOs participate in Borana rangeland management are bush clearing and pond/well construction and maintenance.
Looking at the overall links, theoretically, a density of social networks equal to one represents full connections of all nodes, in our case, however, the overall density of social networks in rangeland management is very low (D = 0.076) in that from 2760 possible links, actors make only 207 (7.6%) ties, which is low poor compared to what other studies, for example, Sandström and Rova (2010), recommend. Closely looking at the network links, only about 32% of links were found to be reciprocated. It was also found that government actors exhibited strong hierarchical vertical links across formal governance tiers with very limited horizontal links at all governance scales.
Despite being on the same governance level, having interdependent goals, and physically close to one another, there was little horizontal reported collaboration among actors operating at the same governance level. An examination of the network densities shows that the densities of all horizontal networks were low; 0.16 for district-scale government actors, 0.056 for the zone level, 0.014 for the  (Table 1). Most of the actors playing a key role in the network were government organizations. Non-government organizations such as SAVE and GDPI (Gayo Pastoralist Development Initiative) also reported a high level of collaboration indicating they are actively involved in the rangeland governance.
Looking at the indegree (popularity or visibility of an actor to others (Dkamela et al., 2014), YDPDO (Yabello District Pastoral Development office) (18) received the greatest number of nominations as a collaborator followed by BZPDO (Borana zone Pastoral Development office) (15) ( Table 1).
The perceived dominance of the YDPDO and BZPDO is because the formal functions in controlling pastoral issues are mainly assumed by these government organizations. Overall, almost all actors from the top ten popular actors belong to government organizations. To visualize each actor's connecting or bridging roles, we analyzed betweenness centrality scores for each actor. The betweenness centrality measures how frequently an actor is situated between two other actors, and high betweenness centrality scores reflect that an actor holds a favorable position for facilitating and controlling communication flows, which in turn indicate a position of brokerage (Hannemann and Riddle, 2011). In our case, the betweenness centrality index was found to be 14% indicating poor network closure and hence, poor governance outcomes (Sandström & Rova, 2010).

Note
The red arrow indicates non-reciprocated collaboration and the blue arrows stand for reciprocated relationships. Pink Square: NGOs, Black Square regional: GOs, Red square: Zonal level GOs, Green Square: Private organizations, Blue Square: Local Organization, Gray Square: District Level GOs, Light Blue: Federal Level GOs.
Actors such as BZEPOAO, YDPDO, YADELAO, YDCPO, and SAVE had higher betweenness centrality scores (Table 1), reflecting their role in brokering the connections of different actors who are themselves disconnected. Despite the dominant role of connecting different actors, SAVE, one of the non-government actors, holds the second rank in connecting different actors, indicating the organization is playing a considerable bridging role.

Conclusions
Our study was premised on the assumption that collaborative governance involving various state and non-state actors is a venue to reverse the current deterioration of SES resilience cross-boundary connections in that there was virtually no direct reported collaboration among stakeholders spanning two levels of governance; actors tend to collaborate with actors who share similar institutional mandates following formal institutional setup, to the nearest level up or down the formal governance hierarchy.
Unfortunately, multi-sectoral and multi-scalar interactions of actors were scarce in the Borana rangeland governance system. For instance, there was no direct collaboration between district and regional level actors or between zone and federal level actors outside of the formal institutional structures, and thus the rangeland governance system failed to synchronize governance assets in responding to the dynamics in SES. In Ethiopia, for more than four decades, the issues of participation and collaboration have been widely discussed in the policy arena. However, the issue is practically absent on the ground. In line with this, one of our key informants from the Yabello Drylands Pastoral and Agricultural Research Center, stated: Nowadays, it is common to hear the issues of participation of stakeholders and collaboration here and there while designing development and research projects here and there. But I am not sure that we have the technical capacity to do so. We just say it to attract the attention of donors to get funds.

The Importance of Actors' Connecting Roles in the Network
To examine the importance of each actor in decision-making related to rangeland management, we assessed degree centrality, and betweenness centrality ( Table 1). The number of collaboration links reported by a given actor (outdegree) reflects the role of that actor in the governance network. In Borana rangeland governance networks, BZEPOAO inform a better design of joint action for adaptive governance of the rangeland SES. of the Borana rangelands by addressing the complexity posed by climate and socio-political change. It has become clearer that, as opposed to hierarchical and top-down forms of governance, network forms of governance facilitate both horizontal and vertical interconnection of actors, bring different perspectives and sources of knowledge, develop shared values, and develop innovative solutions to rangeland environmental problems. In view of this, by analyzing social networks, we attempted to identify the structural gaps that result in misfits in Borana rangeland SES.
As evident from the low level of network density, there are very few horizontal and vertical interactions and linkages among actors in the Borana rangeland governance. This does not mean that having high social relations guarantees increased joint action under all circumstances, but such a poor linkage considerably blocks flows of knowledge, experiences, and other resources indicating the inability of the governance system to facilitate the solidification of shared values and joint action. It has been argued that the environmental problem is an inherently complex issue demanding the integration of scattered resources and capacities distributed across scales, sectors, and domains of governance systems (Bodin and Crona, 2009).
More importantly, in countries where problems of limited human and other resources are very common, a governance system that enables the integration of scattered resources and coordination of action is essentially recommended. Unfortunately, the rangeland governance system in Borana is hampered by the absence of adequate network heterogeneity and closure, which in turn blocks the building of collaborative strategies and efficient use of available resources to address complex problems. Thus, we suggest that a policy environment that can create conditions for more collaboration, strengthen actors' ties, foster the development of trust and social capital, and enable the design of effective collective efforts should be put in place.
We also attempted to identify the actors that were playing the most important roles in reaching actors who are not otherwise connected. In the network, one of the government actors, the Yabello District Pastoral Development Office (YDODO) was ranked highest as the most important collaborative actor by other organizations (in-degree value) followed by the Borana zone Pastoral Development office (BZPDO), reflecting that these organizations are more responsive in dealing with rangeland management issues and attempts to enhance the Borana SES will benefit from involving these key organizations. One of the limitations of our investigation was that we focused only on the general collaboration of actors and failed to investigate various relational patterns in the network such as information and other resource flows, policy influence, power relationships, etc. Further research on the flows of these resources could