Building Health Communities: Local Health Care Networks in Maryland

Abstract

Combatting chronic disease (prevention and treatment of obesity, diabetes, heart health, and stroke) requires action at the local level, both to educate the public and to provide health services. Effective collaboration among local organizations devoted to educating the public about, and treating patients of, these diseases is a key component of successful health care. To better understand local efforts, a social network analysis of five local health care networks spanning eight counties in Maryland was conducted. The purpose of this exploratory research was to discover whether collaborative networks exist at the local level, to map the networks, and to assess their strengths and needs.

Keywords

collaboration healthcare networks

Networks and networking principles play an increasingly important role in governance. As is the case with many public services, networks are being promoted as a means of effectively delivering public health. In this paradigm, local health departments play a coordinating role, acting as brokers and bridges between intersectoral networks of collaborative partners (Institute of Medicine [IoM], 2012; National Association of County & City Health Officials [NACCHO], 2016, 2017). A body of work examining local health care networks is beginning to develop (e.g., Harris, Leider, Carothers, Castrucci, & Hearne, 2016; Hogg & Varda, 2016; McCullough, Eisen-Cohen, & Salas, 2016). This study adds to that scholarship by mapping and examining five local health care networks in Maryland. Using the distinction between serendipitous and goal-directed networks (Kilduff & Tsai, 2003), this study more closely examines health care networks and is able to make network management recommendations based on the type of network identified. The article begins with an overview of network and network management literature and reviews serendipitous and goal-directed networks. A shift toward collaboration and networking in health care follows, along with a case description. Methods and results of the study are presented and discussed; the article concludes with a call for additional lines of research and highlights the necessity of developing mechanisms to support network management in practice.

Networks, an Overview

It has been nearly 20 years since a “transformation of governance” was described (Kettl, 2000). That transformation was based largely on the increase of boundary-spanning connections, decentralization, devolution, and a reliance on street-level actors to deliver services. In other words, governance was adopting network principles and practices. Networks have long been framed as a third form of governance, alongside market-based and hierarchical approaches (Considine & Lewis, 1999; Miller, 1994; Powell, 1990). As Keast, Mandell, and Brown (2006) note, these three “primary modes of social organization . . . represent ways of organizing society” (p. 27). The growth of network governance has given rise to concerns about the hollowing of the state (Milward & Provan, 2000; Rhodes, 1997) and the impact that decentralization, a flat organizational structure, and dispersed accountability will have on traditional structures of government.

Networks and networking, however, appear to be here to stay, as they offer distinct benefits. Fundamentally, networks accommodate resource scarcities, as they proceed from conditions of interdependence and facilitate strategies to optimize shared resources (Keast, Mandell, Brown, & Woolcock, 2004; Klijn, 2001). Networks are “structures of interdependence” (O’Toole, 1997, p. 45); the justification for forming a connection with another person or organization is that a go-alone strategy is not possible (Agranoff & McGuire, 2001a; Imperial, 2005). Networks form as actors realize that the only way to achieve their goals is through collaborative action and resource sharing (Agranoff & McGuire, 1998). Networks are dynamic (Keast et al., 2004). They are decentralized, lacking a formal chain of command or central locus, networks are able to distribute information and resources efficiently (Bogason & Musso, 2006; O’Toole, 1997; Raab & Milward, 2003). They are characterized by fluid participation and so can accommodate turnover and changes in personnel. Owing to their fluidity and lack of a centralized structure, they do not rely on traditional forms of authority or accountability, but on relational power. They are driven by the connections between actors rather than by the positions of those actors in a network (Keast et al., 2004; Raab & Milward, 2003). In other words, in a network, it does not matter where you sit, but who you know and how you work with them. Operating effectively in this environment requires a distinct set of skills different from those required management of a bureaucracy or traditional hierarchical organization.

A growing body of literature addresses the challenges and strategies for managing within networks. Agranoff (2007) distinguished four types of public networks, noting that different sorts of networks may require different management techniques. He also raised the question of whether a network analog of traditional management activities can be identified, proposing activating, framing, mobilizing, and synthesizing as the network equivalents of traditional POSDCORB activities (Agranoff & McGuire, 2001b). Taking a different approach, Provan and Kenis (2008) assess three network management strategies: shared governance, where management and governance responsibilities are distributed throughout a network; a lead organization approach, where one organization in the network takes on management duties; and a network administrative organization approach, where an organization outside the network is contracted to manage the network. Provan and Kenis note that the choice of an appropriate management strategy is influenced by a number of factors, including size and complexity of the network and degree of goal alignment among members. This means that knowing what sort of network one has is critical to managing that network effectively.

Goal-Directed and Serendiptious Networks

Kilduff and Tsai (2003) propose two ideal types of what they variously term “trajectories,” “organizational principles” or “processes of change”; distinguishing between goal-directed and serendipitous networks. A goal-directed network starts “with the establishment of a goal” (Kilduff & Tsai, 2003, p. 93). It “exhibits purposive and adaptive movement toward an envisioned end state” (p. 92). In other words, “goal-directed network trajectories develop around specific goals that members share” (p. 89). Goal-directed networks share a common goal of which members are aware and toward which they are consciously working. As Human and Provan (2000) put it, actors “see themselves as part of the network and are committed to network-level goals” (p. 329). A larger network-level purpose serves as an organizing and screening mechanism: new members may be attracted to the network by the opportunity to pursue the stated network goal; thus, members tend to self-select based on fitness with articulated goals (Kilduff & Tsai, 2003, p. 92). Likewise, a common purpose facilitates organization among actors and the pursuit of common resources (p. 92). As a result, one of the key features of a goal-directed network is the “emergence of an administrative entity that acts as a broker to plan and coordinate the activities of the network as a whole” (p. 89). Goal-directed networks, then, are relatively formalized structures; they are self-aware (in that actors know that they are in a network, know why they are in that network, and can articulate shared goals); and have some processes for coordination.

Serendipitous networks, on the contrary, grow from individual actors working independently to establish ties to others (Kilduff & Tsai, 2003, p. 93). In this type of network

individual actors make choices about who to connect with, what to transact, and so on, without guidance from any central network agent concerning goals of strategy. Actors form ties or partnerships based on their own interest. Serendipitous networking can provide conduits through which information and other resources flow. (p. 90)

There are no network-level goals that drive interactions. There may be little formal organization or processes of communication or coordination. As a result, the network develops haphazardly; “at any point in time, any specific pairs of actors may or may not share goals” (pp.89-90).

These two types impact network structure (Kenis & Provan, 2009, p. 450). Goal-directed networks have clear boundaries between members and nonmembers and have clearly articulated criteria for eligibility along with statements of benefits and obligations. They are likely to be highly centralized—organized around a leader or set of leaders who articulate network goals and recruit members (Kilduff & Tsai, 2003, p. 95). Serendipitous networks, on the contrary, tend to exhibit a looser structure and lack a centralized core. Most members will be connected to only a small subset of the larger network; meetings tend to be local, and trust bonds are strong among localized connections. Overall network boundaries are ill-defined (p. 96).

Provan and Kenis build on the work of Kilduff and Tsai (e.g., Kenis & Provan, 2009; Provan & Kenis, 2008), distinguishing between what they term self-initiated or voluntary (serendipitous) networks and mandated or contracted (goal-directed) networks. Of particular relevance here is their observation that serendipitous networks are more common. However, while goal-directed networks occur less frequently, they are the more-studied form (Provan & Kenis, 2008, p. 231). One of the contributions of this article is to add to the body of work examining serendipitous networks, and to call for additional work on this type of network. The article also emphasizes that type matters—serendipitous networks have specific characteristics, strengths, and limitations, and should be managed with those particular features in mind. These skills and knowledge will be important for health care managers, as the United States adopts a network-centric approach to health care.

U.S. Health care Policy—A Shift Toward Networks

The American Health Care System regularly is ranked low among developed nations (Institute for Healthcare Improvement [IHI], 2018; Tandon, Murray, Lauer & Evans, 2000; World Health Organization, 2000) and is criticized for high costs and lack of commensurately high outcomes (Cha, 2017). Consequently, U.S. health care is undergoing a transformation, with national policy recommending a multi-pronged approach to improve patient care and raise community health while reducing costs (Hester, Stange, Seef, Davis & Craft, 2015, p. 1; IHI, 2018). This transformation emphasizes population health, a concept that recognizes health outcomes as the product of many determinants, that roots health in the community, and that sees health as a communal outcome (Kindig & Stoddardt, 2003). Put simply, “population health is a shared responsibility” (NACCHO, 2017, p. 24). This shifts the focus of public health care provision away from the traditional model of direct provision of services and toward a collaborative approach in which “public health agencies can convene or join partnerships aimed at creating environments in which people can be healthy” (IoM, 2012, p. 21).

As it is rooted in the community, population health efforts are grounded in local organizations, with local health departments playing a key role. Here, too, the focus has shifted away from direct provision of services, and beyond the “steering” oversight of contract- and grant-funded providers. Instead, local health departments are being encouraged to develop multi-sectoral collaborative partnerships; NACCHO (2017) notes that “multi-sectoral partnerships play an increasingly critical role in the movement to improve health, equity, and economic prosperity” (NACCHO, 2017, p. 29), and views local health departments as “chief health strategists” (NACCHO, 2016). In this role, healthy communities are realized through a wide range of actors working in concert. The task of the local health department is to develop and maintain collaborative partnerships with and between actors from all sectors. The IoM lists “partnership development and community mobilization” as a foundational public health capability and views partnership development as “a mission-critical need” for health (IoM, 2012, p. 157). The NACCHO definition of a functional local health department includes the following elements:

Collaborates with other local responders and with state and federal agencies to intervene in other emergencies with public health significance (e.g., natural disasters).

Engages the community to address public health issues.

Develops partnerships with public and private health care providers and institutions, community-based organizations, and other government agencies (NACCHO, 2005, p. 2).

NACCHO further emphasizes the need to “develop relationships,” “engage the community,” “engage health systems,” and “develop partnerships.” This echoes the IoM’s inclusion of “mobilizing community partnerships” as 1 of the 10 essential public health services (IoM, 2011, p. 32). All of this means that local health departments must become effective at operating in a complex environment, working across sectoral boundaries, managing multi-faceted partnerships, and communicating efficiently to a dispersed and disparate set of collaborators. In short, they must become adept at managing networks.

Local Health care in Maryland

As noted above, national health policy emphasizes the importance of networks of collaborative partnerships at the local level as a driver of health. In support of this goal, the Centers for Disease Control and Prevention (CDC) has focused on developing “clinical and community linkages to better support chronic disease self-management” (CDC 1305, 2016, para 4), “supporting community prevention strategies” and “linking community programs to clinical services” (CDC 1422, 2015, para 2). To support these goals, grants were awarded to states to identify and support local health care communities. As an awardee, Maryland identified five areas across the state for study. The purpose was to determine whether local collaborative relationships exist and to assess the state of any partnerships that might be found. These areas are: (a) Garrett and Allegany counties, and (b) Washington county in the western end of the state; (c) the independent city of Baltimore in the central portion of Maryland; and two areas on Maryland’s eastern coast: (d) Caroline and Dorchester counties, and (e) Somerset, Wicomico, and Worcester counties. The location of these five areas are shown in Figure 1.

Figure 1.

Five local health care areas in Maryland.

Cumulatively, these five areas include eight counties plus the independent city of Baltimore, encompassing 4,659 square miles of area home to 1,120,450 people. Networks were postulated to exist in each of these areas. Local health departments were assumed to have a central role, as Maryland has a strong county system and relies on county offices to coordinate local efforts. However, there was no explicit mandate for counties to do so; networks had not been formally developed, and this is the first exploratory study to describe networks in these areas. Basic demographic information about each region is summarized in Table 1.

Table 1.

Demographic Overview*.

Network	Western Maryland		Washington county	Baltimore	Caroline and Dorchester		Lower Shore
County	Garrett	Allegany	Washington	Baltimore city	Caroline	Dorchester	Somerset	Wicomico	Worcester
Population	30,005	75,528	149,585	620,961	32,579	32,384	25,768	102,370	51,540
Population	105,533		149,585	620,961	64,963		179,678
Area (sq mile)	656	430	467	92	326	983	610	400	695
Area (sq mile)	1,086		467	92	1,635		1,705
Population density (people/sq mile)	45	169	321	6,700	99	33	42	256	74
Population density (people/sq mile)	97		321	6,700	40		105
Per capita income	25,348	21,674	27,066	25,707	24,943	27,870	16,631	26,241	32,419
Median household income	45,432	40,551	56,228	42,241	52,465	47,093	35,154	52,278	56,773
Persons in poverty, %	14	20	12	24	14	18	26	15	11
High school or higher, %	88	88	87	83	83	85	81	88	89
Bachelor’s or higher, %	19	17	20	29	16	21	15	29	30
Median age	38	39	37	35	37	41	36.5	35.7	43
% under 18 years	22	18	23	22	24	22	17	22	18
% 65 or older	17	18	14	12	13	18	14	13	23

2010 Census data.

As may be seen in Table 1, these five areas do share some common demographic characteristics, particularly with regard to age—both median age as well as percentages of the population under 18 and over 65—and in attainment of high school diplomas. There is greater variation across these counties with respect to higher education as well as to poverty levels, although there is no evidence of correlation between these categories. Interestingly, there is also variation across these measures within areas, as evidenced in the differential between the adjacent counties of Somerset and Worcester counties in the Lower Shore. However, the defining characteristics of these five areas emphasize the split between the urban setting of Baltimore and the rural eastern and western arms of Maryland. As the 29th most populous city in America, there is a clear difference between that city’s population density and those of the counties in the remaining four areas. That differential is greater than can be shown in Table 1, though. The most populous county in these areas, Washington County, holds roughly one third of its population in the county seat of Hagerstown. Similarly, the next two most populous counties, Wicomico and Allegany, each have roughly 1/3 of their population in two towns, with the rest of each county consisting of small towns and villages and unincorporated areas. Hagerstown, in Washington County, is the only municipality with a population greater than 35,000 (other than Baltimore). Thus, a central challenge for all areas is how best to address the needs of their populations and the challenges posed by the distribution of that population.

Methods

Data collection and analysis was undertaken independently for each area. Data collection was accomplished in two steps. First, a snowball sample was administered. Snowball sampling is an established network analytic technique useful in exploratory research, particularly when members of a network are not known to researchers (Atkinson & Flint, 2003; von der Fehr, Sølberg, & Bruun, 2016). One or more initial respondents, assumed to be in a network, are identified and asked for their contacts in that network. Those contacts are then asked to identify their own contacts; the process is repeated until no new contacts are generated. While it does have limitations, including the potential for selection bias and gatekeeper bias, snowball sampling has been shown to be an effective means of generating accurate data, particularly for hidden or opaque networks (Atkinson & Flint, 2003; Doreian & Woodard, 1992; von der Fehr et al., 2016). For this study, staff at each local health care department were used as seeds for the snowball survey. These contacts were requested to identify organizations, or departments or divisions of a large organization, and to provide contact information for their main contacts at each organization with whom they worked in the fight against chronic disease.¹ These organizations were added to the seed list that was distributed to all newly identified organizational contacts with a request to add additional organizations and contacts. This process was repeated iteratively until no new organizations or contacts were identified.

After the snowball survey was completed, a second survey, adapted from Cross and Parker (2004), was sent to all contacts within each network to assess the quantity and qualities of connections between organizations. After gathering organizational information, this survey listed all organizations identified in the snowball and asked respondents to select those organizations with whom they worked directly. Subsequent questions focused on those partner organizations, eliciting information on the length of connection, type, and frequency of communication, how well respondents understood the skills and knowledge of their partners, and whether respondents more often sent information and resources to, or received information and resources from, selected partner organizations.

Both surveys were administered via Qualtrics. Up to three requests were emailed for each survey; in addition, telephone outreach was used to increase response rates for the network survey.

The unit of analysis is the organization. Surveys were intentionally fine grained. Divisions and departments of large organizations were listed separately, as these could be clustered later during analysis. Similarly, multiple contacts at an organization were frequently identified during the snowball sample. All contacts were included in the network analysis survey. Multiple responses from a single organization or department were combined as part of the data preparation process. In several instances, respondents from a single organization or division jointly completed a survey. All survey responses were held anonymously; only organizational or departmental identifiers were used.

Data were analyzed using NodeXL (Smith et al., 2010). Data were checked, cleaned, anonymized, and imported into NodeXL. All visualizations were also done within Node XL.

Ideally, complete data for a network will be made available. That is, all network members will respond to the survey instrument. Frequently, this is not the case; missing data through non-response is a common and potentially serious problem for network analysis (Robins, Pattison, & Woolcock, 2004; Stork & Richards, 1992). However, a unique aspect of network analysis is that “nonparticipation by a respondent does not necessarily mean that the respondent is not included in the study” (Borgatti & Molina, 2003, p. 339). Partial information on nonrespondents is available via ties from other actors to nonrespondents (Huisman, 2009, p. 3). A variety of techniques may be used to account for missing data and to incorporate partially observed actors into the network. Here we follow Robins et al.’s (2004) pragmatic approach of using “whatever information is available to strengthen interpretations, rather than to discard data when it is incomplete” (pp. 277-278). We do so using reconstruction, a common approach to network analysis (Stork & Richards, 1992, p. 198). Reconstruction uses “both fully described links (two descriptions)” as well as links indicated by only one respondent. “The assumption is that if A describes a relationship with B, that, indeed, a relationship does exist between them” (Stork & Richards, 1992, p. 197). As Huisman (2009) notes, “partial information . . . is used to obtain (better) estimates of the structural properties of the actors and the network” (p. 3).²

Results

From an initial pool of 15 contacts, the snowball survey identified a total of 678 contacts working in 593 organizations across all five networks. Table 2 breaks these results down by network. It lists individual contacts identified through the snowball sample as well as the number of organizations identified in each network. Survey responses for the network analysis survey are presented, along with the response rate (responses/individual contacts) as well as organizational coverage rate (responses/organizations in network). As shown in Table 2, survey responses cluster between 57% and 63%; due to the presence of multiple contacts at some organizations in each network, the organizational coverage rates are somewhat higher. While 100% participation is ideal, it is not necessary for network analysis. Network reconstruction facilitates the inclusion of partial data on nonrespondents, allowing incoming connections from respondents to nonrespondents to be included in the network. Organizational response rates fall within the standards established by Wasserman and Faust (1994), Kossinets (2006), Huisman (2009), and others.

Table 2.

Survey Responses.

Network	Individual contacts	Organizations	Survey responses	Contact response rate (%)	Organizational response rate (%)
Western Maryland	65	58	41	63	72
Washington county	104	99	59	57	60
Baltimore	337	288	195	58	68
Caroline and Dorchester	83	81	52	63	64
Lower Shore	79	67	46	58	69

An important result is that networks exist in each of the five areas under study. As this is an exploratory project, it was not known whether networks would be found. As will be shown below, each area has a functional network: there are only a few isolated (nonconnected) organizations and no clusters of organizations disconnected from each larger network. All networks involve organizations from a range of sectors.

The total number of organizations in each network is shown in Table 3. Network size varies between 58 and 287 organizations; network size does not correlate to population, population density, geographic size, or other metrics.

Table 3.

Network Organizations by Sector.

		Western Maryland		Washington county		Baltimore		Caroline and Dorchester		Lower Shore
Sector	Business	3	5%	18	18%	23	8%	3	4%	10	15%
	Government	10	17%	11	11%	55	19%	16	20%	24	35%
	Health System	25	43%	35	35%	62	22%	17	21%	12	18%
	Education	11	19%	8	8%	37	13%	10	12%	8	12%
	Nonprofit	9	16%	27	27%	110	38%	35	43%	14	21%
	Total	58		99		287		81		68

Table 3 breaks out these totals by sector. Respondents were asked to classify their organizations by sector using CDC categories and definitions; the research team confirmed these responses as well as filling in missing responses. Two minor amendments were made. First, organizations spanning two or more sector categories were placed in a sector based on their primary function. Thus, a nonprofit hospital would be classified as a “health system” organization rather than a “nonprofit.” Second, for this report, the CDC categories of “philanthropy” and “community” were folded into a single category—“nonprofit.”

Sector totals also vary by network, reflecting a range of strategies undertaken in each network. For example, over one third of Baltimore city’s network is comprised of nonprofit organizations, reflecting the important role that these organizations play throughout the city. Twenty percent of Washington county’s network comes from the private sector. Many of these are small local businesses, an indicator of the success that this network has had with recruiting local employers as partners. Similarly, the Lower Shore network—Somerset, Wicomico, and Worcester counties—shows strong participation from government agencies. Closer examination reveals that these include local fire departments, state and county parks as well as local parks and recreation departments, and local library systems. This highlights the success this network has had engaging governmental partners that directly service the public—of bringing health information and resources into public spaces so as to make information and services available to the public at their convenience rather than mandating that the public come to government to receive services, support, or information.

Network Measures

Table 4 displays overall metrics for each network. Nodes are the number of organizations in each network; edges are the number of connections between those organizations. Density is the ratio of actual connections to all possible connections. Note that density tends to be lower for serendipitous networks (Kilduff & Tsai, 2003, p. 91); note also that high densities can be detrimental to networks.³ Distance is the count of steps needed to connect a node to another node. Max distance is the largest number of steps needed to connect a node to another node in the network while average distance is the average across a network. In network 2, for example, no organization is more than five steps away from every other contact in the network; on average most organizations can reach any other organization in the network in just over two steps. Together, these measures tell us that network organizations are in relatively close connection to one another and that there are no outlying organizations at a significant distance from the rest of the network.

Table 4.

Overall Network Metrics.

	Western Maryland	Washington county	Baltimore	Caroline and Dorchester	Lower Shore
Nodes	58	99	287	81	68
Edges	338	381	2,102	570	257
Density	0.10	.04	0.03	0.09	0.06
Max distance	3	5	4	3	4
Average distance	2.0	2.3	2.4	1.9	2.2
Reciprocated edge ratio	0.24	0.12	0.14	0.15	0.19

Several standard network measures were calculated. These include the following:

Density—a measure of network completeness. It is the fraction of reported connections over all possible connections in a network.

Geodesic distance—the number of connections lying along the shortest path between two nodes. Both average and maximum distances were calculated.

In-degree centrality—the number of connections coming in to a particular node.

Out-degree centrality—the number of connections that a particular node reports having to other nodes throughout the network.

Betweenness centrality—how often a given node lies on the shortest path between any two other nodes in the network.

Table 4 presents basic information about these networks, including node and edge count, and maximum and average distances for each network.⁴ A more detailed picture of each network is given in Tables 5 to 9. These display in-degree, out-degree, and betweenness centrality for the top 10 organizations in each network. As the local health departments have clear interests in the networks and served as the seeds for the snowball surveys, they are indicated by name in the following tables. All other organizations have been obscured; they are listed in these tables by their sector; numbers are included after sector labels to allow tracking of central organizations across centrality measures. For example, “Health System 1.1” is the first health systems organization listed in network 1; “Health System 1.1.2” is a subsidiary division or subordinate department of Health System 1.1; and “Health System 1.2” is the second health system organization in network 1.

Table 5.

Western Maryland Centrality Scores.

In-degree centrality		Out-degree centrality		Betweenness centrality
Sector	Score	Sector	Score	Sector	Score
Allegany County HD	15	Health System 1.1	32	Garrett County HD	738.967
Health System 1.1	15	Nonprofit 1.1	30	Health System 1.1	407.424
Education 1.1	12	Health System 1.4	29	Health System 1.2	346.746
Garrett County HD	12	Health System 1.7	26	Allegany County HD	282.746
Health System 1.2	12	Allegany County HD	25	Health System 1.4	250.353
Health System 1.3	11	Health System 1.2	24	Health System 1.9	211.854
Education 1.2	10	Health System 1.8	24	Health System 1.8	199.295
Health System 1.4	10	Health System 1.9	22	Nonprofit 1.2	186.505
Health System 1.5	10	Health System 1.9	22	Health System 1.7	182.591
Health System 1.6	9	Garrett County HD	19	Health System 1.10	89.877

Note. HD = health department.

Table 6.

Washington County.

In-degree centrality		Out-degree centrality		Betweenness centrality
Sector	Score	Sector	Score	Sector	Score
Washington County HD	17	Nonprofit 2.1	29	Nonprofit 2.3	1,610.316
Health System 2.1	16	Nonprofit 2.2	29	Nonprofit 2.1	1,384.887
Health System 2.1.2	16	Government 2.2	28	Health System 2.1	1,075.934
Government 2.1	14	Nonprofit 2.3	28	Government 2.2	1,035.100
Health System 2.1	14	Health System 2.1.4	26	Washington County HD	916.616
Nonprofit 2.1	11	Washington County HD	25	Nonprofit 2.2	805.329
Education 2.1	11	Health System 2.1.2	22	Health System 2.1.4	505.011
Government 2.2	10	Education 2.2	20	Health System 2.2	464.454
Health System 2.1.3	10	Health System 2.2	19	Health System 2.3	375.867
Health System 2.2	10	Health System 2.1.5	19	Education 2.2	282.761

Note. HD = health department.

Table 7.

Baltimore City.

In-degree centrality		Out-degree centrality		Betweenness centrality
Sector	Score	Sector	Score	Sector	Score
Nonprofit 3.1	46	Baltimore city HD 2	94	Baltimore city HD 2	7,995.081
Baltimore city HD	41	Nonprofit 3.4	92	Nonprofit 3.4	7,056.278
Government 3.1	37	Nonprofit 3.2	79	Nonprofit 3.2	7,026.511
Nonprofit 3.2	37	Nonprofit 3.1	69	Nonprofit 3.1	5,438.999
Education 3.1	34	Nonprofit 3.5	61	Health System 3.3	4,210.666
Baltimore city HD 1	28	Nonprofit 3.6	61	Education 3.1	4,004.001
Nonprofit 3.3	26	Education 3.2	59	Nonprofit 3.5	3,350.614
Nonprofit 3.1.1	24	Health System 3.2	57	Nonprofit 3.6	3,151.394
Government 3.1.1	24	Health System 3.1	53	Education 3.2	3,125.039
Health System 3.1	24	Baltimore city HD 3	53	Baltimore city HD 3	3,119.361

Note. HD = health department.

Table 8.

Caroline and Dorchester Counties.

In-degree centrality		Out-degree centrality		Betweenness centrality
Sector	Score	Sector	Score	Sector	Score
Nonprofit 4.1	20	Nonprofit 4.4	59	Nonprofit 4.4	979.414
Caroline County HD	18	Dorchester County HD 2	58	Dorchester County HD 2	966.513
Health System 4.1	18	Dorchester County HD	50	Dorchester County HD	828.210
Government 4.1	16	Nonprofit 4.1	45	Nonprofit 4.1	493.092
Dorchester County HD 1	16	Health System 4.3	43	Health System 4.3	420.130
Health Systems 4.2	16	Education 4.1	37	Caroline County HD	311.947
Dorchester County HD	15	Health System 4.2	34	Health System 4.2	311.345
Nonprofit 4.2	15	Dorchester County HD 3	34	Education 4.1	303.653
Nonprofit 4.3	14	Education 4.2	21	Dorchester County HD 3	233.808
Education 4.1	12	Nonprofit 4.5	19	Nonprofit 4.6	126.351
Government 4.2	12	Caroline County HD	17	Nonprofit 4.7	102.615

Note. HD = health department.

Table 9.

Lower Shore.

In-degree centrality		Out-degree centrality		Betweenness centrality
Sector	Score	Sector	Score	Sector	Score
Worcester County HD	14	Nonprofit 5.4	33	Worcester County HD	1,711.331
Wicomico County HD	12	Worcester County HD	32	Nonprofit 5.4	1,132.430
Health System 5.1	12	Wicomico County HD	30	Wicomico County HD	716.179
Nonprofit 5.1	9	Education 5.2	25	Education 5.4	445.454
Nonprofit 5.2	9	Nonprofit 5.6	13	Nonprofit 5.6	310.380
Nonprofit 5.3	9	Somerset County HD 1	12	Nonprofit 5.8	163.223
Nonprofit 5.4	8	Nonprofit 5.3	11	Nonprofit 5.2	116.522
Education 5.1	8	Nonprofit 5.7	10	Health System 5.1	101.553
Health System 5.2	8	Education 5.3	10	Nonprofit 5.5	70.716
Health System 5.3	8	Nonprofit 5.1	9	Nonprofit 5.3	69.785
Nonprofit 5.5	8	Nonprofit 5.8	9	Health System 5.3	66.293

Note. HD = health department.

A visualization of each network is paired with the table listing the top centrality scores, to place those scores in the context of the full network structure, and to facilitate comparison between networks. In each visualization, nodes are sized by in-degree centrality and are color-coded by sector.

Theory suggests that serendipitous, or voluntary, networks will have lower densities, be less highly centralized, and demonstrate stronger local ties than network-wide connections (Kilduff & Tsai, 2003). These results only partially support this conceptualization. As shown in Table 4, overall network densities are relatively low. Tables 5 to 9 show that each network does have a central core of highly active organizations. There is significant overlap of top-ranked organizations across centralization measures. Moreover, as may be seen in these tables, top-ranking organization are connected to a substantial portion of each network, with the exception of Baltimore. For in-degree centrality, the top-ranked organizations tend to have connections with between 12% and 25% of the overall network. For out-degree centrality, the proportion is higher, topping out at more than half of the network in three cases. These highly connected organizations form a central core for each network, as may be seen in Figures 2 to 6. While some local clusters may be seen in each network, a distinct central core is also present in each network, largely due to the impact of highly active organizations.

Figure 2.

Western Maryland full network.

Figure 3.

Washington county full network.

Figure 4.

Baltimore city full network.

Figure 5.

Caroline and Dorchester counties full network.

Figure 6.

Lower Shore full network.

Tables 5 to 9 reveal variations among these networks. For example, the strong role of nonprofits in the Baltimore network may be clearly seen in the dominance of that sector among the central core of that network. These tables also highlight some similarities across networks. All measures of centrality drop in all networks. Betweenness centrality is calculated on a log scale, so the rates of drop for that metric should be assessed accordingly. In-degree and out-degree centralities are simply counts of connections between organizations. Across both measures, and for all five networks’ top-ranking organizations, these scores drop by roughly half. In all networks, these most-central organizations are in the core of the network and represent the most well-connected organizations. Nevertheless, the level of connections reported by these core organizations decreases rapidly.

Analysis of network metrics revealed a discrepancy that is at least marginally present in all networks. Some organizations report outgoing connections to other organizations in their network, but do not have any reported incoming connections. That is, while an organization reports connections to other network members, none of the organizations in that network report a connection to the organization in question. These organizations are listed in Table 10.

Table 10.

Organizations With Out-Degree But no In-Degree Centrality, by Network.

Note. HD = health department.

As can be seen in Table 10, every network has at least one organization where a connection is not recognized by the remainder of that network. In several cases, these are organizations that see themselves as well-connected, often reporting connections to one quarter to one third or more of the entire network. It is problematic that these connections are not recognized or reciprocated by partner organizations. A related concern is the precipitous drop in all networks from total reported connections to reciprocal connections—that is, those instances where two organizations each report an outgoing connection to the other. Results on reciprocal relationships are presented in Table 11. As these are emergent networks—they coalesced around common goals rather than being developed according to an explicit mandate or a set of conscious guidelines—it is not surprising that network relationships are not well understood by participants. However, a mutually acknowledged connection is the foundation of a collaborative partnership; increasing awareness of network relationships and organizational activity within these networks should increase efficiencies as well as effectiveness.

Table 11.

Reciprocal Connections.

	Western Maryland	Washington county	Baltimore	Caroline and Dorchester	Lower Shore
Total connections in network	338	381	2102	570	257
Reciprocal connections	78	46	292	82	48
Total organizations in network	58	99	287	81	68
Organizations in reciprocal relationships	21	17	61	21	18

Discussion

The networks described here are serendipitous—these networks were not “set up”; there is no evidence that there was any plan to establish or develop these networks nor to maintain or evolve them. Thus, the first result—that functional networks exist—is signficant. As can be seen in Figures 2 to 6, each local region consists of a single network. There are few if any isolates and no clusters or groups disconnected from the network as a whole. The formation of five independent instances of coherent, integrated networks across a broad domain (chronic health) and incorporating 60 or more organizations, should be seen as a success and as a foundation for further developing policy goals for building collaborative cross-sector partnerships at the local level.

The fact that every region in the study developed a coherent network with few isolates or diads and no isolated clusters is indicative of the power and appeal of networking as a means of organization. As Kenis and Provan (2009) have noted, serendipitous networks may be more common, but they are less-frequently studied than are goal-directed networks. These results show that large and complex networks do emerge and may persist for some time. The mechanisms by which networks emerge and self-organize continue to deserve more investigation.

Impressively, relationships between network organizations are often long lasting. Respondents were asked to indicate the duration of their relationship with each partner organization. In each network, 25% to 30% of network organizations reported maintaining multiple relationships for 10 years or longer. Building on this, respondents report good working relationships with their partner organizations. Asked about the extent to which they understand the skills and knowledge of their contacts at each partner organization, more than half of all respondents in each network strongly agreed, with an additional one third agreeing to this statement. Overall, in each network, more than 90% of respondents agreed that they understood their partner organizations. While this emphasizes the strength of particular ties within these networks, it also highlights a central challenge for serendipitous networks: organizations understand and have good working relationships with their immediate partners. This does not translate into an understanding of the network as a whole. The disparity between network size and number of reciprocal connections (Table 11) is one indicator of the differential between particular linkages and connections to the entire network.

Put bluntly, the networks are not self-aware—network organizations do not know that they are part of a larger network, are not aware of the size or structure of the network, and are not familiar with potential network partners. From a research perspective, this was assumed to be the case, as this was an exploratory study designed to identify organizations and map the networks. Practical implications were revealed during the survey process, as it became clear that organizations knew only their own partners. In phone calls and emails, survey respondents divulged that they did not know there was a larger network in operation. Responses ranged from skepticism about the extent of the networks and their activities to incredulity that an organization unknown to them was doing similar work in their community. In general, respondents were curious to know the shape and composition of their network. For them, the value of this study lay in uncovering which organizations were participating in each network.

Interestingly, respondents—particularly the local health departments—were often more interested in the absences in their network. Health department officials wanted to know which local organizations were not identified as network members and which organizations did not report connections and partnerships. Direct feedback from the people involved in network partnerships made it clear that for them, the value of this study lay in identifying participating organizations and in disclosing which organizations were working together and which were not.

All five networks show robust cross-sector connections; three networks face the additional complexities involved in coordination across county boundaries. This may be clearly seen in the Western Maryland (Figure 2) and Lower Shore (Figure 6) networks. Both figures reveal two different network management approaches. In both figures, a secondary hub-spoke arrangement may be seen in the top half of the network diagram, centered around a blue node; in each figure, this node is one of the county health departments; the other county health department(s) are near the center of the lower half of the network diagram. The difference between strategies is particularly emphasized in the Western Maryland network, where the Garrett county health department serves as the single hub at the top of the network, in contrast to the Allegany county health department which is embedded in the web of the lower half of the network. The point is not that one approach is better, but that health departments and network organizations may not be aware that each county is using a different strategy. Thus, communication, management, and governance procedures are not likely to be consistent throughout the network, increasing opportunities for missed communication and duplication of efforts while lowering efficiencies.

Similarly, the Washington county network shows a nonprofit organization serving as a secondary hub and the sole connection to a number of nonprofit organizations. As there is no overt management system in place, there is not widespread awareness of the role that this hub nonprofit plays in the network, nor a recognition that this organization may not function within the schedules or protocols of the larger governmental and health systems organizations in the network. In this instance, failure to work effectively with the hub nonprofit endangers 10% of the network.

This underscores a challenge that serendipitous networks face: they function at the level of individual connections rather than as a network. These five networks developed “opportunistically,” as individual organizations made connections; there were no “conscious efforts to build coordination” (Provan & Kenis, 2008, p. 231). This is not surprising, but it does pose management challenges that likely impact the effectiveness and efficiency of the networks. Most obviously, there is no management strategy in any of these networks. These are emergent networks, so it is not surprising that there is no plan for development, maintenance, or management. However, relying on a network to collectively manage itself is not realistic, especially given the size and complexity of these networks (Provan & Kenis, 2008).

As the structure of the networks has not, until now, been made overt, effective management of the network has not been possible. Knowledge of the network’s structure and constituent parts is a necessary precondition to effective management, but it is only the first step. Some coordination is required to establish governance protocols for each network. Critically, resources must also be allocated to the network for management and maintenance. This can be especially problematic for serendipitous networks, particularly for those operating at a local level. Networks are responses to resource scarcity; local organizations are often those with the least stability of human and financial capital. With no founding mandate or predetermined management structure, devoting resources to the network—to the space between organizations—can be difficult.

Limitations, Contributions, and Recommendations for Research

This project is dependent on self-report survey data and is limited by those responses. The unit of analysis is organizational; respondents are individuals within organizations. It is likely that not all individuals within an organization view network partners in the same light; different respondents may have answered surveys differently than their co-workers. It is also likely that respondents may not have been aware of, or may not have recalled, all relevant network connections. As noted, none of the networks achieved a 100% response rate, leading to missing data. Response rates were within norms for mitigating the effects of missing data, and reconstruction was employed to include partial data on nonrespondents (Huisman, 2009; Kossinets, 2006; Stork & Richards, 1992). Missing data do distort the structure of a network, as it overemphasize data from respondents (Huisman, 2009). This tends to lower network density, as not all connections are reported, and to overstate in-degree centrality against other centrality measures, as in-degree can include partial data on nonrespondents (Borgatti, Carley, & Krackhardt, 2006; Robins et al., 2004).

This study presents a snapshot of five health care–focused networks in Maryland. It was undertaken to determine whether these networks existed and was founded on the assumption that they did and that county and city health care departments were active in such networks. It was further assumed that survey respondents would be able to compartmentalize chronic disease as a discrete aspect of health care. Practically, it is not reasonable to expect that a chronic disease network exists independently of general practitioners, nursing or urgent care providers, or of other aspects of health care. Separating chronic disease out as a focus of study may create artificial boundaries within a larger and fuzzier network of health care organizations. Similarly, the study began with predetermined geographic boundaries. The five regions of interest were identified by the Maryland Department of Health; it is reasonable to expect functional networks to transcend these borders. Finally, this is a case study and is subject to limitations of that approach, particularly with regard to generalizing from these cases to other types of networks.

However, this study does make important contributions and highlights a need for additional research. In public administration, comparative studies of networks are relatively rare. More comparative research is needed if we are to develop standards for evaluating networks (Hoppe & Reinelt, 2010). As Goggins (1986) points out, comparing cases that are similar and comparable, such as these five networks, gives a measure of control and provides a starting point for the development of broader principles (p. 333). In addition, longitudinal assessment of networks—the comparison of a network over points in time—is also needed. Many organizations in these five networks report long-standing connections; network formation and maintenance has long been an area in need of more study. Longitudinal analysis may give a robust picture of how networks persist and change over time. A third means of comparison, that of surveying and separately mapping multiple respondents across a network’s organizations, may also provide insights into network structure and address the limitation noted above that network analysis depends on the perceptions of the respondent.

There is also a need for more research into serendipitous networks. Network theory has focused on goal-directed networks; in practice, most networks are serendipitous. Some findings in this study suggest that these networks may not conform to the theories, needs, and structure of goal-directed networks. Research should test whether goal-directed network theories and management techniques can be applied to serendipitous networks.

Conclusion

For at least two decades, networks have been heralded as a growing trend in public administration and a key component of a shift from government to governance (Agranoff & McGuire, 2001b; Considine & Lewis, 1999; Kettl, 2000; O’Toole, 1997). Government continues to promote networks as a means of accomplishing the ends of government and as an efficient method for delivering services to the public. The CDC and IoM’s calls for local health departments to function as chief health strategists, managing collaborative cross-sector partnerships rather than providing services directly or overseeing contracts, is one aspect of this larger shift. However, it is still not clear whether government has the capacity and the skills necessary to undertake this shift successfully, particularly when serendipitous (voluntary and emergent) networks are involved. As scholars have noted, these types of networks may be more prevalent than goal-directed (contracted, mandated) networks but are less-frequently studied. While serendipitous networks may be a common means of coordinating local actors and of sharing information and resources among organizations and across sectors, there has been little investigation of the ways in which their serendipity impacts their performance and stability or of the ways in which their management strategies and needs may diverge from goal-directed networks.

This article reports basic exploratory findings on five local health care networks. Encouragingly, serendipitous networks were found in all five areas under study. While findings confirm some aspects of serendipitous networks, other results seemed to align more closely with principles expected of goal-directed networks. More work is needed to better understand serendipitous networks. Research should be undertaken to assess how wide spread serendipitous networks are and to model their structure based on observations and original data. We need to better understand how serendipitous networks form, what keeps them together, what management techniques help bind and maintain the network, and how information and resources can be best shared among network actors. In addition to scholarship, more practical work is also needed to better manage serendipitous networks. A network management strategy is a critical component of an effective network (e.g., Agranoff, 2007; Agranoff & McGuire, 2001b; Kenis & Provan, 2009). However, none of these networks have a management plan. Local health care officials and other network participants knew of their own local connections but were not aware that they were part of a larger network. There was no provision for network-wide communication, governance, or coordination. Crucially, there was no provision for funding or resources devoted to the maintenance or management of these networks. Consequently, the main finding may not be that these networks exist, but should instead be the foundational question of why, lacking knowledge, resources, or a strategy to manage them, do these networks persist? If we continue to pursue a network-centric approach to governance, we would do well to understand these issues.

Footnotes

Acknowledgements

This research was funded by the Maryland Department of Health, Center for Chronic Disease Prevention and Control (contract no. OPASS 16-17137G). The research was made possible by the Schaefer Center for Public Policy. The author is particularly indebted to Ken Weaver for his assistance throughout this project.

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) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project was funded by the Maryland Department of Health, Center for Chronic Disease Prevention and Control (contract no. OPASS-16-17137G).

ORCID iD

Aaron Wachhaus

Notes

Author Biography

Aaron Wachhaus is an associate professor and director of the doctoral program in public administration at the University of Baltimore. He is a governance theorist and network scholar. He studies the emergence of collective organization and seeks to better understand the impediments to effective cross-sector collaboration.

References

Agranoff

(2007). Managing within networks: Adding value to public organizations. Washington, DC: Georgetown University Press.

Agranoff

McGuire

(2001a). American federalism and the search for models of management. Public Administration Review, 61, 671-681.

Agranoff

McGuire

(2001b). Big questions in public network management research. Journal of Public Administration Research and Theory, 11, 295-326.

Agranoff

McGuire

(1998). Multinetwork management: Collaboration and the hollow state in local economic policy. Journal of Public Administration Research and Theory, 8, 67-91.

Atkinson

Flint

(2003). Sampling, snowball: Accessing hidden and hard-to-reach populations. In Miller

R. L.

Brewer

J. D.

(Eds.), The A-Z of social research. London, England: SAGE. Retrieved from https://proxy-ub.researchport.umd.edu/login?url=https://search.credoreference.com/content/entry/sageuksr/sampling_snowball_accessing_hidden_and_hard_to_reach_populations/0?institutionId=975

Bogason

Musso

(2006). Introduction: The democratic prospects of network governance. American Review of Public Administration, 6, 3-18.

Borgatti

Molina

(2003). Ethical and strategic issues in organizational social network analysis. The Journal of Applied Behavioral Science, 39, 337-349.

Borgatti

S. P.

Carley

M. K.

Krackhardt

(2006). The robustness of centrality measures under conditions of imperfect data. Social Networks, 28, 124-136.

Centers for Disease Control and Prevention. (2015). State and local public health actions to prevent obesity, diabetes, and heart disease and stroke. National Center for Chronic Disease Prevention and Health Promotion. Retrieved from https://www.grants.gov/web/grants/search-grants.html?keywords=CDC-RFA-DP13-1305

10.

Centers for Disease Control and Prevention. (2016). State and local public health actions to prevent and control diabetes, obesity, heart disease, diabetes, and associated risk factors and promote school health. National Center for Chronic Disease Prevention and Health Promotion. Retrieved from https://www.grants.gov/web/grants/search-grants.html?keywords=CDC-RFA-DP14-142202PPHF15PPHF2015

11.

Cha

A. E.

(2017, May 18). ‘An embarrassment’: U.S. health care far from the top in global study. The Washington Post. Retrieved from https://www.washingtonpost.com/news/to-your-health/wp/2017/05/18/an-embarrassment-u-s-health-care-far-from-the-top-in-global-study/?noredirect=on&utm_term=.9c77cc2f0179

12.

Considine

Lewis

(1999). Governance at ground level: The frontline bureaucrat in the age of markets and networks. Public Administration Review, 59, 467-480.

13.

Cross

Parker

(2004). The hidden power of social networks: Understanding how work really gets done in organizations. Boston, MA: Harvard Business School.

14.

Doreian

Woodard

(1992). Fixed list versus snowball selection of social networks. Social Science Research, 21, 216-233.

15.

Goggins

(1986). The “too few cases/too many variables” problem in implementation research. The Western Political Quarterly, 39, 328-347.

16.

Harris

J. K.

Leider

J. P.

Carothers

B. J.

Castrucci

B. C.

Hearne

(2016). Multisector health policy networks in 15 large cities. Journal of Public Health Management Practices, 22, 520-528.

17.

Hester

J. A.

Stange

P. V.

Seeff

L. C.

Davis

J. B.

Craft

C. A.

(2015). Toward sustainable improvements in population health: Overview of community integration structures and emerging innovations in financing (CDC Health Policy Series, No. 2). Atlanta, GA: Centers for Disease Control and Prevention.

18.

Hogg

R. A.

Varda

. (2016). Insights into collaborative networks of nonprofit, private, and public organizations that address complex health issues. Health Affairs, 25, 2014-2019.

19.

Hoppe

Reinelt

(2010). Social network analysis and the evaluation of leadership networks. The Leadership Quarterly, 21, 600-619.

20.

Huisman

(2009). Imputation of missing network data: Some simple procedures. Journal of Social Structure, 10, 1-29.

21.

Human

Provan

(2000). Legitimacy building in the evolution of small-firm multilateral networks: A comparative study of success and demise. Administrative Science Quarterly, 45, 327-365.

22.

Institute for Healthcare Improvement. (2018). The IHI triple aim: The Institute For Healthcare Improvement. Retrieved from http://www.ihi.org/Engage/Initiatives/TripleAim/Pages/default.aspx

23.

Imperial

(2005). Using collaboration as a governance strategy: Lessons from six watershed management programs. Administration & Society, 37, 281-320.

24.

Institute of Medicine. (2011). For the public’s health: Revitalizing law and policy to meet new challenges. Washington, DC: The National Academies Press.

25.

Institute of Medicine. (2012). For the public’s health: Investing in a healthier future. Washington, DC: The National Academies Press.

26.

Keast

Mandell

Brown

Woolcock

(2004). Network structures: Working differently and changing expectations. Public Administration Review, 64, 363-371.

27.

Keast

Mandell

Brown

(2006). Mixing state, market, and network governance models: The role of government in “crowded” policy domains. International Journal of Organization Theory and Behavior, 9, 27-50.

28.

Kenis

Provan

(2009). Towards and exogenous theory of public network performance. Public Administration, 87, 440-456.

29.

Kettl

(2000). The transformation of governance: Globalization, devolution, and the role of government. Public Administration Review, 60, 488-497.

30.

Kilduff

Tsai

(2003). Social networks and organizations. London, England: SAGE.

31.

Kindig

Stoddart

(2003). What is population health? American Journal of Public Health, 93, 380-383.

32.

Klijn

E.-H.

(2001). Rules as institutional context for decision making in networks: The approach to postwar housing districts in two cities. Administration & Society, 33, 133-164.

33.

Kossinets

(2006). Effects of missing data in social networks. Social Networks, 28, 247-268.

34.

McCullough

J. M.

Eisen-Cohen

Salas

S. B.

(2016). Partnership capacity for community health improvement plan implementation: Findings from a social network analysis. BMC Public Health, 16, Article 566.

35.

Miller

(1994). Post-progressive public administration: Lessons from policy networks. Public Administration Review, 54, 378-386.

36.

Milward

Provan

(2000). Governing the hollow state. Journal of Public Administration Research and Theory, 10, 359-379.

37.

National Association of County & City Health Officials. (2005). Operational definition of a functional local health department. Washington, DC: NACCHO.

38.

National Association of County & City Health Officials. (2016). Statement of policy: The local health department as community chief health strategist. Washington, DC: Author.

39.

National Association of County & City Health Officials. (2017). The changing public health landscape: Findings from the 2017 forces of change survey. Washington, DC: Author.

40.

O’Toole

(1997). “Treating networks seriously: Practical and research-based agendas in public administration. Public Administration Review, 57, 45-52.

41.

Powell

W. W.

(1990). Neither market nor hierarchy: Network forms of organization. In Staw

B. M.

Cummints

L. L.

(Eds.), Research in organizational behavior (Vol. 12, pp. 295–336). Greenwich, CT: JAI Press.

42.

Provan

Kenis

(2008). Modes of network governance: Structure, management, and effectiveness. Journal of Public Administration Research and Theory, 18, 229-252. doi:10.1093/jopart/mum015

43.

Raab

Milward

(2003). Dark networks as problems. Journal of Public Administration Research and Theory, 13, 413-439.

44.

Rhodes

R. A. W.

(1997). Understanding governance: Policy networks, governance, reflexivity and accountability. Philadelphia, PA: Open University Press.

45.

Robins

Pattison

Woolcock

(2004). Missing data in networks: Exponential random graph (p*) models for networks with non-respondents. Social Networks, 26, 257-283.

46.

Smith

Ceni

Milic-Frayling

Shneiderman

Mendes Rodrigues

Leskovec

. . . Dunne

(2010). NodeXL: A free and open network overview discovery and exploration add-in for Excel 2007/2010/2013/2016. Available from http://nodexl.codeplex.com/

47.

Stork

Richards

(1992). Nonrespondents in communication network studies. Group & Organization Management, 17, 193-209.

48.

Tandon

Murray

Lauer

J. A.

Evans

D. B.

(2000). Measuring overall health system performance for 191 countries (Global Programme on Evidence for Health Policy Discussion Paper No.30). Geneva, Switzerland: World Health Organization.

49.

von der Fehr

Sølberg

Bruun

. (2016). Validation of networks derived from snowball sampling of municipal science education actors. International Journal of Research & Method in Education, 41, 38-52.

50.

Wasserman

Faust

(1994). Social network analysis: Methods and applications. Cambridge, UK: Cambridge University Press.

51.

World Health Organization. (2000). The world health report 2000: Health systems: Improving performance. Geneva, Switzerland: Author.