- Split View
-
Views
-
Cite
Cite
Jörg Raab, Remco S. Mannak, Bart Cambré, Combining Structure, Governance, and Context: A Configurational Approach to Network Effectiveness, Journal of Public Administration Research and Theory, Volume 25, Issue 2, April 2015, Pages 479–511, https://doi.org/10.1093/jopart/mut039
- Share Icon Share
Abstract
This study explores the way in which network structure (network integration), network context (resource munificence and stability), and network governance mode relate to network effectiveness. The model by Provan and Milward (Provan, Keith G., and H. Brinton Milward. 1995. A preliminary theory of interorganizational network effectiveness: A comparative study of four community mental health systems. Administrative Science Quarterly 40 (1):1–33) on the effectiveness of designed and goal-directed interorganizational networks is extended and tested on the basis of 39 crime prevention networks (Safety Houses) in the Netherlands. Ten cases were subjected to in-depth analysis through documentation reviews, interviews, observations, and a survey among network participants. In the other 29 cases semistructured interviews were conducted with the network managers. The data for all 39 cases were analyzed with crisp-set Qualitative Comparative Analysis. The results revealed two different configurations for network effectiveness. Effective networks are centrally integrated networks that have been in existence for at least 3 years (age) and which show a high degree of stability. In addition, they either have considerable resources at their disposal or they have been set up with a network administrative organization. The results confirm core insights from Provan and Milward’s earlier study but also show that administrative resources can serve as a substitute for financial resources (and vice versa). The article concludes with suggestions for the further development of a configurational theory of network effectiveness.
Fundamental changes in technology and economic production and exchange, and growing interdependencies within and between societies in the past two decades (Castells 2000; Raab and Kenis 2009), have prompted attempts by many governments to enhance the effectiveness of public services by setting up or facilitating network-based interorganizational collaborations. As a consequence, horizontal, non-hierarchical forms of communication and cooperation across organizations have become more prevalent in the search for new and comprehensive solutions to societal problems (Klijn 2005). In this article, we analyze interorganizational networks, understood as deliberately formed groups of three or more autonomous but interdependent organizations—often providers of public services—that pursue a common goal and generate a collective output (Provan, Fish, and Sydow 2007).
There is, of course, nothing new about interorganizational networks and consortia. Nonetheless, there has been a considerable increase in the number and importance of networks in public management, healthcare, innovation, research and development, and the creative industries (Raab and Kenis 2009). It has been argued that the rise of this mode of governance in public management is attributable to a combination of factors (Isett et al. 2011). First, networks have emerged as a response to the constraints that the New Public Management reforms encountered in the marketization of public services. Longer-term and relatively stable relationships evolved between public agencies and private or non-profit organizations as a mechanism for the joint production of the services they provided. Second, complex, sometimes even ‘wicked’ problems can only be tackled via multiple organizations with access to diverse resources and expertise. Third, in Europe today and the wider world, policy problems are increasingly crossing territorial, political, and industrial boundaries, where there are no classical hierarchical institutions to address them. In addition, as Prahalad and Krishnan (2008) argue, private sector organizations are increasingly encountering consumers who demand tailor-made services, products, and solutions. Since access to resources is globalized, and since resources come from multiple organizational sources and not from one vertically integrated body, a network response is called for. This rationale similarly applies—albeit to a lesser extent—to citizens and their demands on public organizations: it is assumed that organizations that operate within networks are able to recombine resources continuously to satisfy the individualized demands of consumers and citizens.
Despite the increase in attention to interorganizational networks and to their study by scholars (Cropper et al. 2008) and despite the strong demand among academics and practitioners for knowledge on the effective design and governance of interorganizational networks, relatively few empirical studies on the effectiveness of public networks exist to date (see also Provan, Fish, and Sydow 2007; Turrini et al. 2010 for the most recent meta-analysis). Empirical studies that compare multiple networks are challenging to conduct. As a consequence, empirically tested knowledge that can be generalized more broadly is in short supply. In particular, most empirical studies of networks were conducted in the United States and do not exceed four to six networks.
In this article, we make three contributions to the development of a theory of network effectiveness. We begin by combining theoretical insights on structural and contextual conditions, drawn mainly from Provan and Milward (1995), with structural and functioning network characteristics (Turrini et al. 2010), especially mode of network governance as recently suggested by Provan and Kenis (2008). Second, we test this theoretical model against 39 crime prevention networks in the public sector in the Netherlands to ascertain how mode of governance is administered at the network level and how it affects the performance of the network as a whole at the community level. The sample is unique as it represents the entire population of a certain type of network in one country at a specific point in time. Since most of the empirical research on public service networks comes from the United States, we enhance the external validity of the core theoretical model on network effectiveness. Third, we apply a configurational approach and a relatively new analytical technique, Qualitative Comparative Analysis (QCA), to discern combinations of conditions that enable public networks to be effective. We contend that this is the appropriate theoretical and methodological approach for the further development of theory in this field. In a configurational approach, causal factors are not examined in isolation or in terms of their additive net effects, but are identified as necessary and sufficient conditions that collectively lead to a certain outcome (see Ragin 1997). The potential value of the configurational approach in the analysis of public sector networks is reflected in the growing attention it is receiving in conceptual and conference papers (Cristofoli, Maccio’, and Markovic 2012; Lemaire and Provan 2010; Raab, Lemaire, and Provan 2013).
In this study, interorganizational collaboration to reduce recidivism and prevent crime is coordinated in networks known as “Safety Houses” (SHs). It is expected that the frequent exchange of information and integration of the services of multiple law enforcement and social service organizations can reduce recidivism or, at least, the gravity of crimes committed by multiple offenders over time. The SHs represent a mix of organizations for the implementation of services and the diffusion of information. Thus, a SH is comprised of a network of organizations providing a related set of services to a common client group within a specific geographical area in the Netherlands. In this study, we test hypotheses on the combined effects of age, network structure (network integration), network context (resource munificence and stability), and formal mode of governance on network effectiveness. Data were collected from September 2009 to July 2010, with data collection for each network lasting about 1 month.
THEORY
The Basic Model
Despite large gaps in our knowledge of the governance and mechanics of networks (Provan, Fish, and Sydow 2007), some major scholarly efforts have appeared in the past two decades to substantiate claims that networks are gaining in importance and can be an effective form of governance (O’Toole 1997). Amid the ongoing discussion of the effectiveness of public sector networks, especially in the implementation of services, Provan and Milward (1995) were the first to conduct a systematic exploration of the factors that might play a role. They studied four mental healthcare networks in the United States and formulated several propositions, which together formed a preliminary theory of network effectiveness. This study can be regarded as seminal, given the very strong reception that it received within and outside the field of public management.1 However, despite frequent citations, there have been very few direct attempts to replicate it and refine or extend the original theoretical framework.
Recently, though, Turrini et al. (2010) performed a meta-analysis on empirical studies in public management that attempt to explain network effectiveness. Building on the framework of Provan and Milward (1995) and Provan and Sebastian (1998), they proposed two groups of independent network variables: structural characteristics and functioning characteristics. In addition, they identified contextual characteristics, both as independent and as moderating variables for the relationship between structural and functioning characteristics and network effectiveness. Structural characteristics include, for example, type and intensity of external control (Provan and Milward 1995), integration mechanisms and tools (Jennings and Ewalt 1998; Provan and Milward 1995), and the inner stability of a network (Ferlie and Pettigrew 1996). Functioning characteristics include variables such as steering the network processes (Agranoff and McGuire 2001; Klijn 1996) and traditional managerial work (Page 2003). Contextual characteristics consisted, amongst others, of the stability of the wider system in which the network is embedded or its level of resource munificence (Provan and Milward 1995).
The dependent variable “network effectiveness” can encompass very different outcomes depending on the tasks and goals of the network, the stakeholders, or the focus of the researcher. Turrini et al. (2010) list, for example, client-level effectiveness (Jennings and Ewalt 1998); network-level performance, such as the implementation of a performance measurement system (Van Raaij 2006); and community effectiveness (O’Toole and Meier 2004). Our research was based on Provan and Milward’s original framework, which explains network effectiveness with structural characteristics (centralized integration, external control) and contextual factors (system stability and resource munificence), but we altered and developed it in five ways.
First, we dropped external control, since the Netherlands is a unitary state, and control was much the same for all networks, that is, there was no variation on that factor.
Second, we added mode of governance as an explanatory factor, which falls in the category of “functioning characteristics” as suggested by Turrini et al. (2010). Since Williamson’s (1975) and Powell’s (1990) publications on network relationships, the discussion around networks as a mode of governance has, until very recently, focused mainly on the difference between this mode of governance and markets or hierarchies and the conditions under which it exists and functions. The reference point was, therefore, these two classic types of governance. As a result, networks were often treated as if they were all the same, despite a growing body of empirical evidence that showed that they come in different shapes and forms and have different modes of governance (Raab and Kenis 2009).
Consistent with this more recent view, Provan and Kenis (2008) suggest three ideal modes of network governance, which, they argue, can lead to effective outcomes depending on a number of contingencies: the self-governed network, where participating organizations jointly coordinate the activities; the lead organization model, in which the network is governed primarily by one organization, often the organization with the greatest vested interest in the network; and the network administration model, where a separate organization is set up to coordinate and facilitate the activities and to represent the network externally. As in our case, networks for the implementation of services are often deliberately created. The design, that is, the formal mode of governance, is systematically managed (Milward and Provan 2006) and is, therefore, a vital part of such networks.
Third, we introduce the configurational approach and QCA as the corresponding analytical technique (Fiss, Marx, and Cambré 2013; Ragin 1987) to the empirical analysis of and theory development on network effectiveness (see also Raab, Lemaire, and Provan 2013). A configurational perspective in public management and organization studies starts with the idea that “organizations are best understood as clusters of interconnected structures and practices, rather than as modular or loosely coupled entities whose components can be understood in isolation” (Fiss 2007, 1180). Consequently, this study responds to recent pleas in organization and management sciences for the application of configurational analysis instead of variance-based analysis (Fiss 2007, 2011; Kogut 2001).
QCA combines set theory and Boolean algebra and deviates from the perspective that “the more we observe variable X, the more we should observe outcome Y,” which tends to dominate current theorizing in organization and management studies. In this “net effects thinking” (Ragin 2008), “each independent variable is assumed to be capable of influencing the level or probability of the outcome regardless of the values or levels of other variables (i.e., regardless of the varied contexts defined by these variables)” (Ragin 2008, 177–78, emphasis in the original). In contrast, in the configurational perspective, contextual effects are the core issues. The effect of an individual characteristic often depends on the context, that is, the presence or absence of one or more (necessary or sufficient) conditions.
Fourth, we pick up the original (configurational) theoretical ideas of Provan and Milward (1995) and treat them in the appropriate manner both theoretically and methodologically. Provan and Milward (1995) already formulated their propositions on network effectiveness not in a linear additive way but in a configurational or set theoretic way. They refer explicitly, for example, to necessary and sufficient conditions and combine several factors into configurations that lead to network effectiveness as the outcome but then graphically represent the theoretical model in a linear additive fashion. We are not implying that they made an error. Basically, they followed the methodological standard at the time. We merely wish to emphasize that a configurational approach is and should be the core of a network theory of effectiveness; that is, as the complex interplay between necessary and sufficient conditions can explain the occurrence of outcomes such as high or low levels of effectiveness, data should be analyzed with corresponding set theoretic methods, such as QCA, instead of variance-based statistics. “Testing hypotheses” in this study, therefore, means constructing (non-probabilistic) evidence to either support or reject hypotheses derived mainly from Provan and Milward (1995) on necessary and sufficient conditions for network effectiveness(see Greckhamer, Misangyi, and Fiss 2013 for the underlying rational).
A configurational approach to network effectiveness, therefore, has important implications for the kind of theory we develop and for the type of methods that are applied. Hence, the configurational approach with QCA as an analytic tool was particularly useful in this study because (1) it allowed the comparison and analysis of a medium number of cases (39 in this study); (2) it offered an opportunity to compare configurations of variables (network age, system stability, network structure, resource munificence, and mode of governance) and to test for equifinality (more paths/configurations leading to the same result) (Fiss 2007); (3) it enabled the identification of necessary and sufficient conditions (Fiss 2007); and (4) it allowed the investigation of configurations leading to effective and ineffective outcomes, which is rare in research on public sector networks.
Fifth, we include network age as a factor. It has been argued (Kenis and Provan 2009, 445) that, after the initial formation, it may take 3–5 years for healthcare and social service networks to realize an improvement in outcomes for clients with serious illnesses (Annie E. Casey Foundation 1995). During the exploratory stage of our research, we noticed that networks take a long time to get organized and achieve visible results. We, therefore, included age as a factor with the idea of investigating how it plays out in conjunction with the other factors.
It is not our intention to ignore work that looks at other variables that influence network effectiveness, such as leadership (McGuire and Silvia 2009), or other process or behavioral factors, such as managing the unity-diversity tension in networks (Saz-Carranza and Ospina 2011). We do feel, however, that in our case the theoretical model suggested by Provan and Milward (1995) is the best starting point for the further development of a configurational theory of network effectiveness. This is because their model is based on systematic empirical research on whole networks in the healthcare and social service sector, because their study is widely regarded as seminal (Turrini et al. 2010) and because, in principle, they are already applying a configurational approach.
Network Effectiveness
Provan and Kenis (2008, 4) define network effectiveness as “the attainment of positive network-level outcomes (i.e. positive from a community stakeholder’s perspective, as discussed by Provan and Milward 2001) that could not normally be achieved by individual organizational participants acting independently.” From this perspective, the effectiveness of the network as a whole is more important than the effectiveness of individual organizations (Kenis and Provan 2009; Provan and Milward 1995), since networks can, for instance, improve the integration of critical services in the healthcare sector, enhance the community capacity to respond to natural disasters, cope with public problems such as crime, and stimulate regional economic development (Ansell 2000; Moynihan 2009).
As in the case of organizational effectiveness, there is no consensus on what network effectiveness is exactly and how to measure it. Looking at the arguments of Herranz (2010) and Kenis and Provan (2009), this does not come as a surprise. First, networks are multidimensional and can be analyzed at several levels. Second, assessment criteria are an element of value and not of fact (Simon 1976). As stated by Kenis and Provan (2009, 443), “Elements of value are implicit or explicit imperatives about the preferred state of a system that cannot be completely derived from the elements of fact; nor can they be proven empirically (Simon 1976)”; that is, ultimately, assessment criteria are by nature normative. Assessment criteria can, therefore, be determined by the researchers or by one or several stakeholders without one being scientifically superior to the other.
Some effectiveness criteria may, however, be more important than others in political or legal terms. Provan and Milward (2001) suggest that this dilemma can be mitigated by including the interests of different stakeholders at different levels of analysis. As a consequence, network effectiveness should be evaluated at three levels: community level, network level, and organizational/participant level. Turrini et al. (2010) share this distinction but place particular emphasis on the network level, with criteria such as “network capacity of achieving goals,” “network sustainability and viability,” and “network innovation and change” (546). However, a sole focus on the outcomes of individual organizations is generally regarded as insufficient for the evaluation of network performance. Networks must be built and maintained at all three levels, because network effectiveness at one level does not necessarily guarantee network effectiveness at all three (Provan and Milward 2001).
Though it is a commendable goal to assess every network simultaneously at all three levels, this is not always feasible. For example, network goals at community level might not be achievable at the point of measurement, or evaluation questions for all three levels might be too burdensome for the respondents, or the resources might be insufficient for collecting data for all three levels. We know of only one study that attempts to achieve a simultaneous assessment of all three levels and further develop the framework by Provan and Milward (2001). It was performed by Herranz (2010) on three multisectoral workforce development networks in Boston, Massachusetts, by linking network coordination orientation (bureaucratic, entrepreneurial, and community) with analyses at organizational, network, and community levels. Given a certain orientation, a network can then be evaluated on the basis of specific performance indicators for each level.
It seems that, so far, researchers have usually had to choose one level to focus on. If one has to choose, the community level is the best option, since the outcomes set the goal for public sector networks, and “overall network effectiveness will ultimately be judged by community-level stakeholders” (Provan and Milward 2001, 423). One could also argue that, for public sector networks, effectiveness at community level is the cumulative outcome of processes and results on the organizational and network levels.2
In their article “A preliminary theory of interorganizational network effectiveness” (1995), the point of departure for this study, Provan and Milward do not distinguish different levels of analysis for effectiveness. They mention the need to consider the views of different constituents but state that, “For this research, we felt it was critical to tie effectiveness measures to enhanced client well-being” (1995, 8). They apply “aggregate indicators of client well-being” (Provan and Milward 2001, 416), which can be regarded as a community-level outcome. We follow this approach, but instead of using aggregate indicators of client well-being, we focus on the changes in the incidence of a problem, another of the possible effectiveness criteria listed by Provan and Milward (2001), at the community level. In this study, we examine network effectiveness at community level, that is, in terms of the achievement of the national target for lowering recidivism. Not only is this as consistent as possible with the approach used by Provan and Milward (1995), but it is also appropriate, since it was the explicit goal for setting up SHs in the Netherlands. Besides, in this study, our primary interest is the network effects at the community level and not the organizational level. Additional information about effectiveness at the organizational level, as suggested by Herranz (2010), may, of course, enhance our understanding of the multidimensional and multilevel effects of network governance. But this goes beyond the scope of our article.
Below we explain the theoretical rationale for the relationships between the outcome variable, network effectiveness, and the different explanatory factors we measured.
HYPOTHESES
Network Age
It is usually assumed that networks for the implementation of services take some time to get started. After all, building trust between participants; deciding and implementing the design (governance mode); and developing rules for monitoring and control, accountability, and conflict resolution are long-term processes (Van Raaij 2006), especially in interorganizational networks, which are not built bottom-up on established interpersonal and professional ties. Time may also be a significant factor in networks that have to bridge cultural and legal gaps between players (e.g., social service and law enforcement agencies) or that need stable, institutionalized relationships and procedures (e.g., service implementation networks in the healthcare and social services) (Provan and Milward 1995). We, therefore, expect a minimum age to be a necessary but not sufficient condition; network effectiveness can never be explained by age alone. In the exploratory stages of our research, we found that 3 years was an important threshold. We, therefore, hypothesize that age (3 years) is a necessary but not sufficient condition for the effectiveness of SHs (H1).
System Stability (Context)
System stability concerns the number of changes in the network and the wider system and the prolonged activity of the network members. Provan and Milward (1995) argue that constant changes undermine network effectiveness because the energy, attention, and resources that members have to reinvest in building and rebuilding relationships and coordination procedures would otherwise be spent on the ongoing operational integration and coordination of services. System changes are another source of uncertainty, as Provan and Milward (1995) found for vulnerable client populations in particular. This also applies to a considerable extent to the SHs. People with psychological, addiction, or serious personal problems may feel confused and disillusioned by frequent changes to procedures, organizations, and contact persons. Provan and Milward (1995) also argue that system stability alone is not a sufficient condition, and that it has to be paired, for example, with high resource munificence. Following Provan and Milward (1995), we argue that system stability is a necessary but not sufficient condition for the effectiveness of SHs (H2).
Resource Munificence (Context)
Resource munificence concerns the amount of funding a SH receives from its members or external parties. Obviously, it is highly unlikely that goals will be reached without proper resources. However, as Provan and Milward (1995) argue, resource munificence is a necessary but not sufficient condition; in other words, effectiveness is difficult to achieve without adequate resources, but adequate resources do not automatically guarantee effectiveness. We, therefore, argue that resource munificence is a necessary but not sufficient condition for the effectiveness of SHs (H3).
Network Structure
Network structure concerns the degree of integration in the network. An integrated system is required to provide the network clients with a full range of non-overlapping services (Provan and Milward 1995). The integration of differentiated units as a necessary condition for goal achievement is a prominent argument in organization science in general and contingency theory in particular (Lawrence and Lorsch 1967). It was recently discussed by Saz-Carranza and Ospina (2011) in relation to public sector networks as the unity-diversity tension. There are three types of network integration in the literature on interorganizational networks: density-based integration, centralized integration, and integration through clique overlap. Network density describes “the general level of linkages among the points in a graph” (Scott 2000, 69). Network centralization describes “the extent to which this cohesion is organized around particular focal points” (Provan and Milward 1995, 10). In integration through clique overlap, effectiveness is enhanced when small cliques of agencies have overlapping linkages in different kinds of ties (Provan and Sebastian 1998).
Our research focuses on the extent to which networks are integrated through centralization (i.e., via a central organization) and not through density (i.e., via direct cooperation between all or a majority of the organizations) at the operational level, also referred to as task integration. First, as Provan and Milward (1995, 24) argue, centralized integration is beneficial for effectiveness, because it can facilitate both integration and coordination, which decentralized systems are less able to achieve because of multiple players and linkages. Second, once a system is centralized, the central agency is in a better position to monitor and control the activities of network members and prevent free riding. This is especially the case in service implementation networks, where goal-directed coordination is crucial. In addition, Provan and Milward (1995) maintain that network effectiveness is highest if integration takes place through centralization and not simultaneously through density, since this implies two possibly contradictory coordination logics. Such a dual integration makes systems “unnecessarily complex and inefficient,” because the energy, resources, and attention that go into building and maintaining redundant ties will adversely affect efficiency. Since the SHs are to a large extent service implementation networks for which centralized integration is assumed to be beneficial for effectiveness, we argue that centralized integration is a necessary but not sufficient condition for their effectiveness (H4).
Mode of Network Governance
Recently, Provan and Kenis (2008) suggested three ideal types of network governance, which they call governance modes: shared governance, governance by a lead organization, and governance by a network administrative organization (NAO). They argue that each mode is effective under different conditions. The mode of network governance is determined by two factors: brokered/non-brokered governance and participant/external governance (Provan and Kenis 2008). “Brokered” is the extent to which network governance is administered by and through a single organization (brokered) or by several or all organizations collectively (non-brokered). Participant/external governance refers to whether network governance is administered by and through organizations that participate in the network (participant-governed) or by a third—non-participating—party (external), which may be voluntarily set up by the participating organizations or mandated in the formation process (Provan and Kenis 2008).
In this study, only broker-governed networks were observed. Being brokered, the SH could either be formally managed by one or two network members or by a separate NAO that did not provide direct operational services for the clients or for law enforcement. All the networks in this study had an administrative entity that either fulfilled a neutral NAO role or functioned as an extension of one of the organizations in the network, the latter indicating governance by a lead organization. Moreover, the SHs, as is the case with many service implementation networks, consisted of a differentiated set of specialized organizations. This distinction is, perhaps, most apparent between organizations in the welfare sector and those in law enforcement. Although both are deemed to be necessary for the reduction of recidivism, their cooperation creates a specific tension between the cultures of care and repression. In case of lead organization governance, the lead organization may come from one of these two “camps,” creating possible problems for internal legitimacy and authority in coordinating the entire network. In this case, the lead organization might be seen as more inclined to either follow the logic of repression (law enforcement organizations) or of care (welfare organizations). We argue that a neutral NAO, which is not part of the primary process but independent in coordinating and monitoring network activities, is therefore better able to manage this tension. This will lead to greater integration and is, therefore, beneficial for overall network goal achievement.
The theoretical rationale for managing the tension between diversity and unity in networks through an NAO was confirmed in a recent study by Saz-Carranza and Ospina (2011). They showed that an NAO is advantageous to integrate, unite, and coordinate a diverse set of partners. In addition, Provan and Kenis (2008, 245) argued that an NAO is also advantageous for managing the network tensions of efficiency/inclusiveness and internal/external legitimacy. Although lead organization–governed networks tend to favor efficiency over inclusiveness, NAO-governed networks are likely to attempt to balance both. Again, this is very important in the SHs where two groups of organizations follow very different logics. An NAO provides the possibility to integrate both camps in the strategic decision making (inclusiveness) while performing operational activities relatively independently (efficiency).
The same logic holds for managing the tension between internal and external legitimacy. Although lead organization–governed networks will favor external legitimacy, NAO-governed networks, as Provan and Kenis (2008, 245) argue, will address both sides of the internal-external tension, but in a sequential manner. As networks in the public sector with government funding, SHs have to focus both on internal and external legitimacy, as they have to demonstrate their capacity both to network participants and outside stakeholders. Therefore, a sequential balancing seems likely to be beneficial to effectiveness. We, therefore, argue that being governed by an NAO is a necessary but not sufficient condition for the effectiveness of the SHs (H5).
A Configurational Model of Network Effectiveness
When we combine the necessary conditions of hypotheses 1–5 and the development of Provan and Milward’s basic framework (1995) into a configurational theory of network effectiveness, the following sufficient pathway emerges (H6): SHs that are at least 3 years old, have sufficient resources, are stable, are centrally integrated, and governed by an NAO will be effective (H6).
Overview of Hypotheses
H1: Age (in existence for at least 3 years) is a necessary but not sufficient condition for the effectiveness of SHs.
H2: System stability is a necessary but not sufficient condition for the effectiveness of SHs.
H3: Resource munificence is a necessary but not sufficient condition for the effectiveness of SHs.
H4: Centralized integration is a necessary but not sufficient condition for the effectiveness of SHs.
H5: Being governed by an NAO is a necessary but not sufficient condition for the effectiveness of the SHs.
H6: SHs that are at least 3 years old, have sufficient resources, are stable, are centrally integrated, and governed by an NAO will be effective.
METHODS
Research Setting
The empirical research was conducted in the field of crime prevention in the Netherlands. Thirty-nine SHs (Veiligheidshuizen) were studied. SHs can best be described as networks for service implementation and information diffusion (Milward and Provan 2006). Though the SHs focus strongly on information sharing between relevant organizations about the complex problems surrounding people at risk of committing crimes, they implement services through partner organizations, which are collectively selected and continuously coordinated in the network. The manager of a SH, therefore, is a network manager. In daily operations, the SHs consist of representatives of partner organizations in crime prevention (such as the police, the public prosecution service, rehabilitation agencies) and social services (such as mental healthcare agencies, local authorities, housing associations) (COT Instituut voor Veiligheids- en Crisismanagement 2008; Hulsen and Moors 2009; WODC and Adviesbureau Van Montfoort 2008).
The general aim of SHs is to reduce crime and improve public safety by focusing on four themes: youth, habitual offenders, domestic violence, and probation (COT Instituut voor Veiligheids- en Crisismanagement 2008; Ministerie van Justitie, Ministerie van Binnenlandse Zaken en Koninkrijksrelaties, and Projectdirectie Veiligheid begint bij Voorkomen 2008; Parket Generaal 2009; WODC and Adviesbureau Van Montfoort 2008). The first SH was established in Tilburg in 2002 after community leaders in the city had pointed out that the high rate of local juvenile delinquency could not be tackled by the respective organizations individually. As is fairly common in the delivery of public services in the Netherlands, the organizations in question joined forces in a consultative collective body to prevent conflicting approaches between crime prevention and social service agencies. The representatives of the partner organizations were then authorized by their own organizations to share their information and develop custom-made solutions to reduce public nuisance and criminality and to help their clients reintegrate in society.
Given the promising results of the first SH, the Dutch government promoted the establishment of SHs nationwide. Most SHs are, therefore, mandated, top-down networks with a relatively high level of external control. The term “external control” is applied here in the sense of Provan and Milward (1995), referring to the extent to which funding is direct and non-fragmented. SHs are publicly funded, mainly by the Ministry of Justice via the public prosecutor and local authorities (Centrum voor Criminaliteitspreventie en Veiligheid 2009b). The institutional context is the same for all SHs, and external control can be regarded as a constant, since the type of funding and the control by the Ministry of Justice and the local authorities fall under a national framework and are almost identical in each case. The level of funding across SHs differs, since municipalities are free to make additional investments. However, the failure of the Dutch government to provide a blueprint for the organization of SHs as networks (Ministerie van Justitie, Ministerie van Binnenlandse Zaken en Koninkrijksrelaties, and Projectdirectie Veiligheid begint bij Voorkomen 2008; WODC and Adviesbureau Van Montfoort 2008) has resulted in a notable variation in structure and mode of governance. Accordingly, SHs in the Netherlands provide a very interesting case for the investigation of whole goal-directed networks.
At the time of data collection, all the SHs had a fixed location for meetings, since it had emerged that SHs without a fixed location were plagued by collaboration problems and delivered disappointing results. The number of partners varied between 8 and 34 (mean = 18.79; standard deviation = 5.94). The variation in size did not necessarily mean variation in the diversity of the partner portfolio, since large SHs very often showed more overlap in their partner portfolios. Nonetheless, some differences in emphasis on either crime prevention or social services were observed.
Data Collection
To acquire in-depth insight into the cases, two rounds of data collection were held (see also Eisenhardt 1989). First, 10 SHs were studied via document reviews, semistructured and group interviews, observations, and a questionnaire. The documents covered all available paperwork, including annual reports, mission statements, business cases, and network evaluations. There were no annual reports for SHs that were still in the start-up phase. The network managers or coordinators, the municipal representatives and the public prosecutor (in their capacity as management committee members), and the chairpersons of the case consultation meetings were interviewed. Chairpersons and process managers (if applicable) were interviewed in a group setting. This had been specifically requested for efficiency reasons and information exchange. A topic list was used for all the interviews and the main theoretical concepts were operationalized as discussed below. The questionnaire was distributed to all network members, and several expert meetings were held to validate the preliminary results. If contradictory evidence emerged from two different sources, a third source was used for verification. We also considered the thoroughness of a source, the insights and involvement of a respondent, and potential political bias (some SH managers openly admitted that annual reports often paint a biased picture). The second round of data collection was for the other 29 SHs and took the form of semistructured interviews with the network managers or coordinators. In addition, a review was conducted of all available documents, such as annual reports and network evaluations, to ensure that the entire population of SHs in the Netherlands at that time was included in the analysis. An overview of the data collection is presented in table 1.
. | Round 1 . | Round 2 . |
---|---|---|
Documents studied | 89 | 49 |
Semistructured interviews | 27 | 29 |
Group interviews | 11 | |
Observations | 13 | |
Questionnaire | 75 | |
Expert meetings | 10 | 3 |
Time period of data collection | September 2009–June 2010 | June 2010–July 2010 |
N = 39 |
. | Round 1 . | Round 2 . |
---|---|---|
Documents studied | 89 | 49 |
Semistructured interviews | 27 | 29 |
Group interviews | 11 | |
Observations | 13 | |
Questionnaire | 75 | |
Expert meetings | 10 | 3 |
Time period of data collection | September 2009–June 2010 | June 2010–July 2010 |
N = 39 |
. | Round 1 . | Round 2 . |
---|---|---|
Documents studied | 89 | 49 |
Semistructured interviews | 27 | 29 |
Group interviews | 11 | |
Observations | 13 | |
Questionnaire | 75 | |
Expert meetings | 10 | 3 |
Time period of data collection | September 2009–June 2010 | June 2010–July 2010 |
N = 39 |
. | Round 1 . | Round 2 . |
---|---|---|
Documents studied | 89 | 49 |
Semistructured interviews | 27 | 29 |
Group interviews | 11 | |
Observations | 13 | |
Questionnaire | 75 | |
Expert meetings | 10 | 3 |
Time period of data collection | September 2009–June 2010 | June 2010–July 2010 |
N = 39 |
Concept Measurement
To test the hypotheses, the 39 SHs were compared by applying crisp-set QCA (csQCA; Marx, Cambré, and Rihoux 2013; Rihoux and Ragin 2009).
In brief, researchers use csQCA to build a dichotomous data table based on case knowledge derived from interviews, surveys, and documents. This table is then used to deduce a set of necessary and sufficient conditions for a certain outcome, in this case network effectiveness (Bakker et al. 2011; Fiss 2007).3
As we used csQCA, the categorization of cases was restricted to two options: a condition was either present (1) or absent (0). This meant that the researcher needed substantive knowledge to define meaningful and calibrated thresholds for the categorization of cases (Fiss 2007; Ragin 2000). To achieve reliable data dichotomization, in-depth insights into the individual cases were acquired via case analyses at multiple levels (case survey approach) (Yin 2003) and multiple data sources. We used csQCA because of theoretical arguments and to get a thorough understanding of the cases. Several conditions proved binary by nature: we observed, for instance, that system stability was either disturbed (e.g., by the drop-out of an important partner organization) or undisturbed. Since the scores were either clearly low or high for most conditions, a “camel shaped” (bi-modal) distribution and a single threshold emerged. The expert meetings did not induce us to finetune the categorization. In addition, csQCA provided outcomes that could be discussed in the expert meetings. In this way, we were able to test the validity of the results.
The measurement of the different concepts is discussed below. An in-depth description of the measurement and coding of network effectiveness is then provided in relation to the specific outcome of the SHs. Admittedly, the coding process is open to interpretation. We, therefore, used as much information as possible, applied a clear coding scheme, and based the coding decisions on transparent and justifiable thresholds for network effectiveness and system stability as explained below and listed in table 2 for each concept used in this study. In addition, though csQCA requires extensive data reduction, the other data are not “lost.” One of the key strengths of QCA is that data reduction is conducted to identify underlying mechanisms, but the results are then checked against and interpreted with the rich case knowledge that has been collected. In this study, after determining the results, we returned to the raw data and discussed single cases to ascertain whether the results made sense and could be explained in the light of our field experience, qualitative data, and the specific characteristics of the crime prevention networks.
Concept . | Indicator(s) . | Threshold Value . | Example . |
---|---|---|---|
Age | Number of years between the establishment of SH and the moment of data collection | 3 years | September 9, 2009. Opening Safety House The Hague. Today, Minister Hirsch Ballin, Deputy Minister Albayrak and Mayor Van Aartsen opened the Safety House in The Hague (Centrum voor Criminaliteitspreventie en Veiligheid 2009a). → coded as network age = 0 |
System stability | 1. Entrance or exit of important organizations to and from the network 2. Interruption of network coordination 3. Impact of (internal or external) changes in the network | Unstable if SH satisfies one or more indicators. | “You can state that we are currently going through turbulent times. Because of all kinds of circumstances things did not get off the ground in 2009, which was tied in with illness [of the manager] and I don’t know what else. So, did partner organizations drop out? Well, they dropped out emotionally” (manager SH). → coded as system stability = 0 |
Network structure: density | 1. Extent to which every network member has contacts with every other network member or participates in every case consultation meeting 2. Ratio between the total number of observed links and the total number of possible links. In 2-mode data, observed links E are a function of the number of members U per meeting and the number of meetings A, with and Density | Network density of 47% | “We have a small cluster for juvenile delinquents, for domestic violence, for nuisance. These organizations have little contact with other organizations in the SH […] but the social service institution has a broad network function and participated in all meetings.” → coded as network density = 0 |
Network structure: centralization | Presence of lynchpins, such as chairpersons or process managers, that coordinate the case and information flow between partner organizations at the operational level | Explicit identification of lynchpins | “[The process managers] are an important link between case consultation meetings of crime prevention agencies and social services. We’ve chosen to keep those meetings separate, because . . . social workers are not always willing to share information with the public prosecutor . . . the process manager can share information if needed . . . we are working on this case, but you have to consider the following developments” (manager SH). → coded as centralization = 1 |
Resource munificence | Financial, material, and personnel input, provided by members or external parties, divided by the number of inhabitants in the region of the SH (in budget and annual report) | €1.85 per capita | “Regarding [the year of data collection] we received 200,000 euros from the public prosecutor and the same from the municipality, so also 200,000 euros, and the police contributes 30,000 euros for the workplaces.” [total 430,000 euros, the authors]. → considering the number of inhabitants, coded as resource munificence = 1 |
Mode of network governance | Formal control with respect to network decision making, supervision, and management. All networks had an administrative entity that could either function as an extension of one of the organizations in the network, which indicates lead organization governance (0) or fulfill the neutral role of an external NAO that did not participate in the primary process (1) | Administrative entity in control of the network decision making, supervision, and management functions as a neutral NAO or an extension of the lead organization | “I [network manager] coordinate the network independently. . . . I work for all 23 organizations. The public prosecutor and municipality are the funders and have more influence than other partners. . . . but I draw up the plan and make all the proposals. . . . Partner organizations can submit proposals, but always after consulting me, to check they are suitable, and after approval by the board [of all partner organizations]” (manager SH). → coded as network governance = 1 (NAO) |
Network effectiveness | 1. Reduction of recidivism among clients of the SHs with 5.8% in a 2-year period 2. Reduction of criminality among SH clients with respect to four themes: youth, habitual offenders, domestic violence, and probation | Reduction of criminality and recidivism on at least 3 themes | 1. “The collaboration leads to more insight in the cases and can provide better custom-made solutions, and therefore, I suppose that it contributes to the reduction of recidivism . . . but there is no pure research on that yet”. → coded as network effectiveness = 0 2. “We had 1,100 cases per year [juvenile delinquents], which is now reduced to 600” (chairperson, case consultation meeting youth). → coded as network effectiveness = 1 on theme “youth.” |
Concept . | Indicator(s) . | Threshold Value . | Example . |
---|---|---|---|
Age | Number of years between the establishment of SH and the moment of data collection | 3 years | September 9, 2009. Opening Safety House The Hague. Today, Minister Hirsch Ballin, Deputy Minister Albayrak and Mayor Van Aartsen opened the Safety House in The Hague (Centrum voor Criminaliteitspreventie en Veiligheid 2009a). → coded as network age = 0 |
System stability | 1. Entrance or exit of important organizations to and from the network 2. Interruption of network coordination 3. Impact of (internal or external) changes in the network | Unstable if SH satisfies one or more indicators. | “You can state that we are currently going through turbulent times. Because of all kinds of circumstances things did not get off the ground in 2009, which was tied in with illness [of the manager] and I don’t know what else. So, did partner organizations drop out? Well, they dropped out emotionally” (manager SH). → coded as system stability = 0 |
Network structure: density | 1. Extent to which every network member has contacts with every other network member or participates in every case consultation meeting 2. Ratio between the total number of observed links and the total number of possible links. In 2-mode data, observed links E are a function of the number of members U per meeting and the number of meetings A, with and Density | Network density of 47% | “We have a small cluster for juvenile delinquents, for domestic violence, for nuisance. These organizations have little contact with other organizations in the SH […] but the social service institution has a broad network function and participated in all meetings.” → coded as network density = 0 |
Network structure: centralization | Presence of lynchpins, such as chairpersons or process managers, that coordinate the case and information flow between partner organizations at the operational level | Explicit identification of lynchpins | “[The process managers] are an important link between case consultation meetings of crime prevention agencies and social services. We’ve chosen to keep those meetings separate, because . . . social workers are not always willing to share information with the public prosecutor . . . the process manager can share information if needed . . . we are working on this case, but you have to consider the following developments” (manager SH). → coded as centralization = 1 |
Resource munificence | Financial, material, and personnel input, provided by members or external parties, divided by the number of inhabitants in the region of the SH (in budget and annual report) | €1.85 per capita | “Regarding [the year of data collection] we received 200,000 euros from the public prosecutor and the same from the municipality, so also 200,000 euros, and the police contributes 30,000 euros for the workplaces.” [total 430,000 euros, the authors]. → considering the number of inhabitants, coded as resource munificence = 1 |
Mode of network governance | Formal control with respect to network decision making, supervision, and management. All networks had an administrative entity that could either function as an extension of one of the organizations in the network, which indicates lead organization governance (0) or fulfill the neutral role of an external NAO that did not participate in the primary process (1) | Administrative entity in control of the network decision making, supervision, and management functions as a neutral NAO or an extension of the lead organization | “I [network manager] coordinate the network independently. . . . I work for all 23 organizations. The public prosecutor and municipality are the funders and have more influence than other partners. . . . but I draw up the plan and make all the proposals. . . . Partner organizations can submit proposals, but always after consulting me, to check they are suitable, and after approval by the board [of all partner organizations]” (manager SH). → coded as network governance = 1 (NAO) |
Network effectiveness | 1. Reduction of recidivism among clients of the SHs with 5.8% in a 2-year period 2. Reduction of criminality among SH clients with respect to four themes: youth, habitual offenders, domestic violence, and probation | Reduction of criminality and recidivism on at least 3 themes | 1. “The collaboration leads to more insight in the cases and can provide better custom-made solutions, and therefore, I suppose that it contributes to the reduction of recidivism . . . but there is no pure research on that yet”. → coded as network effectiveness = 0 2. “We had 1,100 cases per year [juvenile delinquents], which is now reduced to 600” (chairperson, case consultation meeting youth). → coded as network effectiveness = 1 on theme “youth.” |
Concept . | Indicator(s) . | Threshold Value . | Example . |
---|---|---|---|
Age | Number of years between the establishment of SH and the moment of data collection | 3 years | September 9, 2009. Opening Safety House The Hague. Today, Minister Hirsch Ballin, Deputy Minister Albayrak and Mayor Van Aartsen opened the Safety House in The Hague (Centrum voor Criminaliteitspreventie en Veiligheid 2009a). → coded as network age = 0 |
System stability | 1. Entrance or exit of important organizations to and from the network 2. Interruption of network coordination 3. Impact of (internal or external) changes in the network | Unstable if SH satisfies one or more indicators. | “You can state that we are currently going through turbulent times. Because of all kinds of circumstances things did not get off the ground in 2009, which was tied in with illness [of the manager] and I don’t know what else. So, did partner organizations drop out? Well, they dropped out emotionally” (manager SH). → coded as system stability = 0 |
Network structure: density | 1. Extent to which every network member has contacts with every other network member or participates in every case consultation meeting 2. Ratio between the total number of observed links and the total number of possible links. In 2-mode data, observed links E are a function of the number of members U per meeting and the number of meetings A, with and Density | Network density of 47% | “We have a small cluster for juvenile delinquents, for domestic violence, for nuisance. These organizations have little contact with other organizations in the SH […] but the social service institution has a broad network function and participated in all meetings.” → coded as network density = 0 |
Network structure: centralization | Presence of lynchpins, such as chairpersons or process managers, that coordinate the case and information flow between partner organizations at the operational level | Explicit identification of lynchpins | “[The process managers] are an important link between case consultation meetings of crime prevention agencies and social services. We’ve chosen to keep those meetings separate, because . . . social workers are not always willing to share information with the public prosecutor . . . the process manager can share information if needed . . . we are working on this case, but you have to consider the following developments” (manager SH). → coded as centralization = 1 |
Resource munificence | Financial, material, and personnel input, provided by members or external parties, divided by the number of inhabitants in the region of the SH (in budget and annual report) | €1.85 per capita | “Regarding [the year of data collection] we received 200,000 euros from the public prosecutor and the same from the municipality, so also 200,000 euros, and the police contributes 30,000 euros for the workplaces.” [total 430,000 euros, the authors]. → considering the number of inhabitants, coded as resource munificence = 1 |
Mode of network governance | Formal control with respect to network decision making, supervision, and management. All networks had an administrative entity that could either function as an extension of one of the organizations in the network, which indicates lead organization governance (0) or fulfill the neutral role of an external NAO that did not participate in the primary process (1) | Administrative entity in control of the network decision making, supervision, and management functions as a neutral NAO or an extension of the lead organization | “I [network manager] coordinate the network independently. . . . I work for all 23 organizations. The public prosecutor and municipality are the funders and have more influence than other partners. . . . but I draw up the plan and make all the proposals. . . . Partner organizations can submit proposals, but always after consulting me, to check they are suitable, and after approval by the board [of all partner organizations]” (manager SH). → coded as network governance = 1 (NAO) |
Network effectiveness | 1. Reduction of recidivism among clients of the SHs with 5.8% in a 2-year period 2. Reduction of criminality among SH clients with respect to four themes: youth, habitual offenders, domestic violence, and probation | Reduction of criminality and recidivism on at least 3 themes | 1. “The collaboration leads to more insight in the cases and can provide better custom-made solutions, and therefore, I suppose that it contributes to the reduction of recidivism . . . but there is no pure research on that yet”. → coded as network effectiveness = 0 2. “We had 1,100 cases per year [juvenile delinquents], which is now reduced to 600” (chairperson, case consultation meeting youth). → coded as network effectiveness = 1 on theme “youth.” |
Concept . | Indicator(s) . | Threshold Value . | Example . |
---|---|---|---|
Age | Number of years between the establishment of SH and the moment of data collection | 3 years | September 9, 2009. Opening Safety House The Hague. Today, Minister Hirsch Ballin, Deputy Minister Albayrak and Mayor Van Aartsen opened the Safety House in The Hague (Centrum voor Criminaliteitspreventie en Veiligheid 2009a). → coded as network age = 0 |
System stability | 1. Entrance or exit of important organizations to and from the network 2. Interruption of network coordination 3. Impact of (internal or external) changes in the network | Unstable if SH satisfies one or more indicators. | “You can state that we are currently going through turbulent times. Because of all kinds of circumstances things did not get off the ground in 2009, which was tied in with illness [of the manager] and I don’t know what else. So, did partner organizations drop out? Well, they dropped out emotionally” (manager SH). → coded as system stability = 0 |
Network structure: density | 1. Extent to which every network member has contacts with every other network member or participates in every case consultation meeting 2. Ratio between the total number of observed links and the total number of possible links. In 2-mode data, observed links E are a function of the number of members U per meeting and the number of meetings A, with and Density | Network density of 47% | “We have a small cluster for juvenile delinquents, for domestic violence, for nuisance. These organizations have little contact with other organizations in the SH […] but the social service institution has a broad network function and participated in all meetings.” → coded as network density = 0 |
Network structure: centralization | Presence of lynchpins, such as chairpersons or process managers, that coordinate the case and information flow between partner organizations at the operational level | Explicit identification of lynchpins | “[The process managers] are an important link between case consultation meetings of crime prevention agencies and social services. We’ve chosen to keep those meetings separate, because . . . social workers are not always willing to share information with the public prosecutor . . . the process manager can share information if needed . . . we are working on this case, but you have to consider the following developments” (manager SH). → coded as centralization = 1 |
Resource munificence | Financial, material, and personnel input, provided by members or external parties, divided by the number of inhabitants in the region of the SH (in budget and annual report) | €1.85 per capita | “Regarding [the year of data collection] we received 200,000 euros from the public prosecutor and the same from the municipality, so also 200,000 euros, and the police contributes 30,000 euros for the workplaces.” [total 430,000 euros, the authors]. → considering the number of inhabitants, coded as resource munificence = 1 |
Mode of network governance | Formal control with respect to network decision making, supervision, and management. All networks had an administrative entity that could either function as an extension of one of the organizations in the network, which indicates lead organization governance (0) or fulfill the neutral role of an external NAO that did not participate in the primary process (1) | Administrative entity in control of the network decision making, supervision, and management functions as a neutral NAO or an extension of the lead organization | “I [network manager] coordinate the network independently. . . . I work for all 23 organizations. The public prosecutor and municipality are the funders and have more influence than other partners. . . . but I draw up the plan and make all the proposals. . . . Partner organizations can submit proposals, but always after consulting me, to check they are suitable, and after approval by the board [of all partner organizations]” (manager SH). → coded as network governance = 1 (NAO) |
Network effectiveness | 1. Reduction of recidivism among clients of the SHs with 5.8% in a 2-year period 2. Reduction of criminality among SH clients with respect to four themes: youth, habitual offenders, domestic violence, and probation | Reduction of criminality and recidivism on at least 3 themes | 1. “The collaboration leads to more insight in the cases and can provide better custom-made solutions, and therefore, I suppose that it contributes to the reduction of recidivism . . . but there is no pure research on that yet”. → coded as network effectiveness = 0 2. “We had 1,100 cases per year [juvenile delinquents], which is now reduced to 600” (chairperson, case consultation meeting youth). → coded as network effectiveness = 1 on theme “youth.” |
Outcome Variable: Network Effectiveness
As discussed above, in this study network effectiveness was determined at the community level. To quote Provan and Milward (2001, 416): “Community-based networks must be judged by the contribution they make to the communities they are trying to serve.” Since the SHs were set up to reduce recidivism in the communities they covered geographically, we felt that the appropriate indicators were perspectives on the reduction of recidivism, which represented “changes in the incidence of the problem” (Provan and Milward 2001, 416) as the effectiveness criterion (Kenis and Provan 2009).
Respondents were first asked whether their SH had achieved a 5.8% reduction in recidivism in a 2-year period. This corresponded to a 25% reduction in 7 years. Both percentages were set as nationwide targets by the Dutch Ministry of Justice (Ministerie van Justitie and Ministerie van Binnenlandse Zaken en Koninkrijksrelaties 2007). Respondents based their answers on the number of cases handled by their SH in which new crimes had been committed in the past 2 years. Respondents usually stated either that they could not show any reduction (yet) (coded as not effective, 0) or that the reduction was significantly higher than the 5.8% (effective, 1). For example, statements such as “We are currently working on the formulation of goals. I really don’t know if it is resulting in [the reduction of recidivism.] It feels good, it has to result in success, but then I am parroting the experiences of other SHs” (manager SH), or “There’s a lot to be done, . . . At this stage, I am contacting partners, making appointments, and collecting money” (manager SH) were coded as low community-level effectiveness (= 0). If there was evidence of a reduction, respondents were asked to expatiate for the four themes in which SHs were expected to be active: (1) youth; (2) habitual offenders; (3) domestic violence; and (4) probation (WODC and Adviesbureau Van Montfoort 2008). Statements such as “We have about 125 highly active habitual offenders . . . we made about 40 probation agreements [with habitual offenders]. During the past two years, less than five [habitual offenders] violated the agreement, which means that once they did not stick to it. Therefore, [with respect to habitual offenders] the goal is amply achieved” (manager SH) were coded as effective (= 1) for “habitual offenders.” And statements such as “We had 1,100 cases per year [juvenile delinquents], which is now reduced to 600” (chairman, case consultation meeting youth) were coded as effective (= 1) for the “youth” theme.
If a SH was effective in at least three of the four themes, that is, had a score of 3 or 4, it was counted as being overall effective (threshold) and coded as 1. SHs that were effective in three or four themes provided evidence for a strong overall reduction in criminality and showed an overall reduction in recidivism that exceeded the national target of 5.8% in 2 years. Respondents were not only consulted about effectiveness in individual themes (based on multiple cases) but also about how this sums up to overall effectiveness. The analysis revealed that SHs that were effective in two themes or less lacked insight into the overall reduction of recidivism or did not attain the national target. These SHs were considered to be ineffective and coded as 0. This binary distinction between “effective” and “ineffective” is due to the crisp set approach but also based on substantive qualitative information; that is, the networks qualified as “ineffective” are clearly more ineffective than effective, as viewed by us and our respondents. As one manager of a SH put it, “With respect to habitual offenders the results are good. . . . But if you refer to the national target, I cannot say if we achieve that, that is just too early.”.
Some of these SHs also demonstrated an improvement in the sense of safety among citizens by producing independent reports. Accordingly, SHs that were effective in three or more themes met the community-level goal. Table 3 shows the distribution of SHs according to the number of themes. Ten out of 39 SHs (26%) were coded as overall effective.
Effective on N Themes . | 0 . | 1 . | 2 . | 3 . | 4 . | |||
---|---|---|---|---|---|---|---|---|
10 | 12 | 7 | 8 | 2 | ||||
26% | 31% | 18% | 21% | 5% | ||||
N = 39 | Mean 1.49 | SD 1.23 | Min 0 | Max 4 |
Effective on N Themes . | 0 . | 1 . | 2 . | 3 . | 4 . | |||
---|---|---|---|---|---|---|---|---|
10 | 12 | 7 | 8 | 2 | ||||
26% | 31% | 18% | 21% | 5% | ||||
N = 39 | Mean 1.49 | SD 1.23 | Min 0 | Max 4 |
Effective on N Themes . | 0 . | 1 . | 2 . | 3 . | 4 . | |||
---|---|---|---|---|---|---|---|---|
10 | 12 | 7 | 8 | 2 | ||||
26% | 31% | 18% | 21% | 5% | ||||
N = 39 | Mean 1.49 | SD 1.23 | Min 0 | Max 4 |
Effective on N Themes . | 0 . | 1 . | 2 . | 3 . | 4 . | |||
---|---|---|---|---|---|---|---|---|
10 | 12 | 7 | 8 | 2 | ||||
26% | 31% | 18% | 21% | 5% | ||||
N = 39 | Mean 1.49 | SD 1.23 | Min 0 | Max 4 |
Explanatory Factors
First, system stability was operationalized as the entrance or exit of important organizations to and from the network, the permanence of network coordination, and the impact of internal or external changes on the network. System stability was coded as 0 (unstable) if one or more indicators showed clear changes (threshold). In other words, SHs were considered to be unstable if important organizations (municipality, public prosecutor, or police) exited or entered the network during the period of observation, if network coordination was interrupted (e.g., long-term absence of SH manager), or if respondents reported that internal or external changes had impacted on the network. If none of these indicators were present, system stability was coded as 1 (stable).
The reason behind this coding is that these three events are likely to hamper the operations of an SH. One SH manager said, for example, “You can say that we are currently undergoing turbulent times. Due to all kinds of circumstances things did not get off the ground in 2009, circumstances like illness [of the manager] and I don’t know what else. So, did partner organizations drop out? Well, they dropped out emotionally.” This statement prompted us to qualify the network as unstable (0). In another case, the manager said that “an important change was the withdrawal of the public prosecutor . . . that took half a year . . .. As the public prosecutor is in control, this situation gravely hampered the operations of the SH.” In this case, the system was also categorized as unstable (0). In a third case, the SH manager illustrated the unstable (0) period as follows: “Many changes, definitely. In this start-up phase, you are busy with everything and everything changes. Recently, we for instance moved to a new property. It was a tremendously turbulent period.”
Second, network age was defined as the period between the establishment of the SH and the moment of data collection. The observed networks had existed for an average of 2.3 years. However, in-depth case knowledge demonstrated that at least 3 years are needed to build a sustainable collaboration, especially if inherent functional tensions exist between the participating organizations. In the SHs we found an inherent tension between law enforcement and welfare functions. It is, therefore, necessary to build sufficient trust between organizations as well as individuals. An age of less than 3 years was coded as low (0), and 3 years or more was coded as high (1).
Third, resource munificence was investigated by calculating the ratio between the total resources of the SHs (financial, material, and personnel input, provided by members or external parties) and the total population in the region. Consequently, comparable figures for per capita spending were acquired. On the basis of in-depth case knowledge, €1.85 per capita was established as the cut-off point between low (0) and high (1) resource munificence. One of the SH managers said: “We have a budget of about €400,000, co-financed by the Ministry of Justice at about €200,000 and the municipality at about €200,000 [confirmed by budget estimate].” As this SH served a region with a population of approximately 94,000 inhabitants, the resource munificence (€4.26 per capita) was coded as high (1). Another SH with approximately the same budget for a region with 550,000 inhabitants was coded as low (0) for resource munificence.
Fourth, the mode of governance was ascertained by determining who was formally in control of the network decision making, supervision, and management. Since no shared governance was observed, the networks could either be governed by one or two network members or by a separate NAO that did not provide direct operational services for the clients or law enforcement measures. All the networks in this study had an administrative entity that could either fulfill a neutral NAO role (coded as 1) or function as an extension of one of the organizations in the network, thus indicating lead organization governance (coded as 0). In one example of a lead organization network, the municipal authority took the leading role and the SH functioned mainly as an extension of the municipality. The municipality participated in almost all the case consultation meetings (12 out of 14), usually as chair (8). Further, the municipality was closely involved in strategic actions, besides providing the bulk of the SH budget and employing the network manager.
Finally, the network structure of the SH was studied. Provan and Milward (1995) found that centralized integration was likely to enhance network effectiveness. However, networks that are both centrally and decentrally integrated are less likely to be effective than centrally integrated networks. As discussed above, centralization can facilitate the coordination and control of network collaboration, but this can be hampered by the complexities arising from the multiple relationships in dense networks. This research, therefore, focused on the extent to which networks are integrated through centralization, but not (simultaneously) through density. Centralization was determined based on whether process managers existed connecting the different partner organizations across case consultation meetings, since the existence of such process managers indicates a centralized structure. Density was determined based on whether all organizations jointly participated in the case consultation meetings. The more organizations that participated jointly and more frequently in these meetings, the denser the structure. Therefore, in principle the structure can be both centralized and dense, if a process manager existed and many organizations frequently jointly participated in case consultation meetings. Accordingly, networks can either be merely centrally integrated (coded 1) or not integrated at all (coded 0), or merely density-based integrated (coded 0) or simultaneously density-based and centrally integrated (coded 0). We only coded central integration as 1, since Provan and Milward (1995) argued that effectiveness will primarily be achieved through central integration but not simultaneously through central- and density-based integration.
Centralization was determined by in-depth information on mechanisms that represent centralization, such as the de facto presence of chairpersons, process managers, and lynchpins in the case consultation meetings. Highly centralized SHs would, for example, be characterized by process managers who select and transfer information to and from partner organizations. It must be noted that the SHs are engineered networks, where in some instances process managers are hired and in others not. Process managers have very frequent contact with the partner organizations and collect and distribute information among them in a SH. Therefore, the use of process managers de facto centralizes the operational communication structure.
Network density was based on the presence of partner organizations in case consultation meetings. Density is usually calculated as the number of existing ties between actors in a network divided by the number of all potential ties. In our case, the number of existing ties is estimated by the average number of participating organizations per meeting and the total number of meetings. The number of potential ties is a function of the total number of organizations in the network. The reliability of the estimated network density was confirmed by in-depth information. For example, we asked the network managers whether members collaborated in small subgroups or whether every member participated in every meeting, thus leading to a density of 100%. As with centralization, we validated the proxy by comparing the figures calculated on the basis of the standard density formula with in-depth qualitative information for the first sample of 10 cases.
Consequently, the measurement of the centralized integration of the networks was based on both quantitative measures and qualitative information on network density and the level of centralization. Table 4 shows the outcome of the coding process in the form of the dichotomized variable scores and their distribution.
Concept . | 0 Score . | 1 Score . |
---|---|---|
Age (below 3 years = 0; equal to/above 3 years = 1) | 26 (66.7%) | 13 (33.3%) |
System stability (unstable = 0; stable = 1) | 16 (41.0%) | 23 (59.0%) |
Network structure (no centralized integration = 0; centralized integration = 1) | 20 (51.3%) | 19 (48.7%) |
Resource munificence (low = 0; high = 1) | 18 (46.2%) | 21 (53.8%) |
Mode of network governance (lead = 0; NAO = 1) | 30 (76.9%) | 9 (23.1%) |
Network effectiveness (low = 0; high = 1) | 29 (74.4%) | 10 (25.6%) |
N = 39 (100%) |
Concept . | 0 Score . | 1 Score . |
---|---|---|
Age (below 3 years = 0; equal to/above 3 years = 1) | 26 (66.7%) | 13 (33.3%) |
System stability (unstable = 0; stable = 1) | 16 (41.0%) | 23 (59.0%) |
Network structure (no centralized integration = 0; centralized integration = 1) | 20 (51.3%) | 19 (48.7%) |
Resource munificence (low = 0; high = 1) | 18 (46.2%) | 21 (53.8%) |
Mode of network governance (lead = 0; NAO = 1) | 30 (76.9%) | 9 (23.1%) |
Network effectiveness (low = 0; high = 1) | 29 (74.4%) | 10 (25.6%) |
N = 39 (100%) |
Concept . | 0 Score . | 1 Score . |
---|---|---|
Age (below 3 years = 0; equal to/above 3 years = 1) | 26 (66.7%) | 13 (33.3%) |
System stability (unstable = 0; stable = 1) | 16 (41.0%) | 23 (59.0%) |
Network structure (no centralized integration = 0; centralized integration = 1) | 20 (51.3%) | 19 (48.7%) |
Resource munificence (low = 0; high = 1) | 18 (46.2%) | 21 (53.8%) |
Mode of network governance (lead = 0; NAO = 1) | 30 (76.9%) | 9 (23.1%) |
Network effectiveness (low = 0; high = 1) | 29 (74.4%) | 10 (25.6%) |
N = 39 (100%) |
Concept . | 0 Score . | 1 Score . |
---|---|---|
Age (below 3 years = 0; equal to/above 3 years = 1) | 26 (66.7%) | 13 (33.3%) |
System stability (unstable = 0; stable = 1) | 16 (41.0%) | 23 (59.0%) |
Network structure (no centralized integration = 0; centralized integration = 1) | 20 (51.3%) | 19 (48.7%) |
Resource munificence (low = 0; high = 1) | 18 (46.2%) | 21 (53.8%) |
Mode of network governance (lead = 0; NAO = 1) | 30 (76.9%) | 9 (23.1%) |
Network effectiveness (low = 0; high = 1) | 29 (74.4%) | 10 (25.6%) |
N = 39 (100%) |
Findings
We used Tosmana 1.3.1 to perform the csQCA analysis. This helped us to identify the configurations for network effectiveness:4
The above formula should be interpreted as follows. Each letter denotes a factor in the study (A = age; S = system stability; I = centralized integration; R = resource munificence; G = NAO governance; E = effectiveness). The symbol “+” denotes the logical operator “or”; “∙” denotes the logical operator “and”; “~” denotes the logical operator “not”; and “→” denotes the logical implication operator (see Fiss 2007). Interpreting the formula accordingly, it can be deduced that networks that combine age (3-year existence) with high levels of system stability and centralized integration and high resource munificence will show high network effectiveness. The second pathway shows that a 3-year existence with high levels of system stability, centralized integration, and NAO governance also leads to high network effectiveness.
This solution for network effectiveness was reached without logical remainders, that is, configurations that are not empirically observed but which can be used to achieve a more parsimonious solution (Rihoux and De Meur 2009).
As figure 1 shows, there is a large “reservoir” of logical remainders. This is because only a fraction of the possible logical configurations is represented by empirical cases. Only 15 of the 32 potential configurations (25, as there are five conditions) correspond to observed cases. Hence, the 17 logical remainders can be used to reduce complexity further and to produce a more parsimonious solution (for a more detailed description of the procedure for including logical remainders, see Rihoux and De Meur 2009).
Note: White: unobserved combinations. Hatched: observed combinations that result in low network effectiveness. Grey: observed combinations that result in high network effectiveness.
For network effectiveness, the configurations including logical remainders are:
Closer examination of the solution, including logical remainders, shows, however, that these unobserved configurations are most likely to lead to network ineffectiveness. Figure 1 shows that all logical remainders are configurations that include low system stability and/or low age and/or which are not centrally integrated. All observed networks that satisfy one of these conditions are ineffective. Therefore, for network effectiveness, we built on configurations without logical remainders.
Consistency
Consistency “assesses the degree to which the cases sharing a given condition or combination of conditions . . . agree in displaying the outcome in question” (Ragin 2006, 292). The aim is to obtain high consistency measures, thereby indicating that a high proportion of cases with a given cause or combination of causes will display the same outcome. The presence of contradictory configurations (the same configuration resulting in both the presence and absence of the outcome) lowers the consistency scores. In small- and medium-sized N csQCA such as the one presented here, the consistency should be close to 1. Since no contradictions occurred in the combinations of configurations leading to an effective outcome as presented in the truth table, the consistency is 1.0, which signals a high validity for the identified causal combinations.
Coverage
Coverage is an assessment of the way the terms of the minimal formulae “cover” observed cases. It is a measure of the “fit” of the model and the reliability of the results (Rihoux and Ragin 2009).
The investigated model contained five binary conditions, so 32 (=25) possible configurations could be created. However, as discussed above, it is unlikely that all configurations will exist in an empirical setting. The truth table for this study (table 5) shows all possible configurations for effective and ineffective outcomes with conditions and outcome defined as present (“Yes”), absent (“No”), or not observed (?).
No . | Conditions . | Outcome . | ||||
---|---|---|---|---|---|---|
A . | S . | I . | R . | G . | E . | |
Age . | System Stability . | Network Structure . | Resource Munificence . | Network Governance . | Network Effectiveness . | |
1 | No | No | No | No | No | No |
2 | No | No | No | No | Yes | No |
3 | No | No | No | Yes | No | No |
4 | No | No | No | Yes | Yes | ? |
5 | No | No | Yes | No | No | No |
6 | No | No | Yes | No | Yes | ? |
7 | No | No | Yes | Yes | No | No |
8 | No | No | Yes | Yes | Yes | No |
9 | No | Yes | No | No | No | No |
10 | No | Yes | No | No | Yes | ? |
11 | No | Yes | No | Yes | No | No |
12 | No | Yes | No | Yes | Yes | ? |
13 | No | Yes | Yes | No | No | No |
14 | No | Yes | Yes | No | Yes | ? |
15 | No | Yes | Yes | Yes | No | No |
16 | No | Yes | Yes | Yes | Yes | ? |
17 | Yes | No | No | No | No | ? |
18 | Yes | No | No | No | Yes | ? |
19 | Yes | No | No | Yes | No | ? |
20 | Yes | No | No | Yes | Yes | ? |
21 | Yes | No | Yes | No | No | ? |
22 | Yes | No | Yes | No | Yes | ? |
23 | Yes | No | Yes | Yes | No | ? |
24 | Yes | No | Yes | Yes | Yes | ? |
25 | Yes | Yes | No | No | No | ? |
26 | Yes | Yes | No | No | Yes | ? |
27 | Yes | Yes | No | Yes | No | No |
28 | Yes | Yes | No | Yes | Yes | ? |
29 | Yes | Yes | Yes | No | No | No |
30 | Yes | Yes | Yes | No | Yes | Yes |
31 | Yes | Yes | Yes | Yes | No | Yes |
32 | Yes | Yes | Yes | Yes | Yes | Yes |
No . | Conditions . | Outcome . | ||||
---|---|---|---|---|---|---|
A . | S . | I . | R . | G . | E . | |
Age . | System Stability . | Network Structure . | Resource Munificence . | Network Governance . | Network Effectiveness . | |
1 | No | No | No | No | No | No |
2 | No | No | No | No | Yes | No |
3 | No | No | No | Yes | No | No |
4 | No | No | No | Yes | Yes | ? |
5 | No | No | Yes | No | No | No |
6 | No | No | Yes | No | Yes | ? |
7 | No | No | Yes | Yes | No | No |
8 | No | No | Yes | Yes | Yes | No |
9 | No | Yes | No | No | No | No |
10 | No | Yes | No | No | Yes | ? |
11 | No | Yes | No | Yes | No | No |
12 | No | Yes | No | Yes | Yes | ? |
13 | No | Yes | Yes | No | No | No |
14 | No | Yes | Yes | No | Yes | ? |
15 | No | Yes | Yes | Yes | No | No |
16 | No | Yes | Yes | Yes | Yes | ? |
17 | Yes | No | No | No | No | ? |
18 | Yes | No | No | No | Yes | ? |
19 | Yes | No | No | Yes | No | ? |
20 | Yes | No | No | Yes | Yes | ? |
21 | Yes | No | Yes | No | No | ? |
22 | Yes | No | Yes | No | Yes | ? |
23 | Yes | No | Yes | Yes | No | ? |
24 | Yes | No | Yes | Yes | Yes | ? |
25 | Yes | Yes | No | No | No | ? |
26 | Yes | Yes | No | No | Yes | ? |
27 | Yes | Yes | No | Yes | No | No |
28 | Yes | Yes | No | Yes | Yes | ? |
29 | Yes | Yes | Yes | No | No | No |
30 | Yes | Yes | Yes | No | Yes | Yes |
31 | Yes | Yes | Yes | Yes | No | Yes |
32 | Yes | Yes | Yes | Yes | Yes | Yes |
No . | Conditions . | Outcome . | ||||
---|---|---|---|---|---|---|
A . | S . | I . | R . | G . | E . | |
Age . | System Stability . | Network Structure . | Resource Munificence . | Network Governance . | Network Effectiveness . | |
1 | No | No | No | No | No | No |
2 | No | No | No | No | Yes | No |
3 | No | No | No | Yes | No | No |
4 | No | No | No | Yes | Yes | ? |
5 | No | No | Yes | No | No | No |
6 | No | No | Yes | No | Yes | ? |
7 | No | No | Yes | Yes | No | No |
8 | No | No | Yes | Yes | Yes | No |
9 | No | Yes | No | No | No | No |
10 | No | Yes | No | No | Yes | ? |
11 | No | Yes | No | Yes | No | No |
12 | No | Yes | No | Yes | Yes | ? |
13 | No | Yes | Yes | No | No | No |
14 | No | Yes | Yes | No | Yes | ? |
15 | No | Yes | Yes | Yes | No | No |
16 | No | Yes | Yes | Yes | Yes | ? |
17 | Yes | No | No | No | No | ? |
18 | Yes | No | No | No | Yes | ? |
19 | Yes | No | No | Yes | No | ? |
20 | Yes | No | No | Yes | Yes | ? |
21 | Yes | No | Yes | No | No | ? |
22 | Yes | No | Yes | No | Yes | ? |
23 | Yes | No | Yes | Yes | No | ? |
24 | Yes | No | Yes | Yes | Yes | ? |
25 | Yes | Yes | No | No | No | ? |
26 | Yes | Yes | No | No | Yes | ? |
27 | Yes | Yes | No | Yes | No | No |
28 | Yes | Yes | No | Yes | Yes | ? |
29 | Yes | Yes | Yes | No | No | No |
30 | Yes | Yes | Yes | No | Yes | Yes |
31 | Yes | Yes | Yes | Yes | No | Yes |
32 | Yes | Yes | Yes | Yes | Yes | Yes |
No . | Conditions . | Outcome . | ||||
---|---|---|---|---|---|---|
A . | S . | I . | R . | G . | E . | |
Age . | System Stability . | Network Structure . | Resource Munificence . | Network Governance . | Network Effectiveness . | |
1 | No | No | No | No | No | No |
2 | No | No | No | No | Yes | No |
3 | No | No | No | Yes | No | No |
4 | No | No | No | Yes | Yes | ? |
5 | No | No | Yes | No | No | No |
6 | No | No | Yes | No | Yes | ? |
7 | No | No | Yes | Yes | No | No |
8 | No | No | Yes | Yes | Yes | No |
9 | No | Yes | No | No | No | No |
10 | No | Yes | No | No | Yes | ? |
11 | No | Yes | No | Yes | No | No |
12 | No | Yes | No | Yes | Yes | ? |
13 | No | Yes | Yes | No | No | No |
14 | No | Yes | Yes | No | Yes | ? |
15 | No | Yes | Yes | Yes | No | No |
16 | No | Yes | Yes | Yes | Yes | ? |
17 | Yes | No | No | No | No | ? |
18 | Yes | No | No | No | Yes | ? |
19 | Yes | No | No | Yes | No | ? |
20 | Yes | No | No | Yes | Yes | ? |
21 | Yes | No | Yes | No | No | ? |
22 | Yes | No | Yes | No | Yes | ? |
23 | Yes | No | Yes | Yes | No | ? |
24 | Yes | No | Yes | Yes | Yes | ? |
25 | Yes | Yes | No | No | No | ? |
26 | Yes | Yes | No | No | Yes | ? |
27 | Yes | Yes | No | Yes | No | No |
28 | Yes | Yes | No | Yes | Yes | ? |
29 | Yes | Yes | Yes | No | No | No |
30 | Yes | Yes | Yes | No | Yes | Yes |
31 | Yes | Yes | Yes | Yes | No | Yes |
32 | Yes | Yes | Yes | Yes | Yes | Yes |
An important part of these unobserved configurations (logical remainders) concerns networks that have been in existence for 3 years but are still unstable (47% of the logical remainders). Such configurations may occur but are likely to be exceptional. This is because networks that have existed for at least 3 years are more likely to be stable unless they have imploded through instability. Networks that have existed for less than 3 years, on the other hand, can either be stable (38% of these “young” networks), or unstable (62%). The number of observed NAO-governed networks was somewhat restricted in this empirical setting, so another 47% of the logical remainders might be explained by mode of governance.
Figure 1 shows that the two pathways to network effectiveness are well covered. The first pathway to effectiveness (A ∙ S ∙ I ∙ R → E) is covered by six unique and two overlapping cases (raw coverage = 80%, unique coverage = 60%). The second (A ∙ S ∙ I ∙ G → E) is covered by two unique and two overlapping cases (raw coverage = 40%, unique coverage = 20%). Consequently, the presence of two pathways to network effectiveness was demonstrated by several cases. The strength of these pathways was improved by investigating the possible pathways that lead to network ineffectiveness. The first pathway to network ineffectiveness identified is ~A → ~E and was covered by 26 cases (raw coverage = 90%, unique coverage = 21%). The second, ~I → ~E, was covered by 20 cases (raw coverage = 69%, unique coverage = 7%). The last pathway to network ineffectiveness, ~R ∙ ~G → ~E, was covered by 12 cases (raw coverage = 41%, unique coverage = 3%).
Raw coverage provides an assessment of the relative empirical importance of each configuration. Thus, from an empirical perspective, a configuration with high raw coverage, such as A ∙ S ∙ I ∙ R → E, is more important than a configuration with a low raw coverage, such as ~R ∙ ~G → ~E. Even though the latter was covered by more cases in total, coverage was calculated for each outcome separately, thus the configuration A ∙ S ∙ I ∙ R → E is covered by 8 out of 10 cases with outcome 1, whereas ~R ∙ ~G → ~E is covered by 12 out of 29 cases with the outcome 0. But even if the raw coverage is low, a configuration can still be important from a theoretical perspective, because from a case comparative perspective as QCA represents, every case contains important information.
Unique coverage is the coverage that does not overlap with other configurations (Ragin 2006, 2008). The relatively low unique coverage scores for the configurations leading to ineffective outcomes indicate a significant overlap in the factors, that is, networks that are young, are also often unstable, not well integrated, and possibly lacking in resources. However, as in the case of raw coverage, a low unique coverage does not indicate that a configuration is not valuable from a theoretical perspective.
All in all, we found compelling evidence that network effectiveness can be hampered, either by a lack of age (at least 3 years), a lack of centralized integration, or a lack of resource munificence in combination with no NAO governance. The overall solution coverage was 100%, since all cases were included in a configuration either leading to the outcome 0 (ineffective) or 1 (effective). These findings are linked to examples and discussed in the next section.
A robustness test was performed to check the validity and legitimacy of the found configurations for the condition “network structure.” This was because “network structure” incorporates two dimensions (density-based integration and centralization) in one condition. Network structure was coded as being either centrally integrated (1) or not integrated (0), density-based integrated (0), or simultaneously density-based and centrally integrated (0). A robustness test was performed to consider the separate implications of integration through density and integration through centralization. Density-based and centralization-based integration were included in the model as separate conditions alongside age, system stability, resource munificence, and mode of governance to ascertain whether the results of the extensive model (with two conditions for network structure) corresponded with the results of the parsimonious model (with one condition for network structure).
The results showed that the configurations for network effectiveness included low density and high centralization as necessary conditions, which perfectly matched our measure of centralized integration. Nevertheless, high density was a sufficient predictor of network ineffectiveness. Consequently, the dichotomous classification of the network structure provided a parsimonious predictor of network effectiveness, although partial deviation from this condition (solely on the density dimension) was enough to cause ineffectiveness. A second robustness test was performed to address the notion of Kenis and Provan (2009) that the network development stage and the performance criteria should match. As the national target prescribed a reduction in recidivism within 2 years, only those (19) SHs that had existed for at least 2 years were included in this additional csQCA. The results were identical to the main findings of the csQCA on the complete set and, therefore, confirmed the robustness of the configurations that lead to effectiveness or ineffectiveness.
We, therefore, conclude that age, system stability, and centralized integration are necessary but not sufficient conditions for the effectiveness of SHs. In addition, two sufficient paths exist with these three conditions: one in conjunction with NAO governance, and one in conjunction with resource munificence (valid for both NAO and lead organization governance). Based on these findings, support for the hypotheses was as follows:
H1 | Age (in existence for 3 years) is a necessary but not sufficient condition for the effectiveness of SHs. | Confirmed |
H2 | System stability is a necessary but not sufficient condition for the effectiveness of SHs. | Confirmed |
H3 | Resource munificence is a necessary but not sufficient condition for the effectiveness of SHs. | Not confirmed |
H4 | Centralized integration is a necessary but not sufficient condition for the effectiveness of SHs. | Confirmed |
H5 | Being governed by an NAO is a necessary but not sufficient condition for the effectiveness of SHs. | Not confirmed |
H6 | SHs that are at least 3 years old, have sufficient resources, are stable, and are centrally integrated and governed by an NAO will be effective. | Partly confirmed. Two more parsimonious paths with four of the five factors. |
H1 | Age (in existence for 3 years) is a necessary but not sufficient condition for the effectiveness of SHs. | Confirmed |
H2 | System stability is a necessary but not sufficient condition for the effectiveness of SHs. | Confirmed |
H3 | Resource munificence is a necessary but not sufficient condition for the effectiveness of SHs. | Not confirmed |
H4 | Centralized integration is a necessary but not sufficient condition for the effectiveness of SHs. | Confirmed |
H5 | Being governed by an NAO is a necessary but not sufficient condition for the effectiveness of SHs. | Not confirmed |
H6 | SHs that are at least 3 years old, have sufficient resources, are stable, and are centrally integrated and governed by an NAO will be effective. | Partly confirmed. Two more parsimonious paths with four of the five factors. |
DISCUSSION AND CONCLUSION
This study examined 39 whole networks to determine the effect of configurations of mode of governance, network structure, system stability, network age, and resource munificence on network effectiveness. To guide our research, we formulated six hypotheses in a configurational format. Specifically, we defined necessary conditions for network effectiveness and a combination of necessary conditions in H6 as a sufficient path for network effectiveness. Hypotheses 1, 2 and 4, defining age (older than 3 years), system stability, and centralized integration, were confirmed since all three were necessary but not sufficient conditions in the two paths leading to network effectiveness. H3 and H5 were not confirmed. H6 was partly confirmed, since two sufficient, more parsimonious paths emerged with the suggested conditions. These two paths, however, contained only four of the suggested five conditions.
Even though H3 and H5 were not confirmed, the findings show that resources and governance mode indeed play an important role, but in a more differentiated way than originally hypothesized. The interpretation of this finding is two-fold. First, it shows that financial resources can to some extent be substituted by administrative resources in the form of a neutral facilitating institution. We identified €1.85 per capita as a threshold for high resource munificence. This finding does not suggest that networks can be effective simply with an NAO and no additional resources, but that networks with a budget of less than €1.85 per capita can make up for this disadvantage with administrative capacity in the form of an NAO. It, therefore, appears that an NAO can be an impartial facilitating organization that can cope better with potential tensions between law enforcement and social service organizations and the network tensions efficiency/inclusiveness and internal/external legitimacy than a lead organization model without considerable resources. This may be tied in with the fact that no matter what organization is in the lead, it will still have to deal with suspicions that it is favoring one over the other functional logic (law enforcement or social services). This leads to conflicts that can be mitigated if resources are available, but which undermine effectiveness if there are no additional gains for the organizations (see Park and Rethemeyer 2012 for a related argument). Therefore, in order to be effective, a network that is governed by a lead organization needs high resource munificence.
The findings for resource munificence confirm earlier findings by Turrini et al. (2010), who list 10 studies that show that resource munificence has a positive influence on network effectiveness. Another interpretation of this finding is that relatively abundant financial resources have a disciplinary effect on the participants in service delivery networks. Individual organizations are inclined to cooperate as long as they see strong financial incentives. The results for mode of governance and structure demonstrate that centralized integration is a necessary condition for SHs that are either lead organization– or NAO-governed (self-governed networks were not observed). As Provan and Milward (1995) state, network integration is essential if clients are to get a full package of services. But whereas density-based integration stimulates information exchange and trust between network partners, centralized integration promotes the efficient coordination of network activities by preventing overlapping or conflicting actions. The study showed that in the case of SHs the network must be centrally integrated to be effective. The complexity of the cases in the network requires well-coordinated activities at the operational level, which is improved by overall centralization.
These findings largely confirm Provan and Milward’s model (1995), as stability and centralized integration were identified as necessary conditions, whereas resource munificence in the form of financial and administrative resources plays an important role. Since external control, the fourth crucial factor in the model, did not vary in our case, it was excluded from our research.
The explicit application of the configurational approach and QCA as a further logical advancement of the work of Provan and Milward (1995) and Turrini (2010) demonstrated the existence of different potential configurations of the necessary conditions that lead to effectiveness. We see this as an important step toward a theory of network effectiveness. We identified two configurations or pathways to network effectiveness, both including the necessary conditions of high network age, high system stability, and high centralized integration. We found that an SH needed to exist for at least 3 years in order to impact significantly on the reduction of crime in society. Around 75% of the 13 SHs that had existed for at least 3 years qualified as effective. This finding makes sense, since it takes a while to create internal legitimacy between network partners: partners need to get to know each other before they can join forces. However, this applies mainly to networks that are not formed bottom-up through personal links between professionals from different organizations, but instead to service implementation networks in healthcare and social services that need more stable, institutionalized relationships, as in the case of the SHs (Provan and Milward 1995; Van Raaij 2006).
It should be noted that SHs deal with the most complex clients in the field of safety and crime prevention and need considerable time for orientation. Also, network effectiveness requires a stable system. Although system stability does not guarantee network effectiveness, an unstable system can have an adverse effect. System stability is required to build trust between network partners, to stimulate the constant exchange of information, and to embed the operational processes. This study confirms that, also in conjunction with other factors, system stability exerts a positive influence on network effectiveness—a finding identified consistently in six empirical studies by Turrini et al. (2010)—and can, therefore, be regarded a necessary condition.
The results further demonstrate that to generate sustainable positive results, networks for the implementation of healthcare and social services require a long time horizon, long-term commitment, and investment in the network by managers and policymakers alike. Networks might provide a solution for the complex or even wicked problems in society, but they are unlikely to succeed in policy areas that are subject to short-term considerations and frequent changes in policy, funding, and political and managerial personnel. Despite the core assumption, that is, that networks are a more flexible form of governance than hierarchies (Powell 1990), the results bear out recent arguments by Schreyögg and Sydow (2010) who claim that to operate successfully new organizational formats, such as networks, need to balance fluidity and stability even though this might pose dilemmas for organizational design.
Our research has developed the rudimentary configurational ideas present in Provan and Milward (1995) to a full configurational model based on necessary and sufficient conditions. Such a model should ultimately include the possible configurations not only for network effectiveness but also for network ineffectiveness. The results on the coverage of pathways for effective and ineffective outcomes (as defined with the threshold of three themes) indicate that the configurations leading to network ineffectiveness are not simply the negations of the configurations leading to effectiveness. This asymmetry of pathways is an important feature of the configurational approach and holds great potential and is necessary for future theory development, since different (combination of) factors might be responsible for network ineffectiveness than for network effectiveness. As stated earlier, networks are often set up to deal with complex or even wicked problems, so they are frequently confronted with multiple dilemmas that are more or less insoluble. The best option may be to avoid ineffectiveness despite the general rhetoric for high performance in public management.
The results of this study should be interpreted in light of its limitations. First, we were able to collect in-depth data in only 10 of the 39 cases. However, given the extensive research prior to the data collection for the remaining 29 networks (by interviewing the network managers), we are confident that the data are valid and reliable. In addition, the use of csQCA implies making a decision for a 0/1 score for every factor and the outcome in every case. The determination of thresholds is, therefore, crucially important. Critics could argue that 0/1 is a crude measure for a complex concept like effectiveness, but common measurement practices in qualitative research are routine and are often left unstated. We contend that measurements must be “adjusted on the basis of accumulated substantive knowledge” (Ragin 2008, 79) in order to interpret and set the most appropriate thresholds. In this study, effectiveness was not a precise ratio variable, indicating that a network with a score of 4 is twice as effective as a network that scores 2. Much of the variation in effectiveness could not be captured by this simple variable. We, therefore, used multiple indicators, subjective and more objective, and in-depth case knowledge to set the threshold. To limit researcher bias, we opted for a relatively high threshold. From the perspective of conventional quantitative research, the more qualitative calibration might be seen as an issue that potentially limits the analysis, but we see it as an essential and explicit step toward interpreting thresholds in the light of case knowledge and the research goal, that is, a clearer understanding of network effectiveness.
Second, in assessing effectiveness, we looked only at outcomes at the community level, not at network or organizational level, and we did so only from the agents’ perspective. When we studied these networks, we could clearly see an association between positive community-level outcomes and network-level outcomes in, for example, the functional operations of the networks, that is, the coordination and integration of services and the commitment of the participants. But we did not assess the outcomes at the level of individual organizations or investigate whether any of the organizations experienced negative effects from network involvement. Similarly, we assumed that the organizational capacities of the involved organizations were comparable across the different SHs, an assumption that is reasonable in a unitary state like the Netherlands, but should be included in the future within this stream of research.
Finally, even though we largely confirmed Provan and Milward’s findings (1995) for networks with a different goal and in a different institutional and national context, the SHs represent just one type of network. Effectiveness still needs to be researched for networks with other purposes, such as pure information diffusion, community capacity, or problem solving (Milward and Provan 2006). There is no “all-purpose organization design” (Snow, Miles, and Miles 2006, 7), and it is empirically very likely that there are other configurations that lead to similar outcomes for different types of networks in different structural and contextual settings. Moreover, we could only observe networks with a lead and NAO mode of governance. This can also be combined with broader theoretical questions regarding the most adequate integration mechanisms for different types and sizes of network in different settings. Whereas we looked exclusively at centralized integration, Turrini et al. (2010) identified other integration mechanisms that have been shown to have a positive effect on network effectiveness, such as steering committees and joint staff activities. Provan and Sebastian (1998) showed that overlapping cliques of players might offer an alternative form of integration in service implementation networks. Future research should, therefore, include self-governed networks and different forms of integration from a configurational perspective.
Despite the limitations, the results of this study clearly confirm the suitability of the configurational approach and set theoretical methods for explaining network effectiveness from a methodological perspective. We have demonstrated that it is possible with set theoretic methods to carry out more systematic comparative studies and to conduct data analysis in small and medium N, which should help the development of theory and testing, a critical issue for whole network research (Provan, Fish, and Sydow 2007).
Methods alone will not advance the study of networks and organizations. What is needed is an approach that embraces the complexity of networks while allowing researchers to identify causal combinations that consistently explain network outcomes. We believe that the further development of a theory of network effectiveness should follow a configurational approach, based on necessary and sufficient conditions as well as equifinality and asymmetry of causal conditions. This will not only build on the work of Provan and Milward (1995) and Provan and Kenis (2008) who were implicitly applying such an approach, but will also extend and refine the framework developed by Turrini et al. (2010) on the basis of their meta-analysis of empirical studies on whole network effectiveness. Based on their insights, prior work on the management (Agranoff and McGuire 2001) and performance (Herranz 2010) of networks, research on network tensions (Saz-Carranza and Ospina 2011), and the results of our study, we suggest that networks be conceptualized as systems or configurations, which include the components of strategy/goals, governance mode, structure, people, and (management) processes. Overall, network effectiveness is then dependent on the quality of the internal alignment of these components and on the fit between the overall configuration and the network context (environment). This is consistent also with the work of Snow et al. (2006), who propose a configurational approach to organizational design, which might be applied to interorganizational networks. Thus, selected structural characteristics can be analyzed in conjunction with formal modes of governance, or the “behavioral dimensions of network governance” (Saz-Carranza and Ospina 2011), such as tension management (Provan and Kenis 2008; Saz-Carranza and Ospina 2011) and leadership (McGuire and Silvia 2009).
One of the main challenges for future network research is, therefore, to identify compatibilities and incompatibilities between the abovementioned network components and the characteristics of the network context in order to formulate recommendations for network design. This could bridge a major gap in the research on public sector networks by delivering knowledge on the combined effects of key network and context characteristics (including the institutional environment). This, we believe, is an exciting perspective for public management scholars and practitioners alike.
REFERENCES
In the Web of Science this study has received 336 and in Google Scholar 1014 citations to date.
We thank an anonymous reviewer for drawing our attention to this point.
For an introduction to the rational, concepts, and techniques of configurational thinking and QCA, see Fiss (2007), Ragin (1987, 2000, 2006, 2008), Rihoux and De Meur (2009), Rihoux and Ragin (2009) as well as http://www.compasss.org/index.htm (last accessed Aug. 5, 2013).
To test if the threshold for effectiveness (being effective in at least three out of the four themes, i.e., >2) provides the best fit to the empirical context, we employed two “robustness tests” with the alternative threshold values for effectiveness >3 and >1. The robustness tests clearly demonstrate that alternative threshold values for effectiveness provide an imperfect fit to the research setting: they either cause a merger between or break up of the empirical groups. Consequently, the csQCA using these alternative threshold values (>1 and >3) provides either no solution at all or nonsensical contradictory solutions that do not explain the empirical distinction between effective and ineffective SHs. Considering the robustness tests as well as the respondents’ indication of the national goal achievement (5.8% in 2 years), setting the effectiveness threshold at >2 was most valid to make a distinction between effective and ineffective SHs at the community level.