From learning to influence: the evolution of collaboration in European Administrative Networks

ABSTRACT

Most European Administrative Networks (EANs) have been operational for over two decades. Yet, our understanding of their evolution is limited. How do EANs evolve? I formulate two hypotheses, premised on the notion that networks help actors address collective action problems by establishing social capital, whose distribution in the network changes over time to respond to changed circumstances. Using a Bayesian network model of the structure of the Council of European Energy Regulators (CEER) at two-time points (1998 and 2014), this paper shows that regulatory relationships shifted from being focused on learning to be aimed at achieving compromise in order to influence European policy-making. Rather than functional, static devices of rule harmonization, EANs are dynamic social networks that adapt to their changing environment. The paper provides a theoretical background and empirical approach to understand and analyse EANs evolution in different sectors.

KEYWORDS:

• European Regulatory Networks
• network evolution
• social capital
• Bayesian ERGMs
• coordination
• learning

Introduction

European Administrative Networks (EANs) have been a distinctive feature of the European Union (EU) governance system since its early days. EANs are sectoral networks of regulators or public administrators from each of the Member States of the EU. Their stated purpose has always been fostering regulatory harmonization across the EU (Dehousse, 1997). Given their scarce proven ability to achieve that purpose (Maggetti & Gilardi, 2011), in order to explain EANs’ persistence the literature has directed its attention to investigating how regulators use networks. The findings are that EANs serve regulators for a variety of purposes, e.g., pool resources (Vestlund, 2015), fight bureaucratic rivalries (Bach et al., 2016), increase their autonomy (Bach & Ruffing, 2013; Yesilkagit, 2011), manage their relationship with their supranational principal, i.e., the Commission (Vantaggiato et al., 2020). These contributions are, however, oblivious to the possibility that the purpose of EANs might change over time. Yet, important policy changes have occurred since the inception of regulatory collaboration in the EU in the late 1990s: successive rounds of legislation have fostered deeper integration of the Single Market as well as areas of social regulation. This paper investigates a pressing, if yet overlooked, research question: how do EANs evolve?

The few existing contributions investigating EANs’ evolution adopt an historical evolutionary perspective finding that, from the perspective of the regulators, the purpose of network collaboration changed over time in ways that dovetail key milestones in the evolution of European integration in the policy sector in question (Mathieu, 2016; Vantaggiato, 2020). Yet, we do not know whether change in regulators’ perceptions of their networks was associated with change in their empirical patterns of collaboration and what these patterns can teach us about how EANs evolve.

In order to conceptualize the evolution of EANs, I rely on the concept of social capital as operationalized in the policy networks literature: actors use policy networks to establish relationships with each other, which are the social capital (Burt, 2005; Lubell et al., 2014) that they can rely upon to deal with collective action problems. A leading argument in this literature is the ‘risk hypothesis’, which posits that interaction between actors generates networks structures that provide bonding social capital that is useful for addressing cooperation problems (where the risk of collaborators’ defection is high), or bridging social capital that is useful for addressing coordination problems (where the risk of collaborators’ defection is low) (Berardo & Scholz, 2010). Analyses then proceed by observing whether network structures are more consistent with bonding (more ‘closed’ structures, e.g., triangles or reciprocal ties) or bridging social capital (more ‘open structures’, e.g., networks centralized on key actors), and then inferring what type of underlying collective action problem is at hand. Recent research in this field, however, has found that, within policy networks, actors always develop both bridging and bonding social capital (Berardo & Lubell, 2016) because they face different problems simultaneously.

EANs are different from policy networks: policy networks typically involve many different actors from the same jurisdiction, whose goal is addressing a common policy problem; in contrast, EANs comprise one type of actor (national regulators), from various jurisdictions, whose goal is producing commonly agreed rules and promoting their harmonization. Still, the risk hypothesis is useful for conceptualizing the fact that network structure reflects the prevailing social capital in the network, that different types of social capital correspond to different social problems and processes, and that network evolution is a process whereby actors form ties corresponding to the social problems they face.

Existing literature suggests that the purpose of European regulatory cooperation is two-fold: learning, i.e., exchanging information and experiences to improve regulatory practice (Vestlund, 2015) and influencing European policy-making (Vantaggiato, 2020). The former task is a type of coordination problem as described in the risk hypothesis. In contrast, the latter task presupposes achieving compromise, i.e., crafting regulatory positions and decisions that can be endorsed by all (Van Boetzelaer & Princen, 2012); this task implies reconciling different preferences and priorities, and should prompt the formation of bonding social capital. The argument of this paper is that EANs evolved from being primarily aimed at learning to being primarily aimed at influencing EU policy-making. I test this argument in the empirical case of the European network of national energy regulators, using interview and questionnaire data collected in 2014.

I rely on Bayesian Exponential Random Graph models (BERGMs) to quantitatively compare the structure of the network in the early stages (i.e., in the late 1990s) and 2014. Bayesian ERGM allows for the incorporation of information about network structure gathered from the elite interviews as well as existing studies of EANs. The findings of the analysis support the current wisdom of the literature. Actors formed bonding and bridging social capital across time, but these are qualitatively different at different moments in time. At the onset of collaboration, national regulators seek ties to counterparts who are more expert than them (bridging) or that face similar challenges (bonding) (Martinsen, Schrama, & Mastenbroek, 2020). Later in time, the deepening of European integration shifts the focus of collaboration on achieving compromise (Mathieu, 2020), prompting regulators to form bonding social capital with a broader set of counterparts, knitting all regulators together in a dense web of interactions.

This paper makes a first step towards conceptualizing and operationalizing the processes of evolution of EANs. Attention is given to how the regulators responded to the evolution of their policy environment and how these changes also changed the purpose of the collaboration and the social dilemmas it embodied. Future research should investigate how the following big change in the evolution of regulatory networks (the creation of the European agencies) influenced the purpose of the collaboration and therefore the structure of the network. The ideas of bridging and bonding social capital and the risk hypothesis provide a useful framework of analysis for the evolution of regulatory cooperation in the EU.

Literature review: network evolution and social capital

There are several open questions in the literature on European Administrative Networks (EANs); a pressing one concerns their evolution over time (Mastenbroek & Martinsen, 2018). Early literature portrays EANs as institutional solutions to achieve regulatory harmonization via exchange of information (Coen & Thatcher, 2008), advice (Martinsen et al., 2020) and best practices (Maggetti & Gilardi, 2014). Several studies have assessed whether EANs achieve their stated aim of regulatory harmonization. The findings present, at best, a mixed record: with no powers to implement decisions made at the EAN level autonomously, the effects of regulatory cooperation on regulatory harmonization have been scarce (Coen & Thatcher, 2008; Maggetti & Gilardi, 2014)

Other contributors focused on another aspect of EANs: control. The arguments proposed depict EANs as instruments of the regulators to fight bureaucratic battles (Bach et al., 2016); of the European Commission to fight European regulatory battles (Tarrant & Kelemen, 2007); of the Member States to influence regulatory provisions at European level (Martinsen et al., 2020). The first argument resonates with findings that regulatory networks offer regulators the chance to loosen the grip of national institutional actors over their decision-making (Ruffing, 2014; Bach et al., 2016). The second and third arguments chime with observations that both the European Commission and the Member States possess incentives and authority to interfere with regulatory collaboration (Tarrant & Kelemen, 2007). These two visions of regulatory collaboration need not be mutually exclusive: EANs help regulators balance their interests and concerns with the demands of other actors in their environment (Vantaggiato et al., 2020). Importantly, however, this balance may change over time. The main limitation of most existing studies of EANs is an implicit assumption that the thrust of regulatory networking is immutable.

A handful of studies adopt a historical institutionalist perspective on EANs and paint a slightly different picture: EANs evolved from informal gatherings of newly appointed officials anxious to understand their job to well-oiled mechanism of influence on policy formulation at the European level (Boeger & Corkin, 2017; Mathieu, 2016; Vantaggiato, 2020). This explanation is underpinned by ample empirical evidence in the form of interview evidence coupled with analysis of policy documents issued by EANs at different time stages. However, we lack a theoretical and empirical approach to conceptualize and operationalize this evolution.

In order to understand how different stages of network evolution have shaped and modified EANs over time, I borrow conceptually and methodologically from the literature on policy networks (Berardo & Scholz, 2010; Leifeld & Schneider, 2012; Scholz et al., 2008; Scott & Ulibarri, 2019; Yi, 2018). Contributors in this literature interpret collaborative relationships as devices aimed at addressing the specific collective action problems encountered by the actors involved. Actors are interdependent insofar as addressing those challenges requires everyone’s participation. Actors’ collaborative ties form a networked structure reflecting the process of building ties with each other, over time, which leave their fingerprints in local network configurations. These relational structures represent the actors’ social capital (Burt, 2005; Putnam, 2001), which they can use to achieve their goals. This basic idea is at the heart of a voluminous body of research spanning political science (Lubell et al., 2002; Scholz et al., 2008), policy sciences (Ansell & Gash, 2012), public administration (Ansell & Gash, 2008; Emerson & Gerlak, 2014) and other fields (Hughes & Pincetl, 2014; Rudnick et al., 2019).

A key argument in this literature is the ‘risk hypothesis’ (Berardo & Scholz, 2010). According to the risk hypothesis, policy actors establish the kinds of ties that best enable them to address the collective action problems at hand. The combination of each actor’s own micro-level network of collaborative ties generates a higher-level structure (a network), which reflects the social capital possessed by the actors. Therefore, social capital can be both an individual level attribute, the personal ‘endowment’ of each actor resulting from the ties they chose to establish, and a network level attribute, reflecting the cumulation of those choices into a single interconnected structure of collaboration.

Collective action problems can be divided into cooperation and coordination problems. Cooperation problems imply a high risk that actors will defect and free-ride on the contributions of others, because they would benefit from doing so. The risk of defection prompts actors to establish many redundant ties to each other in order to monitor each other’s contributions to the common enterprise, which facilitates the development of trust (Berardo & Scholz, 2010). In contrast, coordination problems imply a low risk of defection; a typical example is information exchange and policy learning, where actors search their environment for information in order to learn the specifics of the policy issue at hand and other actors’ beliefs (Berardo & Lubell, 2016). When addressing coordination problems, actors tend to establish efficient ties that allow them to access information from other actors, particularly well-connected ones. Table 1 provides an overview of the network configurations corresponding to these different types of social capital.

Table 1. Social capital in network motifs.

	Function	Network Motif
Coordination Structures for Learning – Bridging social capital
Degree Centralization	Regulators seek information and resources from most advanced/resourceful counterparts
Open Two Paths	Efficient information transmission via hubs
Cooperation Structures for Influencing – Bonding social capital
Reciprocity	Close collaboration
Transitivity	Close collaboration

This literature initially hypothesized that policy networks face cooperation and coordination problems at different stages of development. At the outset, actors in policy networks seek to access information, in order to form their own policy position and to understand the policy positions of other actors; once the information is acquired, actors form policy coalitions with others who share their same policy position (Sabatier, 1987). According to this line of argument, the network evolves from one primarily based on coordination (prevalence of open two paths and degree centralization) to one primarily based on cooperation (prevalence of reciprocity and transitive ties).

Recent literature seeking to test the risk hypothesis, however, found that policy networks tend to develop both bonding and bridging social capital at the same time, which are both identifiable empirically in the same analyses (Berardo & Lubell, 2016; McAllister et al., 2014; Yi et al., 2018). The explanation given to the finding is that actors face multiple collective action problems at the same time. What changes is the relative importance that actors assign to these problems in different time periods (Angst & Hirschi, 2017).

Translating this argument into the realm of EANs requires some adjustments. EANs and policy networks are different: the former are transnational, while the latter are not; the former are homogenous and comprise the same type of actor (national regulators), while the latter are heterogenous and comprise different types of actors (policy makers, knowledge actors, civil society, non-governmental organizations); the former have the goal of promoting convergence and harmonization across widely different political and economic jurisdictions, while the latter typically address a specific policy problem within a single jurisdiction. Yet, the risk hypothesis is relevant to understand EANs, which are, essentially, a group of regulatory professionals who are interdependent insofar as the cross-border dimension of their regulatory tasks is concerned.

Furthermore, it is important to note the institutional structure encasing interactions within EANs, because it plays an important role in mitigating collaboration risks. Interaction between regulators within EANs, for the purposes of EANs’ goals and deliverables, takes place in working groups and regular board meetings typically coordinated by a central secretariat. This institutional structure, which developed over time as EANs consolidated as fixtures of the EU’s regulatory space, allows regulators to monitor each other’s behaviour and reduce their uncertainty about the trustworthiness and motives of other network members (Leifeld & Schneider, 2012). Therefore, these institutional structures should support the development of bonding social capital between regulators, its effects spilling over onto the more informal dimension of regulatory collaboration, which is the one examined in this paper.

In sum, existing literature points to two main stages of EANs’ development, moving from a focus on learning from each other to establish themselves as institutions to a focus on influencing policy-making at the European level as a cohesive unit. In the following, I examine each in turn.

Learning

Learning takes place when actors can obtain and effectively process new information about policy issues they care about, and about how other actors think about those problems (Berardo & Lubell, 2019). To improve their learning, actors will seek counterparts who can provide high-quality information, either because they are very experienced or because they are directly in charge of managing the policy issue in their organization (Berardo & Scholz, 2010). Structurally, this should translate into network configurations of centralization on knowledgeable actors (who will tend to be well-connected) and open-two-paths (see Table 1). Both configurations allow actors to reach distant parts of the network quickly.

Moreover, the policy literature emphasizes that collaborative networks in which the structures of interactions are characterized by high levels of heterophily (i.e., ties connect actors with different characteristics) also facilitate learning (Berardo, 2014). In contrast, homophily (i.e., a tendency to establish ties to actors similar to oneself) is associated with collective action problems of cooperation and the formation of coalitions.

Given uncertainty concerning the remit of their role, I expect the early stages of European regulatory collaboration to embed collaborative ties aimed at gathering information on regulatory practice. This should prompt them to connect to the most advanced counterparts at the time, and to bond with counterparts in similar political economies.

Hypothesis 1: The early structure of European regulatory networks is explained by bridging social capital with centralization on key expert regulators (learning from best in class) and bonding social capital within small clusters of similar political economies (learning from similar peers).

Influencing policy-making

As integration deepened, regulatory collaboration came to be seen as pivotal to achieve policy convergence (Dehousse, 1997). The European Commission co-opted and politically backed regulatory networks by offering them the opportunity to contribute to policy formulation (Eberlein & Newman, 2008). This offered regulators precious opportunities to both ‘upload’ their national policy preferences to the European level and ‘download’ European rules they had contributed to shape into national markets (Bach & Newman, 2010). However, regulatory networks were supposed to forward to the Commission a single, agreed upon regulatory position.

Achieving compromise is a different activity that seeking information and expert advice; it presupposes finding common ground with others. This implies sustained dialogue, exploration of alternatives, and eventually convergence of views and beliefs (Berardo & Scholz, 2010). Reciprocated ties and transitive relationships achieve the opposite goal of bridging social capital structures: they foster redundancy in information. This redundancy, though inefficient, can serve to make cooperation more feasible by increasing credibility among actors (Ulibarri & Scott, 2017), who can monitor the quality of the information they receive across providers. This suits the case of EANs, whose ability to influence policy-making resides in their expertise and their access to high quality information on national markets (Vantaggiato, 2019b). Moreover, since more resourceful and experienced regulators command more influence within the network (Papadopoulos, 2017), I expect them to receive more incoming ties from their peers who consult them for advice and information. In other words, I expect more resourceful regulators to serve as coordinators within the group, enabling it to achieve compromise (Ansell & Gash, 2008).

Hypothesis 2: Long term development of European regulatory networks is driven by increasing bonding social capital throughout the network (influence policy-making) with centralization on key influential regulators (coordinators of compromise).

The opportunity of policy influence does not imply that networks have stopped serving their learning function. Rather, the two rationales for network collaboration may co-exist at the same time.

Case study

This paper focuses on the empirical case of European national energy regulatory authorities. This is one of the most studied EAN in the literature (Maggetti, 2014; Mathieu, 2016; Rangoni, 2017; Vantaggiato, 2019a; Vasconcelos, 2009), one of the first to form, and one which had enormous influence on the development of a European regulatory framework for energy markets that simply did not exist before the inception of collaboration (Mathieu, 2020). Therefore, this network is a ‘crucial case’ of the mechanism proposed in the paper (Seawright & Gerring, 2008), likely to reflect the evolution of other EANs that played a pivotal role in determining the course of the current European rules in their respective sectors, e.g., telecommunications (Humphreys & Simpson, 2008; Thatcher, 2001), aviation (Pierre & Peters, 2009), health (Martinsen & Schrama, 2021) and competition (Vantaggiato et al., 2020).

In the mid-1990s, liberalization reforms in telecommunications and energy swept across Europe inspired by the example of the UK liberalization programme of the late 1980s (Lodge & Stern, 2014); regulators found themselves tasked with implementing and overseeing market reforms for which little precedent existed (Vasconcelos, 2009). In 1996, the European Commission released the first so-called Energy Package of legislation mandating the end of national monopolies. Seeking to learn the ropes of their new profession, and facing the absence of national counterparts, newly established national regulators began looking for advice and information from counterparts abroad: both the pioneers (e.g., the British regulatory authority) and regulators from countries with comparable legislative and market provisions (Vantaggiato, 2020). The European legislation for energy markets released in 2003, which mandated the establishment of national authorities in all Member States, consolidated the importance of the, as yet, wholly informal network of 15 European energy regulators – the Council of European Energy Regulators (CEER).

In 2004, the enlargement brought 10 new Member States into the EU and, therefore, 10 new national energy regulators into the CEER. Regulatory authorities in Eastern Europe had similarly been established in the early 1990s and maintained collaborative relationships with each other (LaBelle, 2012). As energy market integration progressed and deepened, so did interaction between regulators and between them and the European Commission. A European regulatory network with advisory role to the Commission was established in 2003 (Eberlein & Newman, 2008). In 2009, it mutated into the Agency for the Cooperation of Energy Regulators (ACER). Alongside these significant developments, national energy regulators always maintained wholly informal ties to their European counterparts in other Member States. These informal ties are the focus of this study.

Data

The data for this paper derives from fieldwork interviews conducted in 2014–2016 and a web-based survey disseminated to national energy regulators in the European Union in 2015. The network data concerning the first period of regulatory collaboration derives from 10 interviews with former regulators who were directly involved in setting it up (see Appendix). The data concerning the later moment of regulatory collaboration derives from the survey, which was answered by representatives of all European National Regulatory Authorities that are members of the CEER, bar one.

The interviews aimed at reconstructing the emergence of regulatory collaboration in Europe from its inception in the late 1990s. There are no written records from this initial period of European regulatory collaboration in the energy sector. The early days of the collaboration were wholly informal and thus undocumented. As such, interviews with individuals directly involved with the network at that time point represent the best available data collection procedure to investigate network emergence. Interviewees were asked to report on their closest collaborative ties at the time, as well as those of their alters (collaborators). At the time, the European Union comprised 15 members from Western Europe. However, not all of these members had established regulatory authorities by the year 2000 (the upper boundary of the first time period considered in this analysis, when the CEER was founded). Moreover, national energy regulators had already been established in Central and Eastern European countries, and maintained collaborative ties (LaBelle, 2012). These ties – reconstructed via interviews – are included in the reconstruction of the structure of this early network. These data comprise ties established between 1998 and 2000.

The data on the early structure of regulatory collaboration is ‘pseudo-longitudinal’ since it captures collaborative relationships at different time points, although it was gathered at one-time point (Bodin et al., 2019). Therefore, the data have limitations: not all regulators were interviewed about them; many of the ties in the early network are reported ties; interviewees were asked to recall ties they possessed two decades prior. However, several considerations mitigate the importance of these shortcomings: my interviewees are individuals who were directly involved in these developments; they were asked to recall ties with a limited number of alters (maximum of nine in the case of Western European regulators and seven in the case of Eastern European regulators); although established two decades prior, these relationships were formed at a pivotal moment in their professional career, when they created, from scratch, their own national regulatory frameworks as well as the European one. No interviewees had any hesitation recalling their own ties or their peers’. Moreover, interviews were used to triangulate alters’ ties across interviewees, with no discrepancies between accounts.

Coming to the second period of collaboration examined in this paper, the web-based survey was addressed to the Heads of International Affairs departments and to Communication Officers of all EU national energy regulatory authorities. Respondents were asked to reply to the following question: ‘Think of the individuals you (or somebody at your NRA) exchange information with more often. Which NRAs do they belong to?’. This question echoes the question asked in the qualitative interviews. All bar one regulators replied to the question. Regulators were asked to name the regulatory authorities with which they are in contact above and beyond European policy requirements (including participation into the European Agency for the Coordination of Energy Regulators).

Method

For the quantitative network analysis, I rely on comparative analysis of two cross-sectional Bayesian Exponential Random Graph Models (BERGMs) for each of the time periods under consideration (Caimo et al., 2021). ERGMs are a family of statistical models for social networks that permit inference about prominent configurations in the network structure, given the presence of other network structures (Robins, 2015). Namely, ERGMs identify parameters by maximizing the probability of the observed network over the networks with the same number of nodes that could have been observed. This is conditional on a set of network statistics that can include node characteristics (e.g., type of political economy or resources), as well as certain network structural characteristics or ‘motifs’ (i.e., patterns of relationships among a small number of nodes such as those described in Table 1). The underlying assumption of ERGMs is that the observed network structure has emerged from an evolutionary process of tie formation over time, which can be explained by the combination of theoretically and empirically relevant variables as well as network dependency structures (Desmarais & Cranmer, 2012).

The Bayesian approach to ERGMs presents two important advantages for the present analysis. Firstly, BERGMs allow incorporation of information on network structure from previous studies and expert judgement (Caimo et al., 2021) as analytical priors. For the network at time 1, the prior information derives from the qualitative interviews and from existing research. For the network at time 2, prior information on network structure derives from the questionnaire and from existing research. Secondly, BERGMs provide a method for the imputation of missing ties in the network at time 1 which has been found to outperform all other existing methods and to retrieve the functional characteristics of the original data with a high degree of precision (Krause et al., 2020).

At the two time periods, the network had different sizes. Between 1998 and 2000 only a handful of regulatory authorities had been created across the then European Union and Central and Eastern Europe. At time 1, the network comprises 18 nodes. At time 2, it comprises 28 nodes. This difference prevents reliance on temporal models of network structure, such as the Separable Temporal Exponential Random Graph Model (STERGM) (Krivitsky & Handcock, 2014), which allow modelling edge formation and edge dissolution as two separate processes.

In the analysis, I introduce a categorical variable indicating the variety of energy market of each Member State, following the categorization employed in Vantaggiato (2019a). The models include this variable under the homophily term in the network at both time periods. The homophily term tests whether regulators tend to establish collaborative ties to counterparts who oversee similar energy markets. Further, I include in the second model a categorical attribute describing regulators’ staff resources,¹ in order to test the hypothesis that network ties are redirected, over time, to the most resourceful regulatory authorities. The quantitative data have been standardized before proceeding to the analysis by subtracting the mean and dividing by the standard deviation.

All calculations were carried out using the R statistical software. The network analyses were carried out using the R package Bergm (Caimo et al., 2021).

Analysis

The early ties between European regulators were few and sparse, as shown in Figure 1. Even though it was not required, these two groups of regulators had occasional interactions in those early years, inspired by a willingness to compare the challenges posed by market liberalization (interview 1, 2, 3). The ties in Figure 2, resulting from European regulators responses to the electronic questionnaire, show a denser and cohesive structure (in Figure 2, most nodes of the network are unlabelled for reasons of anonymity). Table 2 reports the results of the BERGMs. The model for the first time period is a model resulting from imputation of missing data for the regulators that were named by interviewees but finally not interviewed as part of the research. The model for the second time period is based on nearly complete data (only one regulator failed to complete the questionnaire) and imputation is carried out for the only missing respondent.

Figure 1. Network of European energy regulators (reconstruction), 1998–2000, nodes sized by degree centrality.

Figure 2. Network of European energy regulators, 2014, nodes sized by degree centrality.

Table 2. Bergm models results.

Model	Parameter	Posterior mean	Posterior SD	Acceptance rate
Time 1–1998–2000 (augmented model with missing data imputation)	Density	−2.85	0.67	0.25
	Reciprocity	3.36	0.85
	Homophily by variety of energy market	0.75	0.37
	Open two-paths	−0.40	0.17
	Transitivity	0.76	0.25
Time 2–2014(augmented model with missing data imputation)	Density	−3.50	0.30	0.3
	Reciprocity	1.65	0.38
	Homophily by variety of energy market	1.25	0.22
	Large staff	Reference
	Medium staff	−0.33	0.28
	Small-mid staff	−0.79	0.33
	Small staff	−0.66	0.33
	Micro staff	−0.66	0.31
	Transitivity	0.76	0.10

The mean values in the table should be read and understood as the coefficient for a logit model, i.e., in terms of odds. Any value above 1 indicates a high probability that the tie existed. For instance, a high coefficient for transitivity would indicate that most open two-paths have a high probability of being closed into a transitive relationship.

The priors imposed on the model for time 1 foresaw a very low probability of a tie between any two actors (−5, which, once exponentiated, corresponds to a probability of 0.006). This decision was made to mirror the empirical findings of the interviews. The priors of all other parameters were set to be positive, in line with empirical evidence from the interviews and with hypothesis 1. The results of the model show that the 95% credible interval for all parameters (except for the density and open two-paths) lie in the positive region (see Figure 3). The negative density means that not all of the possible ties among the regulators are realized; this is a common finding for all but the densest social networks. The negative coefficient for open two-paths suggests that triangles are generally closed, meaning that when two regulators have a tie to a common alter, they tend to also share a tie. The fragmentation of this network into three separate components results in locally central regulators (the British and Hungarian regulators) within an overall decentralized structure where bonding social capital is prevalent, as testified by the high reciprocity coefficient. These findings partially confirm hypothesis 1.

Figure 3. Bergm model time 1 (thinner lines represent 95% credible intervals).

The BERGM model for time 2 shows that reciprocity and transitivity as well as homophily parameters lie in the positive region, while density lies in the negative region. Parameters for staff numbers show that regulators with small to micro staff numbers are less likely to have incoming ties than regulators with medium or high numbers of staff (the reference category). Reciprocal ties and triangles remain important determinants of network structure, while on regulators with higher resources tend to be more central: bonding and bridging social capital still co-exist. This model does not include a term for open-two-paths because in this network they are all nested within transitive relationship (see Figure 2). Homophily by type of energy market is also consistently associated with the existence of ties in the network, confirming existing studies (Figure 4).

Figure 4. Bergm model time 2 (thinner lines represent 95% credible intervals).

These results support Hypothesis 1, which expected the ERGM model to feature bonding social capital within small clusters (learning from similar peers) and centralization on advanced regulators. The results also lend support to hypothesis 2: bonding social capital persists, centralization is driven by resources (taken as a proxy for influence) rather than stage of liberalization. The goodness of fit plots for the models are shown in the Appendix. The models provide good fit to the data.

Discussion

Although bonding and bridging social capital co-existed at both time points, they had different purposes in each time period. In the early stages of regulatory collaboration, national regulators set on a mission to collect information from their peers (interviews 1, 3, 4, 5). This information gathering followed two rationales: reaching out to regulators with most experience at the time (i.e., who had liberalized earlier, i.e., the British and Hungarian regulatory authorities in, respectively, Western and Eastern Europe) and reaching out to regulators from countries with similar legal frameworks, with whom to exchange more detailed information about national conditions. The network features very high reciprocity and a tendency to triangulation. Thus, patterns of bridging social capital and bonding social capital co-exist within a framework of learning.

Two decades later, bridging and bonding social capital still co-exist, but reflect a different balance of purposes. Firstly, bonding social capital appears to prevail, with all regulators being embedded in the network and part of a reciprocal or triadic relationship. Secondly, bridging social capital is primarily reflected in centralization on regulatory authorities with the highest amount of (staff) resources. Insofar as staff resources represent a proxy for influence (Papadopoulos, 2017; Vantaggiato, 2019a), this finding suggests change in the purpose of the collaboration. Exchanging information about market characteristics and regulatory criteria across borders is a delicate task (Efrat & Newman, 2017), which relies on the existence of trust to be performed. Since the late 1990s, EANs made regulatory collaboration, previously inexistent, a fact of life in European governance. The densification and cohesion of the later network belies the maturity of the collaboration, which encompasses both formal (i.e., within-CEER) and informal relationships as regulators pursue compromise amid enormous differences between their national markets. The necessity to achieve consensus in order to be able to contribute to European policy-making – since regulators needed to forward the Commission a single ‘regulators’ position’ – allowed individual regulators to essentially veto positions that they could not endorse (interviews 1, 2). The influence and experience of leading regulators conferred them coordinating functions (typical of central actors in policy networks, (see Berardo & Lubell, 2016), allowing the collective to achieve compromise and preventing the collaboration from stalling or falling apart (interviews 3, 4, 5).

Other EANs may experience different evolutionary paths. Some may evolve from highly centralized and connected to decentralized and disconnected, contrarily to what is observed in the case of energy regulators. Extant research on collaborative networks for service delivery states that they typically follow an evolutionary trajectory that begins from the bottom up with a few members establishing a collaborative structure that evolves into a more centralized and formalized one as the network grows in size (Provan & Kenis, 2008). Research on policy networks claims that they evolve from centralized and open to decentralized into separate close-knit communities linked by a few brokers (Berardo & Scholz, 2010). The literature on EANs has no specific hypothesis concerning how EANs evolve. This paper makes a first step in that direction by proposing that EANs evolve from separated close-knit communities of national regulators linked by similar political economy to a single close-knit community of peers that, while maintaining homophilous relationships based on similarity, focuses primarily on achieving compromise in order to influence policy-making.

Conclusions

Studying the process of evolution of EANs helps addressing some of the most prominent gaps in the relevant literature (Mastenbroek & Martinsen, 2018): do EANs change over time? If so, how to conceptualize and measure this change? This paper offers a theoretical and empirical approach to analyse network evolution in the case of EANs and beyond, which is premised on the notion that the purposes of the collaboration leave traces in the structure of the network as micro-configurations of relationships linking network members. These configurations are given theoretical meaning as embedding different kinds of social capital, which reveal the type of collective action problems that network members grapple with.

The network of European energy regulators evolved from an initially scattered set of close-knit collaborative structures focused on learning to a cohesive structure of collaboration focused on achieving compromised in order to inform and influence European policy-making in the sector. Although this conclusion is somewhat already present in the literature (Mathieu, 2016; Vantaggiato, 2020), this paper contributes a quantitative and structural analytical approach that allows for replication to the cases of other regulatory networks. Namely, the analysis investigates the relative prevalence of different types of social capital (Burt, 2005) at different stages of network evolution as reflected in specific micro-level network configurations (or motifs, see Berardo & Scholz, 2010). The analysis leverages data collected at two time points and interprets the findings based on knowledge of the policy context and regulators’ motivations for collaboration at each time point.

Methodologically, the paper unpacks the co-presence of bonding and bridging social capital that is by now a recurrent finding in this type of analyses (Berardo & Lubell, 2019) by combining quantitative network analysis with qualitative interview analysis to disentangle the meaning of the collaborative ties at the two-time points. The paper adopts a Bayesian approach to network modelling which allows the researcher to incorporate relevant analytical priors and to impute missing data based on the model specification, itself grounded in the literature and the qualitative data.

Empirically, the findings show that cross-boundary ties only existed between actors at the forefront of energy market privatization and liberalization reforms in the late 1990s (the UK in Western Europe, Hungary in Eastern Europe). By 2015, the network of European energy regulators comprised a wealth of connections across jurisdictional boundaries, with high triangulation (indicating bonding social capital) and centralization (bridging social capital). High triangulation suggests that regulators monitor each other’s contribution to their common goal of achieving compromise amid the difficult trade-offs posed by energy market integration (Pérez-Arriaga, 2014). High centralization on resourceful regulators suggests that these regulators perform a coordinating function of leadership.

The important feature of this analysis is its potential for replicability for other studies of network evolution over time. Future research should further investigate how structures of regulatory collaboration adapt to exogenous change in the policy environment. In the case of European energy regulators, the 2010s ushered in two important challenges to the policy relevance of their regulatory network: the establishment of the European Agency for the Cooperation of Energy Regulators (ACER) in 2011 and the increasingly pivotal role of market operators in the European energy policy framework (Eckert & Eberlein, 2020; Jevnaker, 2015). These changes have likely triggered renewed change in the purpose of regulatory collaboration; whether in the direction of attempting to compete with these other actors for influence on the Commission or in the direction of complementing these actors’ influence by diversifying their contribution to European energy policy-making is an important empirical question for future research.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Francesca Pia Vantaggiato

Dr Francesca Pia Vantaggiato is a Lecturer in Public Policy.

Notes

1 For data on regulators’ resources I rely on the report on national regulatory authorities’ staff resources released by ACER in 2016, https://bit.ly/2GQnzFP (last Accessed February 9, 2021). The ACER report groups regulators according to the number of Full-Time Equivalent (FTE) staff dedicated to the regulation of energy markets. The ACER document categorises regulators’ staff levels in six categories: ‘large’ (over 220 FTE); ‘medium-large’ (between 170 and 175 FTE); ‘medium’ (between 90 and 140 FTE); ‘medium-small’ (between 50 and 75 FTE); ‘small’ (between 12 and 50 FTE) and ‘micro’ (fewer than 12 FTE). For ease of exposition and analysis, I merged category ‘medium-large’ (comprising the energy regulatory authority of Italy and Spain) into ‘large’.

Appendix

Table A1. Regulators interviewed for the research.

Date of interview	NRA	Role
03/11/2014	Italy	Former regulator
04/11/2014	Italy	Former regulator
07/11/2014	Italy	Consultant
11/12/2014	United Kingdom	Former regulator
11/12/2014	United Kingdom	Former regulator
31/03/2015	Portugal	Former regulator
27/05/2015	Hungary	Former regulator
28/05/2015	Latvia	Former regulator
22/03/2016	NA	Consultant
02/11/2016	Lithuania	Current regulator

Figure A5. Example 4 simulated networks from Bergm imputation procedure for missing data.

Figure A6. Goodness of fit BERGM time 1, 1998-2000.

Figure A7. Goodness of fit BERGM time 2, 2014.