How different multilevel and multi-actor arrangements impact policy implementation: evidence from EU regional policy

ABSTRACT While implementation through multilevel governance (MLG) has often been found to have achieved limited success, the literature agrees that further studies are required to investigate if and how different MLG systems impact policy implementation. By collecting data from the implementation of four European Union (EU) regional development programmes, characterized by distinct organizational and institutional arrangements, this article adopts a performance-oriented approach to EU implementation in order to support or challenge the conventional assumptions about the expected benefits of MLG. The data suggest that effective MLG needs a central coordinating authority that has the power to enforce decentralized actors’ compliance, mobilize the implementation bodies, and provide actors with adequate organizational structures and the resources that they lack.


INTRODUCTION
Does multilevel governance (MLG) improve policy implementation?How do different MLG systems impact policy implementation?Despite recent developments in the field, the political scientist literature agrees that only partial answers have been given to these interrelated research questions, to which this article intends to respond.
Implementation arrangements and structures are of long-standing concern in public policy and public administration research.Since the 1970s, in fact, political scientists have been interested in the impact of multilevel and multi-actor arrangements on policy implementation (Hanf & Scharpf, 1978;Matland, 1995;O'Toole, 1986O'Toole, , 2000;;Pressman & Wildavsky, 1973).This debate gained traction in the context of European integration, since the implementation of several European Union (EU) policies required the development of MLG systems (Benz & Eberlein, 1999;Casula, 2022) in order to facilitate the fulfilment of EU policy targets and objectives (Committee of the Regions, 2009Regions, , 2014)).In this context, implementation through MLG often had limited success, since subnational actors struggled to establish functional governance layers (Gollata & Newig, 2017) and became caught in a number of implementation traps (Domorenok, 2017;Milio, 2010).
The recent literature has confirmed that further empirical studies are required to learn if and how MLG improves policy implementation (Thomann & Sager, 2017a;Trein et al., 2019), thus advancing our understanding of implementation performance in the EU multilevel system (Thomann & Sager, 2017b).After testing their hypothesis that the interplay between Europeanization and domestication is a central explanatory feature of implementation performance, Thomann and Sager (2017b) advocated addressing the practical effectiveness of EU policy through the use of a more evaluative perspective, and exploring the relevant properties of policies and institutions that bring about different implementation dynamics (p. 1398).The present article contributes to this academic debate on the relevance of MLG policy settings to policy implementation from both theoretical and empirical standpoints.It does so by addressing if and how different MLG systems impact policy implementation.
The theoretical contribution of this article consists in its assessment of the way in which components of MLG are integrated to produce policy.In doing so, it investigates how four context conditions, namely: (1) the level of decentralization in the management of policy; (2) the role of regional political factors; (3) the organizational features of policy performance; and (4) the degree and type of stakeholder participationaffect the effectiveness of decentralized policy management by producing different implementation dynamics.For this purpose, the article posits a set of theoretical propositions to guide the empirical analysis in order to support or challenge conventional assumptions on the expected benefits of policy implementation through MLG.
Against this theoretical background, the article's empirical contribution is its examination of how different multilevel and multi-actor arrangements operate across different policy stages, and how they affect policy implementation.An evaluation perspective is adopted to analyse the different implementation dynamics related to the management of four Italian and Spanish regional development programmes financed by EU regional policy (also known as Cohesion Policy).The implementation of Cohesion Policy is a paradigm case of MLG: since the 1990s, EU-funded regional development programmes have been managed by means of multilevel and multi-actor governance arrangements (Hooghe, 1996), and implementation of the partnership principle (Bache & Jones, 2000).More specifically, EU regional policy is implemented through operational programmes (OPs)detailed plans in which the amount of money to be spent during a programming period is determined.The implementation of an OP, drawn up to achieve a countrywide or regional objective, requires constant interaction among EU, national, regional, local and civil society representatives during all stages of preparation, financing, management, monitoring and operations assessment.The way in which policy is designed can vary in each member state (MS) and its regions, depending on domestic policy choices.Thus created are different implementation dynamics with different multilevel and multi-actor arrangements.
The article's findings suggest that, in order to avoid the implementation gaps that may arise between an intended policy and its results (Pressman & Wildavsky, 1973), effective policy implementation through MLG must operate in a context of fully participating technical actors, politicians, and vertical and horizontal partners.But their participation must occur under the supervision of a higher level of government, one able to compensate for the potential shortcomings of MLG.
The remainder of the paper is structured as follows.The next section presents the theoretical background.The third section illustrates the distinct characteristics of the Italian and Spanish contexts for implementing EU Cohesion Policy.The fourth section details the research design and methodology.The fifth section presents the results of the empirical analysis.The sixth section discusses the findings.The seventh section concludes.

THEORETICAL BACKGROUND: POLICY IMPLEMENTATION THROUGH MLG
The concept of MLG made its first appearance in European integration studies as an alternative to the state-centred inter-governmental approach and as a result of the new structures put in place by the Treaty of Maastricht (Marks, 1993).In this role, MLG reflects the importance of distributing the competences of actors at different levels for designing effective EU policies (Bache, 2008).The dispersion of policy-making tasks was seen as more efficient than central state monopoly and other modes of governance (Piattoni, 2010), given that the simultaneous activation of new centre-periphery, domestic-foreign and state-society dynamicswhich is at the core of MLG theorizingwould first ensure wider and fuller participation in decision-making processes (input legitimacy) and then produce better policies (output legitimacy) (Piattoni, 2009, p. 164).
The canonical distinction between Types I and II MLG provided by Hooghe and Marks (2003) suggested that the participation of lower levels of government in Type I MLG (because of their greater insight into the needs of beneficiary groups), and of private actors in Type II MLG (because of their different viewpoints), guarantees the incorporation of more diverse information in policy-making.Having established that more and more EU policies aim at the inclusion of both private actors and different levels of government, Piattoni defined MLG as: 'a class of policy-making arrangements characterized by the simultaneous activation of governmental and non-governmental actors at different jurisdictional levels and such that the interrelationships thus created defy existing hierarchies and rather take the form of non-hierarchical networks' (Piattoni, 2015, p. 326).In a seminal article, Piattoni (2009, p. 176) suggested that the concept of MLG should not be tailored as a mere descriptor of existing governance structures; rather, it should help empirically analyse the implications of these structures in different countries.Given that domestic actors enjoy a great deal of autonomy and discretion in designing their MLG mechanisms (Ongaro et al., 2019), the impact of different multilevel and multi-actor arrangements on policy implementation is an interesting subject for empirical investigation.
To investigate the matter, this article adopts a policy implementation perspective rooted in the public governance literature.As said, the study of implementation arrangements and structures has been of long-standing concern in the public policy and administration literature since the 1970s.In identifying the numerous barriers to effective implement policy programmes (Linder & Peters, 1987), the first studies on implementation research (e.g., Bardach, 1977;Pressman & Wildavsky, 1973;Sabatier & Mazmanian, 1979;Van Meter & Van Horn, 1975) had the merit of conceptualizing implementation as a complex and dynamic process in which multiple actors with contrasting interests and different interpretations of authoritative decisions participate.These pioneering implementation studies were criticized for being overly top-down in their approach.An alternative bottom-up approach in programme implementation research then began to develop in the 1980s (e.g., Elmore, 1980;Hjern, 1982;Hjern & Porter, 1981).These implementation scholars were interested in analysing the role of local networks in affecting a given problem during the implementation process.According to Hull and Hjern (1987), for example, implementation structures and arrangements tend to be poorly hierarchical, and their establishment leads towards the creation of collaborative networks at the operational level that transcend standard organizational boundaries.This sterile top-down/bottom-up dispute has subsided since the 2000s because sufficient evidence has been accumulated to argue that 'variables located at the top or centre can be important, as can contextual or field variables' (O'Toole, 2000, p. 268)and therefore 'to validate partially top-down and bottom-up arguments' (p.268).Efforts at synthesis have been numerous (e.g., Elmore, 1985;Goggin et al., 1990;Sabatier, 1986).They have adopted several theoretical and methodological approaches, often in combination with each other, and they have proved promising for implementation scholarship in the past three decades (O'Toole, 2000, pp. 273-282).Used in particular have been the following: rational-choice institutionalism; the governance approach; the network and network management; formal, rational-choice models; and the policy design and instrument approach.
The study of network and public management is pertinent to the purpose of this article because it 'draws from promising theoretical streams with questions of implementationperformance via governance in the delivery of policy resultsas significant as ever' (O'Toole, 2000, p. 281).In a pioneering article, O'Toole (1997) argued that 'in many cases involving network implementation, empirical scholarship has demonstrated that substantial challenges await those seeking to manage the effort and that significant implementation performance gaps can often be found ' (p. 116, emphasis added).This multi-actor implementation perspective (O'Toole, 1986) has consolidated in recent decades (O'Toole, 2012), and several attempts have been made to link it with the policy design approach (Busetti & Dente, 2018;May, 2012).Ansell et al. (2017), for example, argued that the implementation of well-crafted policy designs cannot be ensured by traditional top-down implementation based on command and control, but instead by an ongoing collaborative design process able to adapt the initial policy design so that it better reflects emerging problems and challenges.This literature has therefore conceived implementation as a problem of cooperation among multiple actors, with 'the degree of coordination required in a given instance [that] is largely determined by the structure of interdependence among those involved' (O'Toole, 1997, p. 120).
This article recognizes that these problems of cooperation increase in the context of EU integration, since the implementation of EU programmes requires the design and consolidation of implementation structures that involve multiple actors at different levels of government.According to this article, the successful establishment of these multilevel and multi-actor implementation arrangements requires an ongoing collaborative design process in which high degrees of coordination are needed to avoid the various disputes that may occur among politicians, bureaucrats, and social and private actors (Ansell et al., 2017) if they see a policy programme as directly relevant to their personal interests (Matland, 1995).
Due to the 'complexity of joint action' (Pressman & Wildavsky, 1973) required by a EU programme's implementation (Bauer, 2006), in the view of this article the achievement of these high degrees of coordination for a better policy performance is expected to happen under the supervision of a higher level of government able to guarantee continuous cooperation among the plurality of actors involved.Hence, this article follows the argument put forward by Homsy et al. (2019, pp. 572-573) that to be effective, MLG requires the presence of a central coordinating authority with the power to enforce solutions on decentralized actors and to provide them with adequate organizational structures.
To discuss this general hypothesis, the article analyses the role that four different context factors, namely: (1) the level of decentralization in the management of policy; (2) the role of regional political factors; (3) the organizational features of policy performance; and (4) the degree and type of stakeholder participationhave on the effectiveness of decentralized policy management.By doing so, the article follows Winter's (2012) recommendation that implementation research can continue to improve only by 'developing and testing partial theories and hypotheses rather than trying to reach for utopia in constructing a general implementation theory' (p.265).Hence, the role played by these context factors in different stages of the implementation process is individually analysed through the use of specific theoretical propositions.Finally, how they are integrated to produce policy in multilevel settings is discussed.Overall, in line with Piattoni's (2015) definition of MLG, the analysis of these four selected dimensions evidences how the main actors within an MLG system (technical actors, politicians and partners) hierarchically and horizontally interact to produce policy.
The analysis of these four specific components of MLG therefore helps to discuss the general hypothesis.It does so by assessing the different multilevel and multi-actor implementation dynamics, and by unpacking those specific elements of the implementation process that are linked to distortions in policy implementation (Lester & Goggin, 1998).In this regard, Pressman and Wildavsky (1973) argued that the 'complexity of joint action' can result in an 'implementation gap' between intended policy and results.Moreover, in research on Europeanization, the compliance concept has been used to capture the degree of 'conformance implementation' (Thomann & Sager, 2017a, p. 1254), understood as the extent to which a policy is implemented from top to bottom (Barrett & Fudge, 1981).
By focusing on policy outputs, this article conceives an implementation gap as (the extent of) non-compliance with intended policy (Thomann, 2015, p. 178) brought about by non-conformance with EU regulations (Thomann & Sager, 2017a).Six theoretical propositions related to the four above-mentioned dimensions are used for the purpose of the article.These propositions interpret the emergence of a possible implementation gap (dependent variable) as a result of a particular context factor (independent variable).Formulated by combining the literature on policy implementation, which discusses the impact of multilevel and multi-actor arrangements on policy delivery, and on MLG in the context of EU studies, these propositions are presented below.In the third section they are connected with specific sub-dimensions of the OP implementation process to afford better understanding of how their analysis enables the exploration of the relevant properties of multilevel policies that generate different implementation dynamics (Thomann & Sager, 2017b, p. 1398).

Decentralization
The implications of how public goods and services are allocated between the central and the local level have been discussed in depth in the policy implementation literature since the 1970s.In their pioneering study, Pressman and Wildavsky (1973, p. xvi) viewed the multiple levels of government in the United States as a cause of implementation delays.They postulated that implementation is more problematic in the presence of many decision-making points.This interpretation of multilevel structures as obstacles to implementation was shared by Bardach (1977), who advocated greater control to remedy unpredictability.This also seemed to hold for EU policy-making, since both regional participation in the EU arena and the devolution of tasks and responsibilities at subnational levels increased the complexity of policy-making due to high transaction costs required for the coordination of multiple jurisdictions (Hooghe & Marks, 2003;Scharpf, 1988).According to Maggetti and Trein (2019, p. 359), the problems may have also been a consequence of opportunist behaviour, and the possible conflicts on competences between the regional authorities and the central state.
To avoid these coordination costs, some scholars suggested that, in multilevel systems, decentralization should be backed up by central control (e.g., Benz & Eberlein, 1999;Matland, 1995).Hanf and Scharpf (1978), for example, emphasized the necessity: to provide central government with the capacity for formulating and putting into effect comprehension and integrated policies, implemented through instruments of central control and designed to ensure that lower units will be more effectively guided by the policy objectives of more inclusive levels of government.(p. 2. emphasis added) Similarly, in the field of EU integration the debate centred on whether subnational authorities are capable of improving EU policy-making without the supervision of central governments (Piattoni, 2010, p. 18;Pollack, 1995).
According to this article, those instruments of central control able to resolve possible problems of cooperation during programme implementation, and which are therefore likely to be effective within a multilevel setting, refer to specific government measures giving a central authority formal responsibility for coordinating and supervising the overall programme implementation, and the relative activities of the decentralized implementation bodies.On the basis of a coordination principle, these central instruments are then expected to help these bodies perform their tasks.The following theoretical proposition is therefore advanced: Proposition 1.In the case of multilevel and multi-actor policies, we find fewer implementation gaps if there are instruments of central control than if there is none.

Organizational features
The importance of organizational features in avoiding implementation gaps in programme implementation has been frequently discussed in policy implementation studies since the 1970s.Pressman and Wildavsky (1973, p. xv) emphasized the significance of establishing adequate bureaucratic procedures and the presence of both sufficient resources and a clear system of responsibilities for effective policy implementation.The positive effects of these elements were also pointed out in the context of EU integration, with several scholars arguing that to avoid implementation gaps, EU policies must have adequate organizational structures and human resources, and enough available support functions in the right combination (e.g., Bachtler et al., 2014;Dimitrakopoulos & Richardson, 2001;Milio, 2010).Moreover, according to Leonardi (2005), MLG requires all relevant actors to fully participate at different levels, and that effective institutions with adequate organizational features should carry out their due tasks.On this issue, some scholars (e.g., Bachtler et al., 2014;Bondarouk et al., 2020) have added that the effective implementation of EU policy programmes needs stable organizational structures throughout the programming period in order to prevent a reorganization of responsibilities among implementing bodies that may induce a loss of accumulated policy knowledge on EU regulations.
The following theoretical propositions are therefore advanced: Proposition 2. In the case of multilevel and multi-actor policies, we find fewer implementation gaps if there is availability of suitably qualified staff than if there is not.
Proposition 3. In the case of multilevel and multi-actor policies, we find fewer implementation gaps if there is organizational stability than if there is not.

Political factors
The literature identifies the interest of regional governments in promoting a policy as an influential political factor determining programme implementation performance in multilevel systems.Indeed, Piattoni states that 'the explanation of the differentiation in policy efficiency … lies in the different capacity and willingness of the regional political class to promote adequate requirements for implementing the funds' (Piattoni, 1998, p. 50).Similarly, for several scholars (e.g., Bache & Jones, 2000;Benz & Eberlein, 1999;Milio, 2010;Pollack, 1995;Smyrl, 1997), the implementation of EU multilevel programmes depends on the institutional capacity of the regional political class to build and sustain policies over the years.Hence, this article argues that leaving the management of a multilevel policy to regional political actors without central control creates implementation gaps related to different contextual political variables.It does so particularly in the presence of a regional government's different interest in promoting and sustaining the policy throughout the implementation period.The following theoretical proposition is therefore advanced: Proposition 4. In the case of multilevel and multi-actor policies, we find fewer implementation gaps if regional government has a political interest in promoting and sustaining a policy over the years than when it does not.

Stakeholder participation
From a normative standpoint, several different actors contributing information, consultation and participation are expected to improve the quality of decision-making (Piattoni, 2010(Piattoni, , 2013(Piattoni, , 2015;;Trein et al., 2019).However, some empirical studies show that the lack of stakeholder participation is a disruptive factor at the implementation stage (e.g., West, 2005).Consequently, Ongaro et al. (2019) emphasize the necessity to create conditions for the empowerment of stakeholders, in order 'to help these actors develop their capabilities and to provide room for their interactive involvement in governance' (p.109).They agree with Piattoni that 'although MLG arrangements challenge the ideal-typical notion of Westphalian state, their ultimate impact is determined by the mobilization capacity of all actors involved' (Ongaro et al., 2019, p. 109, emphasis added).
As a large body of research in network settings/collaborative implementation has shown, in fact, differences in stakeholders' mobilization capacity depend on several factors, including trust and the relative stakeholders' understanding of both the benefits and risks of participating (Ansell & Gash, 2008); their interdependence with other stakeholders (Ansell & Gash, 2008;Innes & Booher, 2018); and the presence/absence of formal and informal incentives and disincentives (Fischer, 2012;Hui & Cain, 2018).Hence, it has become increasingly evident that stakeholders' effective participation in programme implementation is conditional upon their ability to present their views at the negotiating table, and upon their access to adequate technical resources and expertise (Culpepper, 2002;Maloney et al., 1994).Since these requisite resources and capacities are rarely distributed equally among stakeholders, this article maintains that it is first necessary to create conditions for their empowerment and to increase their mobilization capacity (Ongaro et al., 2019), and then to devise new opportunities structures to involve them more closely in the policy process (Ansell et al., 2020).Accordingly, in line with the general hypothesis advanced above, in multilevel settings these factors causing a possible lack of stakeholders' mobilization can be overcome in the presence of a high level of government able to stimulate a wider stakeholders' participation during the implementation phase by creating new opportunities structures.
The following theoretical propositions are therefore advanced: Proposition 5.In the case of multilevel and multi-actor policies, we find fewer implementation gaps if there is mobilization capacity among all the actors involved than if there is none.
Proposition 6.In the case of multilevel and multi-actor policies, we find fewer implementation gaps if a higher level of government creates the conditions for mobilizing all the actors involved than if it does not.
Against this theoretical background, the next section will explain why the study of Cohesion Policy implementation in Italy and Spain is a testing ground for understanding how MLG works.

EU COHESION POLICY IMPLEMENTATION IN ITALY AND SPAIN IN PRACTICE
The adoption of Community Regulation No. 4253/1988 initiated a new era in EU regional policy because it completely reversed the logic adopted since the Treaty of Rome (Graziano, 2013;Leonardi, 2005) While the regulatory context (and the role of the European Commission -EC) changed over the course of these five multiannual programming periods (Casula, 2021a), the 2000-06 programming period was characterized by a decentralization of responsibilities in implementation, which was further stabilized in the following period (Piattoni & Polverari, 2016).From the 2000s, the responsibility for each OP was formally conferred to a domestic institutionthe so-called management authority (MA)that supervises its entire supply chain, that is, from the definition and approval of the OP to its final evaluation.
Depending on the varying emphases to MLG given by the regulatory context and the EC, the role of partners changed after the 1988 reform.In particular, the 'vertical' dimension of the partnership principlethat is, the interaction among European, national and subnational levels during the formulation, implementation and monitoring stages of the OPswas placed side by side with a 'horizontal' one through the involvement of private sector and socio-economic actors (Bauer, 2002).The horizontal dimension was then expanded during the following two programming periods to include 'any other appropriate body representing civil society, environmental partners, [and] non-governmental organizations' (Official Journal of the European Union, 2006, p. 39), in order to strengthen the application of the partnership principle at all OP implementation stages.This led to the establishment of a 'cohesion policy system of MLG based on decentralization of responsibilities and a stronger role for actors on the ground' (Commission of the European Communities (CEC), 2008, p. 18).
Keeping in mind the theoretical background presented in the next section, even if EU regulations establish the 'rules of the game', domestic actors have some manoeuvring space to model their governance arrangements for OP implementation in accordance with their political-administrative structures and traditions (Casula, 2021b).In particular, they can choose between more centralized and decentralized systems for policy management.Moreover, in line with Piattoni's (2015) definition of MLG, and the choice to analyse the four dimensions under investigation in this article, OP implementation can be influenced not only by effective administrative structures but also by both regional political dynamics and the effective participation of vertical and horizontal partners.
Hence, the governance of Cohesion Policy implementation varied significantly among MSs (and sometimes within them), and in some cases it underwent changes over the course of the programming periods.Italy and Spain, for example, opted for different governance systems for implementing EU Cohesion Policy.
The implementation of Italian OPs was completely under the control of subnational governments, with the MA placed under the supervision of regions.Thus, regional actors decided on development strategies and investment priorities, drew up the programming documents, and negotiated them with the EC; they decided on the content of public calls, as well as the features of regional institutions placed under their management and control; they established direct contacts with the final beneficiaries responsible for managing the projects; they were entrusted with creating a regional monitoring system and directly supporting the beneficiaries as they uploaded their data within an online system; they were made responsible for creating and consolidating evaluation units that were meant not only to produce frequent evaluation reports but also to actively support the overall programming.This system significantly differed from the one in Spain, which was centrally coordinated and where all activities were directly managed by a single MA located in Madrid, rather than multiple MAs for each OP placed under the administrative control of the regions.Through a documento de attribution de funciones (function attribution document), some of the MA's function could be delegated to the regions (which thus became Intermediate Bodies), while the MA still remained responsible for overall governance.In particular, regional actors had to manage the stage of project selection/approval on the basis of a common system of national rules, and in accordance with the MA's daily recommendations.Moreover, even if some of the monitoring structures were allocated to the regional level, a single national monitoring system with procedures consistent across regions was present.Only the evaluation was directly managed by regional governments.
In light of the foregoing description of the legal context, the next section will explain the empirical and analytical strategies used to assess the implementation performance of the Italian and Spanish OPs, and the factors influencing it.

RESEARCH DESIGN AND METHODOLOGY
Driven by the logic of comparison in qualitative studies (Casula et al., 2021;Mahoney & Goertz, 2004), this article focuses on a small-N, case-oriented comparison.Case studies and small-N comparisons are praised for their detailed analyses of processes (Rueschemeyer, 2003) and can be useful for both theory-building and theory-testing (Blatter & Haverland, 2012).They can be used to explore the impact of a large number of relevant factors and to confront analytical propositions with many data points (della Porta & Keating, 2008, p. 211).
The four regional Italian and Spanish OPs, financed by the ERDF during the 2007-13 programming period, are chosen here as units of analysis, and compared with each another.Against the above theoretical background, analysis of the regional ERDF OPs financed within the fourth programming period is a basis for understanding how different multilevel and multiactor arrangements affect policy implementation, for three reasons: (1) a decentralization of responsibilities to subnational actors; (2) the reinforcement of institutional capacities in domestic administrations, considered necessary to achieve greater economic efficiency; and (3) the partnership principle going from simple consultation to close cooperation, planned to be strictly applied at all stages of OP implementation.Moreover, the decision to examine the implementation of two OPs for each of the MSs considered allows empirical investigation of the following: (1) if central coordination produces similar effects among the regions; and (2) in the absence of central coordination, how regional contextual factors may lead to implementation gaps.As described in-depth in the third section, in fact, Italy and Spain significantly differ in the extent of their multilevel and multi-actor policy-making Indeed, 'despite the highly decentralized political system in Spain, the central government plays a strong role … in programming as well as in the OP implementation phase' (European Parliament -Directorate-General for Internal Policies -Policy Department B: Structural and Cohesion Policies, 2014), because established in this MS is a more general centralized system jointly with the attribution of responsibility to an overall 'national garrison' (Casula, 2020), which has more instruments of central control with which to supervise the entire OP supply chain.In line with the theoretical background set out in the next section, those instruments extend beyond an administrative basis.They allow, in fact, the presence in Spain of a central coordination authority able to supervise the programme's implementation, and to support the activities of the decentralized implementation bodies.
When a comparison strategy for similar cases (della Porta & Keating, 2008, p. 214) is adopted, the four regionally implemented OPs are similar from an economic standpoint.They were therefore selected against the same backdrop of EU regulations for OP implementation.To clarify, the four regions considered (Campania and Calabria for Italy; Andalusia and Galicia for Spain) were part of the so-called convergence area, which included the most underdeveloped European regions, that is, those with a gross domestic product (GDP) per capita below 75% of the EU average.Given the theoretical propositions advanced, the following subsections describe the analytical and empirical strategies used to investigate possible implementation gaps.This methodological section concludes with the presentation of the data analysis procedure used.

Data
From an analytical standpoint, 13 dimensions were selected for each of the five implementation stages of a funded ERDF OP.These dimensions were adopted from previous analyses in the academic and non-academic literatures (e.g., Bachtler et al., 2014;Leonardi, 2005;Milio, 2010), and modified in view of the policy's technical characteristics during the 2007-13 programming period.Given the aim of this article, and the propositions advanced in the next section, the use of these dimensions enabled us to see, very broadly, how organizational structures were modelled in different subnational contexts, andthrough the use of an evaluation perspectiveto empirically assess whether implementation gaps were present during any of the OP implementation stages.Each dimension was then used to measure a specific component of MLG, and its related proposition.Tables 1-5 summarize this analytical strategy by connecting each dimension (and the relative acronyms used in the body of the article, as indicated in the first column of Tables 1-5) to a specific theoretical proposition (fourth column), on the basis of the main actors and/or structures involved, as required by EU regulations (third column).Tables 1-5 also describe how implementation should have happened ideally, that is, without any implementation gaps (second column).

Methods
With regard to the empirical strategy, a preliminary desk analysis was conducted.EC reports, national and regional programme documents, and evaluation studies were consulted, as well as data related to commitment, expenditure, decommitment and system effectiveness.Furthermore, a total of 80 semi-structured interviews were carried out from 2016 to 2020. 1  All the primary and secondary data collected were then triangulated to explain how the independent variables listed in the six theoretical propositions advanced in the second section affected the implementation process.Used for this purpose was analysis of the presence/absence of implementation gaps for each of the 13 dimensions of analysis selected.
Following Thomann andSager (2017b, p. 1398), an evaluative perspective was adopted to measure the implementation performance of each dimension: specifically, a bottom-up qualitative approach based on selected indicators that adapted the Institutional Development Framework (IDF) method (USAID, 2000) was used.By following previous technical studies that had employed this method for evaluating regional policies (European Community, 1999;Ernst & Young, 2003;Oraže, 2009), it was possible to assess the complete programme management cycle through the use of specific criteria to determine where, along a development continuum, each of the 13 dimensions was situated: the terms 'consolidated/strong' (A), 'significant' (B), 'moderate' (C) and 'absent/weak' (D) were used to categorize variations in the implementation performance.This assessment is summarized in Table 6, where this continuum is used to evaluate the extent to which the systems related to each component of the OP fulfil the EU's requirements.On the basis of this continuum, implementation gaps − quantified as a discrepancy between intended policy (as required by EU regulations and presented in the second column of Tables 1-5), and the actual results that each OP achieved − are classified as totally absent in the case of the 'consolidated/strong' (A)' category and total present in the case of the 'absent/weak' (D)' one.
As regards the empirical exploration of the independent variables, Table 7 summarizes which of the 13 dimensions were used to empirically examine the six theoretical propositions advanced, and it specifies the empirical strategy used to test them.More specifically, Table 7 reports: (1) how the variables listed in the six propositionsinstruments of central control, availability of qualified staff, organizational stability, regional governments' political interest in sustaining the policy, stakeholders' mobilization capacity, and higher level of the government's ability to create conditions for stakeholders mobilizationwere defined for the purposes of the research (fourth column); and (2) how they were applied in practice (fifth column).As summarized in  the sixth column of Table 7, the IDF method was used for their assessment as well, with the elaboration of a development continuum criterion to evaluate the degree of presence/absence of each variable, and the triangulation of all the primary and secondary data collected.For this purpose, the terms 'consolidated/strong' (A), 'significant' (B), 'moderate' (C) and 'absent/weak' (D) were used to categorize variations in the selected variables.
The sources used to empirically assess the variables listed in the six propositions, and to categorize each dimension for the four regional OPs analysed, are presented in the Appendix in the supplemental data online.The implementation performance of the 13 dimensions selected was then categorized as 'consolidated/strong' (A), 'significant' (B), 'moderate' (C) or 'absent/weak' (D) for each of the four regions, and reported in the last columns of Tables 1-5.The third step provided an extension of information concerning the factors influencing implementation for each dimension.It did so through a categorization of the degree of presence/absence of the related variable(s) listed in the propositions that the single dimension was intended to measure.The results of this third step of data analysis enabled the integration of the information related to the main characteristics of the four subnational contexts analysed through the use of the development continuum criteria previously described.This information is reported in the fifth columns of Tables 1-5.
Finally, the aim of the fourth step was to constructfor each of the five stages of implementationa history of implementation performance for the four regional ERDF OPs based on the six theoretical expectations formulated.EU representatives, external evaluators and observers specialized in this research field were consulted to validate the results once they had been completed.These policy narratives are presented in the following section, while the sixth section discusses how the four context conditions analysed integrate with each other to produce policy in multilevel settings.

FINDINGS
The manner in which different multilevel and multi-actor arrangements operated in the four regions, along with their implementation performances, is presented in what follows.By elaborating the five stages of OP implementation, differences between the Italian and Spanish OPs, and between the regions of the same countries, will be described, bearing in mind the theoretical propositions previously presented.
The presentation of the findings follows the structure of Tables 1-5, which steer the progress of this article.As described in the methodological section, Tables 1-5 set out the results of the implementation process as regards the presence/absence of implementation gaps for each stage of an OP.These results of the implementation process are explained in terms of, first, the characteristics of the institutional context of the four regions analysed, and then of the specific component of MLG and the relative theoretical proposition that each of the 13 dimensions was intended to measure.Tables 1-5 categorize variations in both the implementation performance of the 13 dimensions (dependent variables) and the implementation-influencing factors listed in the theoretical propositions (independent variables), based on the assessment proposed in Tables 6 and 7, which summarize how the IDF method was used for the purposes of this article.

Programming
Implementation gaps were not found in the programming stage of the two Spanish ERDF OPs.The implementation process (Prog1) and the programme documents (Prog2)both dimensions classified as 'consolidated/strong (A)'were in fact clearly defined, and responsibility was attributed to a central coordinating authority ('Proposition 1'), which in Spain was identified with the MA (Applica, IsmeriEuropa, 2016a).As reported in Table 1, the variable 'instruments of central control' was then classified as 'consolidated/strong (A)' in the two Spanish ERDF OPs, since 'instruments of central control [were] present, well-defined and well-implemented'.Moreover, by displaying a great capacity for mobilization ('Proposition 5'), stakeholders were able to contribute information that improved the general quality of programming.Due to the presence of stable organizational structures ('Proposition 3'), the negotiation of the OP (Prog3) was speedy and without technical problems. 2financial crisis.Nevertheless, under strong solicitation by the MA, the regional governments decided to improve the use of financial aids, and to involve the banks in the certification of expenditure.Because of these corrective actions, there was no decommitment (ProjectMan3).In Andalusia, moreover, the MA assigned direct responsibility for supporting local beneficiaries to the Regional Development Authority (CEC, 2013c).
Conversely, implementation gaps were found throughout the entire management stage in Campania and Calabria, with ProjectMan1, ProjectMan2 and ProjectMan3 classified as 'absent/weak (D)'.Because this stage was completely decentralized in Italy at the subnational level, the regional administrations were reluctant to introduce the novelties required by EU regulations that concerned the separation between management and control functions (ProiectMan1) (Applica, IsmeriEuropa, 2016b).This resulted in direct intervention by the EC, which suspended payment for around 1.5 years.In fact, systems were not clearly defined, with major processing problems and frequent delays (D).This inevitably affected the stages of project payment (ProjectMan2) and project management (ProjectMan3) due to the participation of both the regional offices and the final beneficiaries at this stage.In the case of ProjectMan2, for example, while the [AQQ18]PA offices were poorly organized, with unstable structures and frequent internal reorganizations ('Proposition 3'), the staff of the final beneficiaries showed a lack of European culture in managing the project dossiers and the relevant expense documentation, with 'nearly all of them lacking highly qualified staff to interact with a regional monitoring system that is often very confusing' ('Proposition 2'). 7 Monitoring Implementation gaps were absent in both the Spanish regions during the monitoring stage for each of the three dimensions considered.As regards Monit1, the use of a monitoring system was perfectly in accordance with European standards and classifiable as one of the most effective and well-structured monitoring systems in Europe. 8In line with 'Proposition 1', this was due to the presence of a single monitoring system applied across the whole of Spain and developed by the MA in Madrid.Monitoring data were available throughout the programming cycle (Monit2), due to the presence, among the final beneficiaries, of qualified staff with a consolidated monitoring culture to ensure that monitoring data were promptly and correctly uploaded to the online system ('Proposition 2').However, monitoring for management purposes (Monit3) was increased within the Spanish regions during the last years of implementation of the 2007-13 ERDF OPs, due to increased political support from the regional governments ('Proposition 4').In 2015, for example, the Galician regional government assembled a regional task force 'to guarantee as much effect as possible of the monitoring data on decision-making'. 9This decision was accepted within the MC (CEC, 2013d), where stakeholders were strongly encouraged to actively participate in these strategic choices ('Proposition 6').
Once again, implementation gaps were found in both Calabria and Campania.In accordance with the analytical section, and in the absence of central coordination, implementation gaps at this stage can emerge due to a lack of interest by the regional governments ('Proposition 4'), which impeded the development of an adequate monitoring system in line with EU standards -Monit1 was classified as 'absent/weak (D)' in both regions.For example, doubts were expressed by the EC on several occasions about the Calabrian monitoring system, described at times as rudimentary. 10This had knock-on effects on the other two dimensions analysed in Calabria and Campania.With regard to Monit2classified as 'moderate (C)' in both regionswidespread training activities were not carried out by the managers of the operational objectives, and their high turnover rate did not allow for the consolidation of a coherent training system for the final beneficiaries ('Proposition 2').Moreover, with regard to Monit3classified as 'absent/weak (D)' in both Campania and Calabria since a lack of use of monitoring for programme management had been foundthe MC was 'a place for decisions that were already made, without concrete strategic actions being proposed, and with poor involvement from partners' ('Proposition 6'). 11Such was the case of Campania, where the regional government was 'interested only in obtaining the financial data when necessary for external publicizing', 12 and not to improve the general programming ('Proposition 4').

Evaluation
With regard to the evaluation stage, fewer implementation gaps were found in Galicia and Campania.On the other hand, evaluation was still far from completion in Andalusia and Calabria: a 'low quality of evaluation activities (C)', and a 'lack of use of evaluation for program management (D)' was found in both regions.
The absence of implementation gaps in both Galicia and Campaniawhere evaluation activities (Eval1) were 'established and well implemented (A)' and an 'institutionalized use of evaluation for program management (A)' (Eval2) was presentmust be attributed to the close attention paid by the regional government to the evaluation stage during those years ('Proposition 4').In Campania, for example, although the regional government underwent some change, evaluation was always acknowledged as 'a practice of decision-making support' (Eval2) (Regione Campania, 2013, p. 71).This also ensured a strong involvement of partners ('Proposition 6'), who found themselves able to use evaluation as support for general programming. 13In both regions, the findings showed a 'strong interest to increase stakeholders' mobilization capacity [,with] actions clearly defined and implemented (A)'.In particular, while training courses in Campania were organized to create the conditions for stakeholder mobilization, partners in Galicia were involved in the design of a regional evaluation plan.
Regional governments were not similarly invested in either Calabria or Andalusia, where the widespread of an evaluation culture was delayed.In Calabria, for example, the Evaluation Unit was not able to start a process of institutionalization (Eval1) (Regione Calabria, 2007), and a real will to consolidate an evaluation culture and improve the performance of the OP was lacking (Eval2). 14

DISCUSSION
The last column of Table 7 summarizes the main empirical points supporting each of the six theoretical propositions formulated from the analysis of implementation performance for the four OPs.The analysis and its findings contribute to knowledge about the dynamics of implementation performance in EU multilevel systems, and they afford better understanding of how different components of MLG integrate with each other to produce policy.They thus further clarify how these data can contribute to supporting or challenging the conventional assumptions about the expected benefits of MLG.
The presence of various types of instruments of central controlparticularly coordinationcan help prevent implementation gaps in multilevel and multi-actor policies ('Proposition 1').The presence of control instruments, in fact, also directly and indirectly impacts on the other three factors considered.Even if the presence of suitable qualified staff ('Proposition 2') and stable organizational structures ('Proposition 3') positively correlates with better implementation performance, it is not true that this is always achieved in the absence of central coordination.This issue is connected with the role played by the third component of MLG considered in this article and related to regional political factors.The presence of a central coordination authority is also able to overcome a lack of interest by the regional governments in sustaining the policy ('Proposition 4').Similarly, although increased participation in decision-making of all vertical and horizontal partners is expected to produce better policies (output legitimacy), partners do not necessarily possess the knowledge and expertise to take part in EU policy implementation ('Proposition 5'), and their mobilization may depend on whether it is stimulated by higher levels of government ('Proposition 6').
These findings contribute to the long-running debate among political scientists on policy implementation in different multilevel settings.In particular, they shed light on those aspects of implementation performance related to the crucial policy design/implementation dynamic (see also Kaufmann et al., 2020;Winter, 2012).Whilst the fact that the design of public polices can play a crucial role in programme implementation was recognized by pioneering implementation scholars, academic interest in this link has waned in recent decades.Nevertheless, according to Saetren and Hupe (2018, pp. 571-572), its study can continue to represent an advancement towards a more general theory in this field of research.This is particularly the case in the context of EU integration, where different MLG policy-settings for policy implementation are needed, and different regional actors and stakeholders are involved.As regards the aforementioned link, the findings presented allow us to advance observable implications on the way in which the presence of a central coordination authority can compensate for the potential pitfalls of MLG settings.More in detail, these findings substantiate the thesis that multilevel and multi-actor structures obstruct implementation in the presence of multiple decision-making points, and in the absence of central control to remedy unpredictability.In other words, while generalization of results beyond these cases requires caution, MLG seems to improve policy implementation in the presence of a coordination authority that, besides possessing traditional command and control instruments, is able to ensure the real involvement of all the actors during the implementation process, doing so in particular through the creation of new opportunities structures that guarantee an equal distribution of those actors' involvement.As the findings suggested, in fact, effective MLG can benefit an authority that mobilizes implementation bodies, not only by providing actors with adequate organizational structures and the technical and professional resources that they lack, but also by creating conditions for consolidating the participation of stakeholders when they lack a similar mobilization capacity.The same central authority can guide an ongoing collaboration design/redesign process which may be needed over the years to reflect emerging management challenges better, and to ensure a high degree of cooperation when multi-actor and multilevel arrangements are established.

CONCLUSIONS
In recent decades, MLG has been increasingly used as an instrument of effective policy implementation.While the greater involvement of various actors at different levels of government is expected to produce better policies, the literature is in agreement that more empirical studies are needed to analyse if and how different MLG systems impact policy implementation.Adopting a performance-oriented approach to EU implementation, this article has contributed to this academic debate by investigating how different context conditions affect the effectiveness of decentralized policy management.By combining traditional literature on policy implementation with the literature on EU studies, this article has developed a set of interrelated theoretical propositions to guide empirical research on the various stages of implementation of an EU development programme in four different regions belonging to two MSs characterized by different organizational and institutional arrangements.
The findings support the hypothesis that the presence of a central coordination authority can compensate for the potential pitfalls of MLG settings, and that it is therefore helpful to guarantee a better implementation performance in the EU multilevel system.However, it should be borne in mind that the empirical research reported in this article adopted a logic of qualitative comparison focused on a small-N comparison.Hence, caution is necessary when proposing possible generalizations of these results not only to other EU policy sectors where MLG systems are used for policy implementation but also to other MSs where Cohesion Policy is implemented.As explained, policy implementation can be affected by different policy traditions, institutional architectures, and other factors related to policy specificity.The decision to decentralize to subnational governments in Italy, for example, led to the consolidation of very different regional governance systems for the management of EU Structural Funds.Those in Spain proved to be more homogeneous (apart from the institutionalization of the evaluation units).The analysis of how these different multilevel and multi-actor systems concretely operate and their effect on EU policy implementation has been the empirical added value of this article for public administration scholars interested in Cohesion Policy.It also assists theoretical reflection on the application of the 'partnership' principle itself.If comparative studies on how this principle operates across the different policy stages that compose an OP and how it affects its policy performance continue to be limited because of the large amount of field research work required, this articlewhich has likewise required a large amount of research over the yearsconfirms the political (and not only technical) conception of the partnership principle.It therefore substantiates the argument that the principle's correct application cannot ignore the institutional and legal framework of each MS within which it is implemented.
To conclude, considering that the perspective adopted in this article seems to be a productive way to provide a comprehensive outlook on research in the field of policy implementation and MLG, the future challenge will likely be the conversion of the theoretical propositions presented into testable hypotheses.This will make it possible to assess the level of intensity with which different factors related to organizational features, subnational political dynamics, and stakeholder participation are able to influence policy performance, when such a central coordinating authority is present.The aim could also be to perform comparative studies in different European policy sectors and countries, and to investigate the effect of different contextual factors and governance models in order to evaluate the usefulness of implementing through MLG.This means, as Piattoni puts it, that not only input legitimacy, but also output legitimacy would be ensured.
Structural Funds worked together to support regional development in Europe, namely: the European Regional Development Fund (ERDF); the European Social Fund (ESF); the Cohesion Fund (CF); the European Agricultural Fund for Rural Development (EAFRD); and the European Maritime and Fisheries Fund (EMFF).

Table 1 .
Explaining the results of the implementation process: the programming stage.: EC, European Commission; MA, management authority; MLG, multilevel governance; OP, operational programme. Note

Table 2 .
Explaining the results of the implementation process: the project selection/approval stage.

Table 3 .
Explaining the results of the implementation process: the project management stage.: CA, certifying authority; MA, management authority; MLG, multilevel governance; PA, payment authority. Note

Table 4 .
Explaining the results of the implementation process: the monitoring stage.: EU, European Union; MA, management authority; MC, monitoring committee; MLG, multilevel governance; PA. Note

Table 7 .
Components of multilevel governance, theoretical arguments, dimensions of analysis and main empirical evidence.