Linking Policy Design and Implementation Styles

Linking Policy Design and Implementation Styles Jale Tosun and Oliver Treib Published in: Howlett, Michael and Mukherjee, Ishani, eds. (2018) Routledge Handbook of Policy Design. London: Routledge, pp. 316–330. Abstract In this contribution, we bring together two strands of research that have so far existed side by side: the scholarship on policy design and the literature on policy implementation. Designing policies involves identifying goals and selecting policy instruments with which the goals can be reached. Policy implementation is about putting these policies into practice and comprises the definition of an implementation structure, decision-making within agencies, target group behavior, and policy results. We propose to connect the stages of policy design and policy implementation by means of the concept of policy feedbacks. Based on the two traditional approaches to studying policy implementation, we distinguish two ideal typical implementation styles. Policy delivery can be organized in a centrally controlled hierarchical manner, where fixed policy intentions are expected to be handed down to addressees with as little deviation as possible (centralized implementation), or it can be organized in a more decentralized manner which leaves implementing actors more leeway for adapting the policy to local circumstances (decentralized implementation). We argue that while decentralized implementation offers more insights into what makes policy work than centralized implementation, the latter is more likely to produce policy feedbacks that inform future policymaking and potential redesign of the policy concerned. 1. Introduction Policy studies address a remarkably wide range of topics and cover a multitude of levels of analysis. The most micro-level research explores the behavior of individuals (e.g. John et al. 2009; Sager et al. 2014; Weaver 2015), whereas the most macro-level research searches for country-comparative patterns of policy decisions (e.g. Falkner et al. 2007; Falkner and Treib 2008; Damonte et al. 2014; Tosun 2013; Mortensen and Green-Pedersen 2015). At first glance, these two extreme ends of public policy do not seem to be conformable with one another, but upon closer inspection, the differences between them start to vanish. What connects these two research perspectives is the mutual interest in how the behavior of individual and collective actors in society can be steered to attain a given goal. An alternative viewpoint that connects the micro- and the macro-level is how past policies’ feedbacks shape future policies (e.g. Mettler and SoRelle 2017). The literature on policy feedbacks typically takes policy outcomes or impacts – here jointly referred to as policy results – as the starting point, but pays limited attention to how decisions regarding policy design or implementation 1 style may matter for producing such feedbacks. This gap is the point of departure for this chapter, in which we seek to connect the stages of policy design and policy implementation by means of the concept of policy feedbacks. Policy design involves two sets of decisions. First, decisions regarding the substance of a policy in terms of what it aims to achieve, and which instruments are conducive to the policy’s goal. Second, the procedure of a policy in terms of which level of government and agency is responsible for implementing the policy concerned (e.g., DeLeon 1988; Howlett and Rayner 2007; Howlett 2009; Howlett et al. 2015; Capano et al. 2016; Chindarkar et al. 2017). The latter also includes considerations regarding the extent to which the individual implementing agencies concerned have some leeway in adapting the policy to local-level conditions. A constellation where policymakers expect fixed policy intentions to be handed down to addressees with as little deviation as possible is referred to as centralized implementation. The alternative constellation is decentralized implementation, which is characterized by leaving the implementing actors more leeway for adapting the policy to local circumstances. The analytical considerations underlying these two implementation styles and their effects for policy delivery have been discussed widely in the literature (e.g. Pülzl and Treib 2007; Damonte et al. 2014; Saetren 2014; Hupe and Sætren 2015; Hupe and Hill 2016). We seek to make a novel contribution to the literature by contending that the choice between centralized and decentralized implementation has implications for the production of policy feedbacks and therefore the dynamics and potential results of subsequent decisions on policy design. This chapter unfolds as follows. First, we introduce the concepts on which we elaborate in this chapter. Next, we introduce our integrated conceptual model, which is followed by suggestions for research designs suitable to test the empirical implications of the model. In the concluding section, we summarize our main arguments and offer some ideas for future research. 2. Clarifications on the key concepts In this section, we offer definitions of concepts that are integral elements of our conceptual model. The concepts concerned are policy tools, policy design, policy implementation, and policy feedbacks. We discuss each of the concepts in turn. Policy tools Recent developments in policy studies make it necessary to position ourselves with regard to the types of policy tools we analyze here. In the last few years, a literature on ‘new’ policy tools 2 based on the insights of behavioral economics has emerged. Commonly referred to as ‘nudge policies’, from the title of the influential book by Thaler and Sunstein (2008), the research program underlying this type of policies corresponds to that of heuristics and biases as put forth by the path-breaking work of Kahnemann et al. (1982). Nudge policies aim to bring about behavioral change by capitalizing on systematic cognitive and behavioral biases and thereby directing individual behavior toward a more beneficial outcome (Grüne-Yanoff and Hertwig 2016: 152). The appeal of nudge policies for policymakers derives from the fact that they promise to bring about effective behavioral change without using the classical, heavy-handed repertoire of policy tools such as coercion or financial incentives. This promise, however, may only hold for a narrow set of policy problems. Often, incentives for non-compliance may be too high and citizens or economic actors may have other reasons than a lack of the appropriate choice architecture for behaving the wrong way, such as a lack of resources or deeply held beliefs (Weaver 2015). There are thus good reasons to hold on to the classical tools of government action. Even though nudge policies have been applied in several contexts (e.g., John et al. 2009), many problems may still be tackled effectively only by resorting to the coercive force of the law, to financial incentives or disincentives, or to direct public service provision (e.g. Hood 1986; Howlett and Ramesh 1993). Most policies still employ these classical techniques, which is why we limit our analytical purview to this type of policy tools (but see Strassheim and Beck 2019). There are several analytical schemes for classifying classical policy tools, among which the socalled NATO scheme is particularly popular. This scheme differentiates between policy instruments based on the principal governing resource they use: policies using information to convince norm addressees of a particular goal (Nodality); policies that employ the coercive force of the law (Authority); policies that provide financial incentives or disincentives (Treasure); and policies that build or reform institutions to achieve their objectives (Organization) (Hood 1986; Hood and Margetts 2007). Another way of classifying policy tools is to differentiate between substantive and procedural policies. Substantive policy tools are intended to directly affect the quality and the quantity of the goods and services provided in society. Procedural policy tools are more diverse than substantive policy instruments and include the provision of information as well as the creation and reform of institutions (see Howlett 2005: 35-36). 3 Our decision to stick to classical policy tools has important consequences for the subsequent analysis. With nudge policies, policy implementation predominantly requires the policy addressees to change their behavior. Consequently, the implementation process is rather straightforward and simple. With classical policies, the implementation process is more complex, involving behavioral changes by the target groups, but also decisions with regard to the overall implementation structure and the involvement of governmental and nongovernmental organizations. Policy design Research on policy design has concentrated on (bundles of) policy tools that public actors choose in order to bring about behavioral changes (see, e.g., DeLeon 1988; Howlett and Rayner 2007; Howlett 2009; Howlett et al. 2015; Capano et al. 2016; Chindarkar et al. 2017; Howlett and Mukherjee 2017). The main question underlying the relevant scholarship is how policies can be constructed in such a way that they produce the desired policy results (Howlett et al. 2015; Colebatch 2017). From this, it follows that research on policy design is tightly linked to both questions of policy success and implementation. In this context, Howlett (2011) argues that the birth of this specific literature in the mid-1980s was motivated by policy scholars’ wish to improve their knowledge of how the choice of policy tools affects the choice of implementation structures and tools (but see Sager and Rielle 2013). While there is a close relationship between implementation and the success of a policy, the literature on policy design also emphasizes that a policy may fail due to an invalid theory connecting policy actions and desired outcomes. Consequently, according to this perspective, poor implementation is not necessarily the reason for poor policy outcomes (see Capano and Woo 2017). More recently, the perspective of research on policy design has moved from single tools to multiple tools and tool mixtures used to address policy problems. The main argument of this analytical perspective is that “efforts of policy makers often have failed due to poor designs which have failed adequately to incorporate this complexity into policy formulation” (Howlett et al., 2015, p. 300). The broadened perspective of tool mixtures opens up an entirely new perspective since the individual elements of the tool boxes selected may conflict with each other. Thus the process of formulating and adopting policy tool mixtures needs to focus more on the actors involved, their preferred policy designs, and their role in decision-making as defined by the respective rules and institutions. Policy tool mixtures are not only more difficult to design than single tools, but also more difficult to implement. 4 Policy implementation The literature on policy implementation deals with how policies are put into practice by administrative actors and to what extent they affect the behavior of societal target groups (for an overview, see Pülzl and Treib 2007). Following Winter (2012) and Vancoppenolle et al. (2015), policy implementation consists of four dimensions: defining the implementation structure, agency decision-making, target group behavior, and policy results. Implementation structure refers to the number and types of organizations involved in the implementation process. It is often the case that two or more ministries or agencies are responsible for implementing a policy. This is especially true with cross-sectoral policies, where the various organizations in charge of policy implementation need to coordinate their activities in what is known as horizontal coordination (e.g. Peters 2015). For example, in order to implement the Paris Agreement on Climate Change, the German government adopted the Climate Action Plan 2050, which stresses the need for horizontal coordination among several ministries at the federal level. In multilevel polities, policy implementation also often includes vertical coordination among organizations located at different levels of government (e.g. Bolleyer and Börzel 2010). Policy implementation may also require the involvement of private actors (e.g. Thomann et al. 2016; Tosun 2017; Tosun et al. 2017). In addition to the types of organizations or actors involved, the implementation structure includes policy-related rules and procedures and the allocation of resources. Agency decision-making refers to the process of making the legal stipulations more concrete and therefore implementable. This requires the members of one or several competent organizations to decide on these concretizations and to elaborate a procedure for the delivery of the services, which also includes considerations about the leeway for managerial discretion and the scope and nature of the involvement of street-level personnel (see, e.g., Sager et al. 2014). Sometimes street-level personnel benefits from great discretion in making decisions relevant for policy delivery such as in social work (e.g. Ellis 2007), whereas in other cases, their leeway for discretion is limited, as is often the case in policing (e.g. Rowe 2007). Target group behavior refers to the role the policy addressees play in the implementation process—their actions as well as their needs (see Pierce et al. 2014). To obtain social benefits, for example, the target group’s involvement in the implementation process consists of filing a request and complying with the conditions attached to the benefits (e.g., regular appointments with case managers). Implementation of other policies may entail more substantive changes to the behavior of the target groups (see Howlett 2005: 47). In the case of promoting biofuels, for 5 instance, policy implementation crucially depends on the behavior of the fuel producers and suppliers as well as on consumers’ buying choices (see Tosun 2017). Policy results are the outcome of the implementation process, which can correspond to the intentions of the policy concerned or deviate from them. For example, if a government seeks to reduce air pollution, it can adopt stricter limit values for the emission of air pollutants. The policy result would correspond to the intentions if levels of air pollution were diminished. The policy result would deviate from the intention of the policy if there were no changes or an increase in air pollution levels. If the policy results deviate from the stipulated intentions, the reason for this can be problems encountered in the implementation process or flaws in the design of the policy. Of these four elements, policy studies have devoted most attention to implementation structure and agency decision-making. With regard to the implementation structure, Howlett’s (2004, 2005) work on ‘implementation styles’ has been particularly influential. Implementation styles denote typical combinations of substantive and procedural policy tools that are employed to achieve given policy goals. Howlett differentiates between the severity of state constraints (with regard to resources and legitimacy) and the nature of the policy target (which relates to exchange or policy actors) in order to identify four ideal-typical implementation styles. Institutionalized voluntarism is the implementation style of choice when the policy targets constitute a large group and state constraints are high. It corresponds to an exhortation-based manipulation of market actors and the institutionalization of networks. For example, in 1985, the Chemistry Industry Association of Canada launched the Responsible Care Initiative, which is a complementary scheme to regulation to facilitate the self-control of the industry and to increase public and political trust in its activities. Regulatory corporatism is chosen when state constraints are high, but the policy targets constitute a small group. This implementation styles regulates market actors and manipulates their interest-articulation system by means of financial incentives. It corresponds to a ‘corporatist’ style economic planning models in industrial policymaking (Howlett 2004: 10). Levi-Faur (2010) refers to regulatory corporatism as a system where strong civil and state regulation co-exist. Of the four forms identified by Howlett (2004, 2005), this one has the most coercive character. 6 The third implementation style, directed subsidization, is expected to be chosen when governmental actors face low constraints and interact with large groups of policy targets. Directed subsidization refers to the extensive use of financial instruments to steer the behavior of market actors, coupled with the use of authority to preferentially recognize networks of actors (Howlett 2004: 10). For example, the German government has decided several years ago to promote the transition to a bio-based economy, which it supports by means of subsidies for the industries concerned as well as investing heavily into research activities that could potentially support this process. Lastly, the fourth implementation style, public provision with oversight, corresponds to a mobilization model, where governmental organizations use resources to provide goods and services to small groups of policy targets. At the same time, the government seeks to manipulate actor networks through information. This implementation style is chosen when the number of policy addressees is small and so are the constraints on the state (see Howlett 2004: 10). The state-based provision of internet in Estonia can be regarded as a case in point for this implementation style. Turning to agency decision-making, we enter the realm of a long established strand of policy studies. This strand of research has long been occupied with controversies between two opposing schools of thought. Top-down scholars stress the role of centrally defined policy goals that were to be executed by a hierarchically ordered administrative structure with as little deviation from the original stipulations as possible (Pressman and Wildavsky 1973; Van Meter and Van Horn 1975; Sabatier and Mazmanian 1979). Proponents of the bottom-up camp, in contrast, argue that policies are often implemented in a much more decentralized way, leaving individual street-level bureaucrats significant leeway in adapting the policy to local circumstances (Elmore 1979; Lipsky 1980; Hjern and Porter 1981). To overcome the often unfruitful clash between the opposing normative assumptions about ‘good’ policy delivery underlying these two schools of thought, a third group of scholars decided to develop integrated analytical models. The hybrid models bring together elements from both approaches and treat the question of whether actual implementation processes resembled more the bottom-up or the top-down approach as an empirical question (Goggin et al. 1990; Winter 1990, 2012; Matland 1995). Policy feedbacks The final concept we need to define before setting out our conceptual model refers to policy feedbacks. Most fundamentally, policy feedbacks are conceived as reactions by the policy 7 addressees or other parts of the society to policy decisions adopted in the past (e.g. Pierson 1993; Béland 2010). Mettler and SoRelle (2017) point to four dominant streams in the study of policy feedbacks. The first stream analyses how policies affect political agendas and the definition of policy problems. The second stream extends the logic of policy feedbacks to how they affect governance through their impact on the capacity of government and political learning by public officials. The third stream concentrates on how policy feedback influences the power of societal groups. The fourth stream concentrates on the study of individual political behavior by examining how policies shape the meaning of citizenship. To this framework, we add a fifth stream of research, which asks how necessary but unpopular policies can be designed to become ‘sticky’ and resilient towards attempts to dismantle them. For example, Jordan and Matt (2014) applied this perspective for the analysis of climate change policies. Of the five streams presented, the second stream on governance is most pertinent for our analysis. The corresponding literature argues that policies may affect future governance as they shape the feasible set of policy choices and the type of organizational arrangements assigned to new policies, and the broader scope of government action as well as political learning by public officials. According to Mettler and SoRelle (2017), such outcomes are likely to emerge if new policies enable governments to develop capacities they did not have previously, which they can then use for the delivery of policies developed subsequently. Following Pierson (1993), whether a policy is subject to redesign depends on whether the feedback is positive or negative. Positive feedback is expected to preserve the status quo, whereas negative feedback is likely to trigger attempts to redesign policies. Policy feedbacks can, in principle, stimulate or reinforce policy learning (Van der Knaap 1995), which, however, depends on the respective policymakers’ willingness and capacity to learn. From this, it becomes clear why policy feedback has been frequently discussed in the context of policy change (e.g. Béland 2010), while this concept has received limited attention from scholars working on policy implementation. 3. An integrated model of policy design, implementation style and policy feedback Our objective in this section is to develop a model that bridges the different strands of literature and integrates the concepts of policy design, implementation and policy feedback. We believe that such an integrated model is useful for, on the one hand, aligning the established literature 8 on policy implementation with the literature on policy design, and on the other hand, to connect policy studies with comparative politics by drawing on the concept of policy feedback. The starting point of the conceptual model is the stage of policy design, which results in two sets of decisions: the selection of the policy tools and the selection of the implementation structures. The selection of policy tools can be examined from the perspective of policy studies and comparative politics alike; both perspective offer important insights and highlight complementary aspect. Starting with policy studies, a recent conceptual note by Chindarkar et al. (2017) stresses the importance of capacity for the quality of policy design outcomes. The authors refer to two types of capacity: governance capacity and analytical capacity. When governance and analytical capacity are high, the authors expect capable design, while in the opposite constellation—where both capacities are low—poor design is the most likely outcome. When governance capacity is high but analytical capacity is low, the authors anticipate capable political design, where the policy is likely to be politically feasible but the quality of policy is low. The complementary scenario (low governance capacity, high analytical capacity) is likely to lead to poor political design, where the quality of the policy is likely to be good, but it may not be politically feasible. The literature on policy design also ties into research on path dependency and incrementalism. Drawing on historical institutionalism, path dependency models presume that the best predictor of future policy design is past policy design (see Pierce et al. 2014). This expectation is clearly at odds with the model put forward by Chindarkar et al. (2017), which rests on the assumption that policy design is the outcome of policy analysis. Nevertheless, the two perspectives can be reconciled: in situations where the analytical capacity of governments is low, the most rational decision to take could be to continue doing what they have done in the past. Comparative politics regards policy design as the outcome of institutional arrangements and the preferences of individual actors. The literature on veto players (Tsebelis 2002) has demonstrated that it makes a difference for the outcome of the policy design process which actors are involved in decision-making and what their policy preferences look like. Policy preferences stem, for example, from the ideologies of political parties that form the government (Tosun and Workman 2017). Governments that are composed of left-wing parties generally tend to display a greater propensity to intervene in the economy, whereas right-wing parties are more in favor of voluntary measures (e.g. Bakker et al. 2015). In parliamentary systems in Europe, political parties often need to form coalition governments that consist of two or more 9 political parties. In such constellations, not only the ideology of one party is likely to determine the selection of policy tools, but the ideologies of several parties and the way in which they reach agreements. Depending on the decision-making dynamics within coalition governments, policy design might be poor due to the need to reach a compromise solution. For the same reason, policy design can turn out to be of a good quality since the political parties in government need to convince one another and engage in an analysis of the existing policy options, which then leads to the selection of the alternative that appears most promising. In the previous section, we followed the conceptual work of Winter (2012) and Vancoppenolle et al. (2015) and differentiated between the implementation structure and agency decisionmaking. When looking at policy implementation as a complementary dimension of policy design, the differentiation between these two elements becomes clear. We conceive the selection of the implementation structure to include decisions on how much discretionary leeway agencies and bureaucrats inside the agencies should be granted. This does not mean that the need for subsequent decisions becomes obsolete, but the general decision about whether the implementation style is centralized or decentralized is likely to be made at the stage of policy design. Above we discussed the four ideal-typical identified by Howlett (2004, 2005) which are useful for explaining variations across different policy domains in one countries as well as across countries. However, for introducing an integrated conceptual model, we must abstract from the analytical considerations on which Howlett’s typology rests. Similarly, we cannot draw on the various models of policy implementation that approach this phenomenon from the top-down, bottom-up or a mixed perspective (see Pützl and Treib 2007) since these rest on specific assumptions and theoretical considerations as well. Yet we draw loosely on the perspective of top-down and bottom-up models of policy implementation as the selection of the implementation structure will most fundamentally set out whether the implementation process is centralized or decentralized. Our simplified definition of the implementation structure consists of two dimensions: first, the number of agencies involved in implementation; second, the degree to which the agencies involved have discretionary powers to adjust the policy design. Centralized implementation is characterized by a small number of implementing agencies that have little to no discretionary power. Resonating with the logic of top-down implementation, a centralized implementation structure seeks to ensure that there is no deviation from the original policy design. A decentralized implementation structure comprises multiple agencies and allows them to deviate 10 to a certain degree from the stipulations of the policy in order to take into account the local context, assuming that the policy results will benefit from these adjustments to the policy concerned. There are several explanations for why the policy design process may result in the selection of a centralized or a decentralized implementation structure. One of the explanations is the political system’s institutional features. For example, we can observe that federal polities tend to opt for decentralized implementation structures, whereas centralized polities are more likely to select centralized implementation structures (see, e.g., Wälti 2004). Another potential explanation is the nature of the policy problem underlying the policy design. For example, it would be virtually impossible to define a centralized implementation structure for a policy that aims to replace conventional fuels by fuel blend that contains a certain share of biofuels. The production and distribution of fuels is the task of private mineral oil companies and therefore policy implementation in this particular case requires these companies to be included in the implementation structure (see Tosun 2017). In such arrangements where public and private actors cooperate, the latter typically also have some degree of discretionary power in implementing the policy (see Thomann et al. 2016). Centralized implementation results in information about the effectiveness of a policy when it is implemented in an identical or at least similar manner. In more abstract terms, the implementation process – if done correctly – produces repeated observations of whether a policy in questions produces the intended policy results. Seen from this perspective, centralized implementation allows, in principle, for gathering information about the quality of the policy tools selected. This implementation structure also allows for inferring insights on how the policy targets and other societal groups perceive of the policy goal and the policy tool concerned. Put differently, policy-makers can gather information on whether the policy goal they seek to attain and the policy tools they employ are accepted and considered legitimate. Decentralized implementation potentially results in variation of the policy design originally adopted by policymakers, which increases the information basis for the evaluation of the policy tool concerned. The success of policy implementation depends on whether the process facilitated learning, capacity-building and support-building in order to address problems associated with it in a decentralized way, consistent with the interests of the actors involved (see Schneider and Ingram 1997; Pierce et al. 2014). Relating this to the literature on polycentric governance (Ostrom 2010), each adapted policy design can be regarded as a ‘policy experiment’ that could improve policymaking through processes of upscaling. Likewise, decentralized 11 implementation can produce insights into which variations of the policy design help to improve the societal acceptance and legitimacy of the policy tool concerned. To sum up, the degree to which the implementation structure is centralized is likely to affect which information governments may infer from the implementation process for future policy processes. For this information to have an impact on future policymaking, it needs to be transformed into policy feedback. To this end, policymakers need to collect this information, analyze it, and decide, in light of the insights gained, whether a new policy process is needed. In line with the literature, we expect positive feedback to support the policy status quo and negative feedback to start a new policy process that aims to improve the design of a policy adopted previously (see Mettler and SoRelle 2017). As we have seen above, decentralized implementation has a greater potential to offer insights into how policies must be designed to be effective and accepted. This richness in information, however, represents a potential impediment to improving future policy design by means of policy feedback. This expectation results from three observations. First, an analysis of variations of one policy requires a high level of analytical capacity, which not all governments possess and are willing to invest (see Wellstead et al. 2011). Second, the multitude of agencies involved in the implementation process may make it difficult to communicate with the actors on the governance level that is responsible for policy design. For example, it could be possible that information on variations of a central policy as it is implemented at the local level only reaches policymakers at the regional level, who are not responsible for the policy design and lack the competence to start a redesigning process. Third, information on the effectiveness of the adapted policy and how it is perceived by policy addressees and the broader society might be inconsistent and potentially force policymakers to decide whether they want to adopt redesigned policy that is likely to be less effective, but finds more acceptance among the target population, or the other way around. To be sure, this is a rather hypothetical case since it requires an even more advanced capacity and willingness to conduct policy analysis. Furthermore, it should be noted that while this model can be applied to individual policy tools and policy mixes alike (see Howlett et al. 2015), the latter entails an additional complication. Altogether, despite the greater level of useful information that may result from decentralized policy implementation, we postulate that this information is less likely to function as a policy feedback than the more sparse information resulting from centralized policy implementation. To be even clearer, we argue that the excess of information can possibly hamper the transformation of information into policy feedback and the initiation of a new policy process. 12 Information originating from centralized implementation is more likely to produce policy feedback, but the information available is more limited. While this can trigger a policy process to redesign the policy concerned, the changes in the policy design are likely to concentrate on one or a few aspects. This can result in a sequence of redesign efforts, of which each concentrates on different aspects. Figure 1 provides an overview of the structure of our conceptual model, including the stages of policy design, the selection of policy tools, the selection of the implementation structure, the outcomes of this selection process (centralized vs. decentralized), and the information resulting from the implementation structure selected. Figure 1: A Model of Policy Design, Implementation Style, and Policy Feedback (1) Low level of analytical capacity required (2) Information channels straightforward (3) Information less likely to be inconsistent Policy feedback more likely Policy design Policy tools Centralized Limited but clear informationon effectivenessand acceptance Decentralized Extensive but diverse information on effectiveness and acceptance Implementation structure Policy feedback less likely (1) High level of analytical capacity required (2) Information channels complex (3) Information more likely to be inconsistent Source: Own elaboration. 4. Toward the formulation and testing of hypotheses The conceptual model illustrated in Figure 1 deliberately omits a number of details in order to highlight the most important elements of the relationship between policy design, policy implementation, and policy feedback. It therefore does not include many of the explanatory factors discussed in the specific literatures on these concepts either. For example, the literature on policy design stresses the importance of analytical capacity (see Chindarkar et al. 2017), which our model only takes into account with regard to the likeliness of policy feedback. 13 Howlett’s (2004, 2005) treatise of implementation styles elaborates on the severity of state constraints and the nature of the policy targets, which would also have to be incorporated into an extended model. And the literature on policy feedback argues that welfare policies are particularly likely to produce policy feedback, which means that the type of policy represents another explanatory variable potentially to be included (see Soss and Schram 2007). The fact that our model does not incorporate theseand otherfactors does not suggest that they are irrelevant. In fact, quite the contrary is the case. We expect these explanatory factors to matter, but how exactly they matter needs to be determined in a next step of theorizing. In this next step, the macro-level model will have to be transformed into one or multiple meso- and microlevel models, which then allow for formulating and testing hypotheses. Before formulating empirically testable hypotheses that capture the role of a set of relevant explanatory variables, we suggest assessing the consistency of the macro-level model first. Only if the basic logic of the model holds is it worth engaging in theoretical refinements and sophisticated testing strategies. For this purpose, we propose adopting a comparative research design that allows for addressing the following three features: the implementation structure, the characteristics of the political system, and the analytical capacity. First of all, we need to determine how often and when policymakers choose a centralized or a decentralized implementation structure. Put differently: do we have variation in the implementation structures selected within and across countries? Drawing on the notion of implementation styles as put forward by Howlett (2004, 2005), one could expect that implementation structures tend to vary across countries, but less so across policy domains within countries. Following the conceptual reasoning by Winter (2012) and Vancoppenolle et al. (2015), the implementation structure is decided individually for each policy, which means that we could expect variation across countries, within countries and even within policy domains. Consequently, before proceeding with the conceptual discussion, we need to develop a clearer understanding of the variation in implementation structures and ideally we would do so for different types of variations: across countries, within countries, but across policy domains, and within countries and policy domains. Depending on the outcome of this exercise, we could extend our theoretical reasoning and add potential explanatory variables for the variation in implementation structure. A second but related research step is to pay attention to the features of political systems. Most importantly, we need to differentiate between unitary and federal polities since this may also determine whether implementation structures are always or mostly centralized or decentralized. 14 The institutional characteristics are also important for learning about possible constraints policymakers may face at the stage of policy design. As already explained above, it is plausible to expect that multi-party governments adopt a different approach to policy design than singleparty governments. Furthermore, it is not only important to differentiate between multi- and single-party governments, but also to take into account the ideological preferences of the governing parties with regard to the choice of policy tools. From this perspective, it appears important to observe whether there is variation in the process and outcome of policy design, depending on the specific institutional characteristics of polities such as the degree of federalism and the partisan composition of governments. The third line of empirical research should look into the varying analytical capacities of governments. Analytical capacity is a straightforward concept, but it is difficult to operationalize (but see Wellstead et al. 2011). Of course, there is information on the staff and budget of agencies, but these indicators do not tap into the concrete tools these agencies use to analyze policy outcomes and what they consider as the benchmark against which the outcomes are evaluated. The operationalization becomes even more challenging when we seek to carry out comparative research, which almost automatically leads to the need to reduce information in order to facilitate comparability. Therefore, with regard to this third dimension, we need to develop valid measurements, which we could then use to produce comparative insights. After mastering this task, a next step is to examine when information about the effectiveness and acceptance of policies actually produces policy feedbacks and whether the reasoning put forward above holds true. To this end, we need to explore variation in the production of policy feedback loops. Once we have a better understanding of whether there is any variation, we should test whether this variation is due to the complexity of information. It is conceivable that policy feedbacks do not or not only stem from the availability and characteristics of policyrelated information produced during the implementation process, but are triggered by other factors. Despite the richness of the literature on policy feedback (see Mettler and SoRelle 2017), our understanding of how (positive and negative) policy feedbacks come about is limited. Therefore, concentrating on cases where we can observe policy feedbacks and contrasting them with negative cases can produce insights that are not only interesting for the conceptual model proposed here, but also for the comparative politics literature on policy feedbacks in general. As the prior explanations have indicated, there are only few established operationalizations for measuring the concepts of importance here. Therefore, it makes sense to start with comparative case studies that address only one of the three dimensions. The question of when policy 15 feedbacks emerge requires even more advanced analytical skills and should be addressed after having examined the other factors. At later stages, the empirical investigation could move on to quantitative analyses. Regardless of whether the research logic is qualitative or quantitative, however, it is appears essential to adopt comparative designs since the questions raised above revolve around variation. Once we have an improved understanding of the variability in the concepts, we can apply more sophisticated methods to establish the causal relationships between the different concepts. Consequently, the exploration and systematic analysis of empirical variation appears to be the next steps to take before being able to further our theoretical and conceptual understanding. 5. Conclusion Public policies can only become effective once they are put into practice. A policy design may be perfect, but it can still fall short of delivering the intended results when the implementation is flawed. This was the key message of the path-breaking book by Pressman and Wildavsky published in 1973. Eager to further scholarship on policy implementation, we decided to bring together three concepts that have as yet existed side by side without being discussed jointly: policy design, implementation style, and policy feedback. We proposed to connect policy design and policy implementation by means of the concept of policy feedback. The latter has mostly been discussed by studies in the field of comparative politics and there are only few policy studies that embrace this concept and adapt it to the phenomena typically studied by scholarship on public policy (see, e.g., Béland 2010). This is a surprising finding since policy feedback theory is widely considered to belong to the group of policy process theories (see Mettler and SoRelle 2017). With this contribution, we offer a perspective for furthering our understanding of policy implementation as well as to bring the concept of policy feedback back into policy studies. In this regard, we are confident that our reasoning about when information derived from policy implementation becomes a policy feedback also contributes to the comparative politics literature on policy feedback. The literature on policy feedback typically assumes that a feedback is given. We challenge this automatism and argue that the availability of information does not necessarily correspond to a policy feedback. This contribution would not have been possible without the public policy literature on the analytical capacity of policymakers (e.g. Wellstead et al. 2011), which demonstrates the value to bringing these concepts together. 16 Policy delivery can be organized in a hierarchical manner, where fixed policy intentions are expected to be handed down to addressees with as little deviation as possible (centralized implementation structure), or it can be organized in a more decentralized manner which leaves implementing actors more leeway for adapting the policy to local circumstances. This chapter has argued that while decentralized implementation offers more insights into what makes policies work than centralized implementation, the latter is more likely to produce policy feedback that can inform future policymaking and potential redesign of the policy concerned. This expectation builds on the complexity of information that can potentially result from a decentralized implementation process. With decentralized implementation, we are likely to observe a policy to be adapted to the conditions at the local level. Seen from this perspective, the individual adapted policies represent policy implementation experiments, of which some will work and other will fail. This variation in the effectiveness of the (adapted) policy constitutes a valuable source for improving the quality of policy design in the future. However, transforming this information into a policy feedback requires a high level of analytical capacity. Even if the necessary analytical capacity is available, inconsistent or conflicting information supplied from the various agencies involved could still hamper the emergence of policy feedback. Consequently, future policy design is more likely to benefit from insights originating from a centralized implementation structure for this arrangements has a greater potential to produce a policy feedback and therefore to guide policy-making. We believe that this is a novel way of looking at policy implementation and its outcomes for the policy process. Most studies of policy implementation attribute decentralized structures a higher capacity to benefit from policy learning than centralized structures. In decentralized arrangements, implementers have flexibility and autonomy to adjust policies in light of particular local conditions and changes in the perception of policy problems (Nilsen et al. 2013). This results in more information about how a given policy can be made operational. The argument we put forward is that despite the lower level of information, centralized policy implementation is more likely to stimulate policy feedback and learning than decentralized implementation since the information is easier to assess and is more likely to reach the policymakers responsible for policy redesign. At this point, we could only present our conceptual model, but owe the readership empirical findings to substantiate our findings. The reason for this is that an empirical test of our claims warrants a complex research design, which requires making decisions on which aspects we would like to keep constant and which ones we want to vary across the cases analyzed. As we elaborated in the previous section, we are convinced that an empirical test of our model can 17 only be attained by adopting a comparative research design. Given the early phase in which our reasoning is, both comparative case study analysis and quantitative analysis can help to refine our argument. The comparison can be attained by looking at different policy domains within one countries or at policies adopted in one policy domain across countries. What is important is to select a research design that allows for varying the implementation structure, the analytical capacity of government, and features of the political system. Altogether, we can conclude that the joint study of policy design, policy implementation and policy feedback is still in its infancy and requires innovative conceptual thinking and theorizing. In addition, comparative empirical investigations are needed to create a continuous process of formulating and testing hypotheses. Perhaps the absence of systematic comparative research is currently the biggest challenge for the advancement of this field. Therefore, we welcome all endeavors to produce and analyze comparative data on policy design, policy implementation, and policy feedback. In so doing, public policy scholars are well advised to take into account the relevant comparative politics research on policy feedback. 6. References Bakker, R., De Vries, C., Edwards, E., Hooghe, L., Jolly, S., Marks, G., ... and Vachudova, M. A. (2015): Measuring party positions in Europe: The Chapel Hill expert survey trend file, 1999–2010. Party Politics, 21(1), 143-152. Béland, D. (2010): Reconsidering policy feedback: How policies affect politics. Administration & Society, 42(5), 568-590. Bolleyer, N., and Börzel, T. A. (2010): Non-hierarchical policy coordination in multilevel systems. European Political Science Review, 2(2), 157-185. Capano, G., Regini, M., and Turri, M. (2016): Changing Governance in Universities. Palgrave Macmillan UK. Capano, G., and Woo, J. J. (2017): Resilience and robustness in policy design: A critical appraisal. Policy Sciences, 1-28. Chindarkar, N., Howlett, M., and Ramesh, M. (2017): Introduction to the Special Issue: “Conceptualizing Effective Social Policy Design: Design Spaces and Capacity Challenges”. Public Administration and Development, 37(1), 3-14. 18 Colebatch, H. K. (2017): The idea of policy design: Intention, process, outcome, meaning and validity. Public Policy and Administration, 0952076717709525. Damonte, A., Dunlop, C. A., and Radaelli, C. M. (2014): Controlling bureaucracies with fire alarms: policy instruments and cross-country patterns. Journal of European Public Policy, 21(9), 1330-1349. DeLeon, P. (1988): The contextual burdens of policy design. Policy Studies Journal, 17(2), 297-309. Ellis, K. (2007): Direct payments and social work practice: The significance of ‘street-level bureaucracy’in determining eligibility. British Journal of Social Work, 37(3), 405-422. Elmore, R. F. (1979): Backward Mapping: Implementation Research and Policy Decisions. Political Science Quarterly, 94(4), 601-616. Falkner, G., Hartlapp, M., and Treib, O. (2007): Worlds of compliance: why leading approaches to european union implementation are only ‘sometimes‐true theories’. European Journal of Political Research, 46(3), 395-416. Falkner, G., and Treib, O. (2008): Three worlds of compliance or four? The EU‐15 compared to new member states. Journal of Common Market Studies, 46(2), 293-313. Goggin, M. L., Bowman, A. O.'M., Lester, J. P., and O'Toole, L. J. (1990): Implementation Theory and Practice: Toward a New Generation. Glenview: Scott. Grüne-Yanoff, T. and Hertwig, R. (2016): Nudge Versus Boost: How Coherent are Policy and Theory? Minds and Machines 26(1), 149–183. Hjern, B., and Porter, D. O. (1981): Implementation Structures: A New Unit of Administrative Analysis. Organization Studies, 2(3), 211–227. Hood, C. (1986): The tools of government, Chatham: Chatham House Publishers Hood, C., and Margetts, HZ. (2007): The tools of government in the digital age, Basingstoke: Palgrave Macmillan. Howlett, M. (2004): Beyond good and evil in policy implementation: Instrument mixes, implementation styles, and second generation theories of policy instrument choice. Policy and Society, 23(2), 1-17. 19 Howlett, M. (2005): What is a policy instrument? Tools, mixes, and implementation styles. Designing government: From instruments to governance, 31-50. Howlett, M. (2009): Governance modes, policy regimes and operational plans: A multi-level nested model of policy instrument choice and policy design. Policy Sciences, 42(1), 73-89. Howlett, M., and Ramesh, M. (1993): Patterns of Policy Instrument Choice: Policy Styles, Policy Learning and the Privatization Experience. Policy Studies Review, 12(1/2), 3–24. Howlett, M., and Mukherjee, I. (2017): Policy design: From tools to patches. Canadian Public Administration, 60(1), 140-145. Howlett, M., and Rayner, J. (2007): Design principles for policy mixes: cohesion and coherence in ‘new governance arrangements’. Policy and Society, 26(4), 1-18. Howlett, M. (2011): Revisiting policy design: The rise and fall (and rebirth?) of policy design studies. In General Conference of the European Consortium for Political Research (ECPR), Section (Vol. 63). Howlett, M., Mukherjee, I., and Woo, J. J. (2015): From tools to toolkits in policy design studies: the new design orientation towards policy formulation research. Policy & Politics, 43(2), 291-311. Hupe, P. L., and Hill, M. J. (2016): ‘And the rest is implementation.’ Comparing approaches to what happens in policy processes beyond Great Expectations. Public Policy and Administration, 31(2), 103-121. Hupe, P., and Sætren, H. (2015): Comparative Implementation Research: Directions and Dualities. Journal of Comparative Policy Analysis: Research and Practice, 17(2), 93-102. John, P., Smith, G., and Stoker, G. (2009): Nudge nudge, think think: two strategies for changing civic behaviour. The Political Quarterly, 80(3), 361-370. Jordan, A., and Matt, E. (2014): Designing policies that intentionally stick: policy feedback in a changing climate. Policy Sciences, 47(3), 227-247. Kahneman, D., Slovic, P., and Tversky, A. (Eds.). (1982): Judgment under uncertainty: Heuristics and biases. New York, NY: Cambridge University Press. Levi-Faur, D. (2010): Regulatory Architectures for a Global Democracy: On Democratic Varieties of Regulatory Capitalism. in: T. Porter and K. Ronit, (Eds) The Challenges of 20 Global Business Authority: Democratic Renewal, Stalemate, or Decay? Suny University Press, 2010, pp. 205-226. Lipsky, M. (1980): Street level bureaucrats. Nova York: Russel Sage. Matland, R. E. (1995): Synthesizing the implementation literature: The ambiguity-conflict model of policy implementation. Journal of public administration research and theory, 5(2), 145-174. Mettler, S., and SoRelle, M. (2017): Policy feedback theory. in: C. Weible and P. Sabatier (Eds.), Theories of the policy process, Boulder, CO: Westview Press, 151-181. Mortensen, P. B., and Green-Pedersen, C. (2015): Institutional Effects of Changes in Political Attention: Explaining Organizational Changes in the Top Bureaucracy. Journal of Public Administration Research & Theory, 25(1), 165–189 Nilsen, P., Ståhl, C., Roback, K., and Cairney, P. (2013): Never the twain shall meet?-a comparison of implementation science and policy implementation research. Implementation Science, 8(63). Ostrom, E. (2010): Beyond markets and states: polycentric governance of complex economic systems. Transnational Corporations Review, 2(2), 1-12. Peters, B. G. (2015): Pursuing horizontal management: The politics of public sector coordination. University Press of Kansas. Pierce, J. J., Siddiki, S., Jones, M. D., Schumacher, K., Pattison, A., and Peterson, H. (2014): Social construction and policy design: A review of past applications. Policy Studies Journal, 42(1), 1-29. Pierson, P. (1993): When effect becomes cause: Policy feedback and political change. World politics, 45(4), 595-628. Pressman, J. L., and Wildavsky, A. B. (1973): Implementation: how great expectations in Washington are dashed in Oakland: or, why it's amazing that federal programs work at all, this being a saga of the Economic Development Administration as told by two sympathetic observers who seek to build morals on a foundation of ruined hopes. Univ of California Press. 21 Pülzl, H., and Treib, O. (2007): Implementing public policy. Handbook of public policy analysis: Theory, politics, and methods, 125, 89-107. Rowe, M. (2007): Rendering visible the invisible: police discretion, professionalism and decision-making. Policing & Society, 17(3), 279-294. Sabatier, P. A., and Mazmanian, D. A. (1979): The Conditions of Effective Implementation: A Guide to Accomplishing Policy Objectives. Policy Analysis, 5(4), 481–504. Sager F., and Rielle Y. (2013): Sorting through the garbage can: under what conditions do governments adopt policy programs? Policy Sciences 46(1): 1–21. Sager, F., Thomann, E., Zollinger, C., van der Heiden, N., and Mavrot, C. (2014): Street-level Bureaucrats and New Modes of Governance: How conflicting roles affect the implementation of the Swiss Ordinance on Veterinary Medicinal Products. Public Management Review, 16(4), 481-502. Saetren, H. (2014): Implementing the third generation research paradigm in policy implementation research: An empirical assessment. Public Policy and Administration, 29(2), 84-105. Schneider, A. L., and Ingram, H. M. (1997): Policy design for democracy. University Press of Kansas. Soss, J., and Schram, S. F. (2007): A public transformed? Welfare reform as policy feedback. American Political Science Review, 101(1), 111-127. Strassheim, Holger and Silke Beck (Eds.) (2019): The Research Handbook of Behavioural Change and Public Policy. Cheltenham: Edward Elgar. Thaler, R. H., and Sunstein, C. R. (2008): Nudge: Improving Decisions About Health, Wealth, and Happiness. New Haven, CT: Yale University Press. Thomann, E., Lieberherr, E., and Ingold, K. (2016): Torn between state and market: Private policy implementation and conflicting institutional logics. Policy and society, 35(1), 5769. Tosun, J. (2013): Environmental policy change in emerging market democracies: Eastern Europe and Latin America compared (Vol. 41). University of Toronto Press. 22 Tosun, J. (2017): The behaviour of suppliers and consumers in mandated markets: the introduction of the ethanol–petrol blend E10 in Germany. Journal of Environmental Policy & Planning, 1-15. Tosun, Jale und Samuel Workman (2017): Struggle and Triumph in Fusing Policy Process and Comparative Research. In C. Weible und P. Sabatier (Hrsg.), Theories of the Policy Process. Boulder, CO: Westview Press, 329–362 Tosun, J., Unt, M., and Wadensjö, E. (2017): Youth‐oriented Active Labour Market Policies: Explaining Policy Effort in the Nordic and the Baltic States. Social Policy & Administration, 51(4), 598-616. Tsebelis, G. (2002): Veto players: How political institutions work. Princeton University Press. Vancoppenolle, D., Sætren, H., and Hupe, P. (2015): The Politics of Policy Design and Implementation: A Comparative Study of Two Belgian Service Voucher Programs. Journal of Comparative Policy Analysis: Research and Practice, 17(2), 157-173. Van Meter, D. S., and Van Horn, C. E. (1975): The policy implementation process: A conceptual framework. Administration & Society, 6(4), 445-488. Van der Knaap, P. (1995): Policy evaluation and learning: feedback, enlightenment or argumentation?. Evaluation, 1(2), 189-216. Wälti, S. (2004): How multilevel structures affect environmental policy in industrialized countries. European journal of political research, 43(4), 599-634. Weaver, R. K. (2015): Getting People to Behave: Research Lessons for Policy Makers. Public Administration Review, 75(6), 806-816. Wellstead, A. M., Stedman, R. C., and Howlett, M. (2011): Policy analytical capacity in changing governance contexts: A structural equation model (SEM) study of contemporary Canadian policy work. Public Policy and Administration, 26(3), 353-373. Winter, S., (1990): Integrating Implementation Research, in: D. J. Palumbo and D. J. Calista (Eds), Implementation and the Policy Process: Opening up the Black Box (New York: Greenwood Press), pp. 19-38. Winter, S., (2012): Implementation. in: B. G. Peters and J. Pierre (Eds) Handbook of Public Administration, 2nd ed. (London: Sage), pp. 255–263. 23