Can power be made an empirically viable concept in policy process theory? Exploring the power potential of the Narrative Policy Framework

Despite the range of analytical foci in current policy process theory, the idea of an empirically sound power concept has not received much attention. While scientifically oriented process frameworks tend to be either implicitly or explicitly based on a pluralist understanding of power, critical theory focused approaches frequently point to power inequality in the policy process but remain vague on its conceptualization. As a result, the concept of power remains underspecified, which renders theoretical understanding of policy-making incomplete. In this article, we argue that it is necessary to integrate an empirically viable power concept into policy process theory which allows researchers to systematically assess the role of structural power imbalances in policymaking, without compromising scientific rigor. To that end, we examine how power has been treated in policy process theory, with focus on the Advocacy Policy Framework (ACF), Social Construction and Policy Design, and—primarily—the Narrative Policy Framework (NPF). In a second step, we explore how Steven Lukes’ three-dimensional power concept can be leveraged by the NPF to bridge the gap between different understandings of power, while also articulating a concept of power amenable to scientific testing within policy studies.


Introduction
Despite the wide range of analytical foci in policy process theory, none of the approaches in the current collection of canonical frameworks address the concept of power in analytical detail (Herweg et al., 2017). In fact, while power as a theoretical concept received a fair share of attention in policy theoretical research in the late 1980s and early 1990s (Kellow, 1989(Kellow, , 1988Lowi, 1979;Schneider and Ingram, 1993;Spitzer, 1987), it now seems to have more or less dropped off the radar (but see, for example, Cook, 2010). While the most popular policy process-oriented theories (i.e., Weible and Sabatier, 2018) rely on implicit pluralist power assumptions, other approaches such as Social Construction and Policy Design (Ingram et al., 2014;Schneider and Sidney, 2009;Schneider and Ingram, 2005) and, to a lesser extent, the Narrative Policy Framework (NPF) (Jones and McBeth, 2010;Shanahan et al., 2017) do explicitly address the concept of power in the policy process. However, the latter either do not conceptualize power as a central aspect (NPF) or do not operationalize the concept sufficiently, as we will argue is the case with Social Construction and Policy Design. As a result, the concept of power remains underspecified and leaves a theoretical gap in the respective frameworks' understanding of the policy process.
In this article, we posit that we must begin integrating an empirically viable power concept into policy process theory in order to better assess how, when and where structural power imbalances in a society, e.g. based on class, race, ethnicity or gender, impact policy, and do so without compromising scientific rigor. We recognize the concept is 'essentially contested' (Gallie, 1955) and difficult to operationalize; however, we argue that it is not justified to sideline the concept of power in policy theory because of said difficulty, nor because of a rift between scientifically oriented policy process theory and critical policy process theory, 1 in which empirical testing seems to be more a domain of the first and assessing the impact of inequality more a task of the latter. We argue that in order to make multiple dimensions of the concept of power empirically testable, it is necessary to analyze the interaction of beliefs and power in policymaking, i.e. to examine how the beliefs of those in power are expressed and how they affect the outcomes for those not in power. The potential structural dominance of certain beliefs among individual policy actors and coalitions, and its impact on relevant populations, should be part of the analysis, which goes beyond merely identifying different beliefs.
Furthermore, we also argue that using power as an explicit theoretical concept, rather than an implicit assumption, would help to clarify the blurred distinction between empirical and normative understandings of power, which tend to align with scientifically and critical theoryoriented approaches, respectively. As succinctly put by Jorgensen, the classic form of pluralism (Kelso, 1978), which is easily identified in the ACF and other policy process frameworks, constitutes "more a normative theory of democracy than …an empirical investigation…" (Jorgensen, 2017: 451). If such normative undercurrents are not transparent, we run the risk of reproducing existing biases in policy research (Bachrach and Baratz, 1962).
In order to find a way forward, we propose using Steven Lukes' classic three-dimensional power conception as a starting point (Lukes, 2004(Lukes, , 1974 for integrating an empirical analysis of power into policy process theory, with focus on the NPF. Despite criticism on the notion of 1 -For the purpose of this article, we compare and contrast what we here term scientifically oriented policy theory and critical policy theory. However, the terminology used to describe these approaches varies, e.g. these two schools of thought have also been described as process vs. content-oriented approaches and as normative vs empirically focused policy theory. While emphases differ, these juxtapositions all serve to highlight similar key differences in the compared theoretical approaches. false consciousness and agency (Bradshaw, 1976;Clegg, 1989;Hayward and Lukes, 2008;Isaac, 1987), Lukes' breakdown of power can provide a useful guide for empirically analyzing different facets of power by helping steer researchers from more obvious and visible expressions of power (first dimension) to less visible expressions, such as issues that do not get put on the agenda (second dimension), and how preferences themselves may be shaped by beliefs, institutions, and other societal structures (third dimension). Furthermore, we illustrate that a Lukes'-based analysis of power could be operationalized via the vehicle of the NPF's policy narratives, which connect power to policy makers' belief systems, and allow for the assessment of the impacts those belief systems have on public policy-both scientifically and normatively.
In the following, we first examine how power has been treated in policy process theory, specifically in the ACF, Social Construction and Policy Design, and the NPF. We focus on these specific approaches because they represent broader schools of thought within policy process theory. The ACF serves as a notable example for scientifically oriented approaches with an underlying pluralistic power conception, while Social Construction and Policy Design takes a critical and explicitly normative view on structural power imbalances. Finally, the NPF unites both elements of scientifically oriented and critical policy process theory, which we argue is valuable when empirically analyzing power dynamics within policymaking. In a second step, we discuss the theoretical implications of this treatment, with a focus on the implications of the pluralist roots of policy process theory. We then elaborate on Steven Lukes' three-dimensional power conception and show how aspects of it have been implicitly used in Social Construction and Policy Design research, as well as the NPF, albeit without sufficiently conceptualizing and operationalizing it. Following from this, we discuss the potential of Lukes' power theory to serve as an empirically viable concept and outline what the role of power theory could be going forward, focusing on the NPF. We argue that the NPF has the potential to bridge the gap between scientifically oriented policy theory and critical policy theory, both of which have substantial shortcomings in their current treatments of power in public policy. For this purpose, we leverage existing policy process research on in-state tuition eligibility of undocumented immigrants in the United States through the lens of the NPF and suggest several power propositions to illustrate an approach to operationalization that can capture all three dimensions of power while maintaining standards of scientific rigor.

Power in Policy Process Theory
In the introduction to the fourth edition of Theories of the Policy Process (TOTPP4), Christopher Weible names power as a concept that, while not part of the volume, is "deserving attention" (Weible, 2017:7). This remark indicates that even though power is generally considered a valuable idea, it has also been sidelined as a supplementary and "ambiguous" (Heikkila and Cairney, 2017: 304) concept, which does not fulfill the criteria of a scientific theory as outlined by Weible's mirroring of the late Paul Sabatier in the introduction of TOTPP4. 2 Interestingly, going beyond the more explicit discussion in the introduction, the frameworks do reference the concept. In fact, all of the theories featured in TOTPP4 use the word power at least once. For instance, the Multiple Streams Framework (MSF) (Herweg et al., 2017: 33), Policy Feedback Theory (PFT) (Mettler and Sorelle, 2017: 104) and the NPF (Shanahan et al., 2017: 189) mention "powerful interest groups". Punctuated Equilibrium Theory (PET) refers to the "power of specialized communities of experts" (Baumgartner et al., 2017: 58) and the ACF describes how actors strategically "exaggerate the power and maliciousness of their opponents" (Jenkins-Smith et al., 2017: 140). However, despite these references the authors do not specify what their understanding of power entails. As a result, the theoretical treatment of power remains largely superficial within TOTPP4's policy process approaches.
In the following we briefly examine the ACF, Social Construction and Policy Design, and the NPF, with attention to the role power plays within each. We focus on these three approaches for several reasons. 3 The ACF is one of the most developed and commonly applied policy process theories, with meta reviews pointing to at least 241 applications since its inception (Pierce et al., 2017;Weible et al., 2009). While the TOTPP4 chapter on the ACF includes only very few references to power, the ACF has previously been criticized for its underlying pluralist assumptions (James and Jorgensen, 2009;Jorgensen, 2017;Parsons, 1995), which cause it to sidestep potential power inequalities and makes it an interesting framework to examine when arguing for a more explicit power concept. Chosen as our representative for a critical theory oriented approach within the policy process literature, Social Construction and Policy Design (Ingram et al., 2014(Ingram et al., , 2007-an expressly normative approach-argues that some groups are disadvantaged in public policy and emphasizes that policy outcomes are best understood when both ideology (social constructions) and political power are considered. Even though the approach was not included in TOTPP4, it was part of the second and third editions and is arguably the policy process framework that pays the most attention to the concept of power. 4 Lastly, if thinking about power as a continuum upon which the various approaches in policy process theory are situated, the NPF can be located between the ACF and Social Construction and Policy Design. ACF-sympathetic in its theoretical structure and empirical applications to-date (e.g. Shanahan et al., 2011), the NPF is at the same time founded upon the notion that social constructions are critical to understanding public policy (Jones and Radaelli, 2015) and acknowledges the use of power to omit and manipulate policy narratives (Peterson and Jones, 2016). While the NPF currently underspecifies the role of power, its theoretical 'middle ground' gives it the unique potential to accommodate both the roles of social constructionswhich are driven by and expressed through narratives-in creating lasting power imbalances in policymaking, as well as the concerns raised in more scientifically oriented frameworks, such as the ACF.

The Advocacy Coalition Framework, Social Construction and Policy Design, and the Narrative Policy Framework
Like all research programs (Lakatos, 1970), the ACF is founded upon a set of explicit and implicit assumptions. One of the explicit assumptions is that policymaking is best understood by examining policy subsystems, which contain actors who are non-trivially cooperating within coalitions to advance a policy position, as the central units of analyses. These non-trivially cooperating actors are, by definition, policy elites. Policy elites can consist of representatives of government departments and select interest groups, as well as involved individuals from the private sector, academics and non-profit organizations (Jenkins-Smith et al., 2017;Sabatier and Jenkins-Smith, 1993). Moreover, while the ACF explicitly assumes policy actors are 3 -We are aware that the analysis of power in policymaking goes beyond these approaches, which have emerged in the U.S. context as 'policy process theory'. For example, Power Resource Theory, which is rooted in a European welfare state context, attributes variation in countries' social policy regimes to the distribution of power (resources) between classes (Esping- Andersen, 1990;Korpi, 1983). 4 -Here we would like to emphasize the term "arguably," as a strong case can also be made for other process frameworks (for example, see Mettler and Sorelle, (2017)). boundedly rational individuals (i.e., Simon, 1947), it also puts emphasis on the importance of understanding their belief systems as motivating forces in decision making (Pierce et al., 2017;Weible et al., 2009). Thus, at its most basic level, the ACF aims at "describing, explaining and, sometimes, predicting" (Jenkins-Smith et al., 2017: 138) how belief systems influence the decisions of policy elites to form or join advocacy coalitions and how this affects policy change and learning.
To operationalize belief systems within the ACF, beliefs are assumed to be hierarchically structured, ranging from more fundamental and general to more instrumental and specific. More precisely, the ACF distinguishes between deep core beliefs, policy core beliefs, and secondary beliefs. While deep core beliefs are strongly ingrained in the individual and highly resistant to change, this is less the case for policy core beliefs, which can perhaps be best thought of as 'applied' deep core beliefs. Finally, secondary beliefs, which describe views held on more specific details of policy, are considered the most susceptible to change. The ACF assumes actors form advocacy coalitions within policy subsystems based on the above outlined belief systems. Coalitions compete in policy subsystems, often disagreeing on policy proposals and strategies as a result of their different belief systems. In short, "policy debates…revolve around diverging preferences" (Jenkins-Smith et al., 2017: 145). It is important to note that not all coalitions have the same levels of influence, that coalitions do not necessarily remain stable, and that defection is common (ibid).
As previously noted, the ACF does not acknowledge power as a causal driver in the policy process. However, despite the lack of an explicit power conceptualization it is possible to identify a pluralist tendency within the framework, which implicitly assumes that policy in an ACF subsystem is "the outcome of a free competition between ideas and interests" (Parsons, 1995: 134). Scholars who argue from a pluralist perspective posit that power can be best measured by analyzing which actors succeed in furthering their own interests in a given situation, by defeating the preferences of their political opponents. Such preference competitions are considered observable expressions of power and thus amendable to measurement. However, it should be noted that power in a pluralist system is often assumed to be openly (not necessarily evenly) distributed between different groups and thus there is at least the potential for all relevant interests or beliefs in a society to participate (Dahl, 1961;Kelso, 1978;Truman, 1981). Whether implicit or explicit, this assumption about power allows for the justification to focus on the observable preference/interest competitions.
An understanding of the distribution of power can be teased out of the ACF. Policy subsystems allow different coalitions of elites to compete and strive for influence, where the accessibility of said subsystems is rarely a focus of ACF scholarship. In this context, the ACF attributes policy stability to dominant belief systems (Pierce et al., 2017). Despite the fact that power is not explicitly discussed, the ACF implicitly connects the concepts of beliefs and power by emphasizing the importance of shared beliefs by majority coalitions. However, there are opportunities for minority coalitions to increase their influence, for example in moments of internal or external crises (Jenkins-Smith et al., 2017). This distinction is important because while a coalition may come out 'on top', causing its belief system to be dominant in shaping a policy, this domination is still considered the result of an observable belief-system competition and not the structural entrenchment of certain values among elite actors, which, for example, might allow for control of the agenda, where non-decisions are at least as critical as decisions, if not more so (Bachrach and Baratz, 1962). Importantly, that a minority coalition might come out on top still only speaks to non-trivially cooperating coalition participants in the policy subsystem, and not those that might fall into the ACF's trivial category, who remain theoretically-and thus empirically-invisible to the ACF.
In contrast to the ACF, which concentrates on policy change and learning driven by coalition belief systems, Social Construction and Policy Design (Ingram et al., 2014(Ingram et al., , 2007 focuses on policy designs as a reflection of policy makers' social constructions, derivative of beliefs and underlying social norms and their ability to have those derivatives reflected in public policy. Furthermore, the idea of political power is used explicitly and constitutes one of the two theoretical dimensions used to understand populations targeted by policy. These two dimensions are political power and perceived deservedness. When the dimensions are intersected, a two-bytwo of four quadrants of target populations is illuminated: Advantaged (high power/high deservedness), contenders (high power/low deservedness), dependents (low power/high deservedness) and deviants (low power/low deservedness).
According to Social Construction and Policy Design, the location of a target group within the two-by-two matrix can have a lasting impact on the way it benefits from or is burdened by policy. The approach further proposes that both the perceived deservedness and political power of target groups will not only impact policy designs, but also influence future policy stasis and change (Ingram et al., 2014: 129). In short, Social Construction and Policy Design is a transparently normative approach which focuses on how political power manifests itself in policy designs via dominant value paradigms (Schneider and Ingram, 1993). The two-dimensional typology explains why belief systems have an impact on policy, but it also examines why belief systems alone may not be enough to explain an actor's or a coalition's actions, if power is not considered. This, for instance, becomes relevant when looking at dependents, who are generally considered deserving of policy support, but do not possess much political power. Consequently, deserving target populations may be less likely to receive benefits, despite the driving force of sympathetic belief systems. Conversely, Social Construction and Policy Design also rejects the notion that power alone is sufficient to explain the distribution of benefits, as the power "capacity" might be there, but the belief system's willingness to deploy that capacity may not. When political power-in a pluralist sense-is considered in isolation, researchers are likely to focus only on the easily observable expressions of power and neglect the role of structural elements and societal norms in policymaking and how they can create disadvantages and advantages for various groups (Ingram et al., 2014;Schneider and Ingram, 1993). By integrating these two dimensions, the approach touches upon what Lukes terms the second and third dimension of power, which are elaborated further below.
Finally, the main goal of the NPF is to describe the role of policy narratives at different points in the policy process and determine how policy narratives shape public policy (Shanahan et al., 2017). In particular, the NPF aims to analyze how policy narratives shape policy at the macro (e.g., institutional and cultural), meso (e.g. subsystem), and micro (individual) levels of policymaking and how the different levels potentially interact. To accomplish this, the NPF treats policy narratives as objects (i.e., "things" in the world) that have distinct components (form and content), within which are narrative structural elements of a setting, characters, plots, and a moral, as well as content elements such as belief systems and policy narrative strategies (Jones, 2018;Jones et al., 2014). This structural approach to policy narratives is necessary to make the NPF's various policy narrative components and elements amenable to quantification. 5 As such, the NPF has been shown to be quite compatible with policy process theories such as the ACF (Shanahan et al., 2011) or PET (Peterson, 2019), which have similar "positivist" tendencies. Derivative of Jones (2018), Table 1 lists the NPF elements and their definitions.

Setting
Consists of policy consequential phenomena such as legal codes, geography, demographics, evidence, etc., some of which are agreed upon and some of which are contested.

Characters
Consists of victims who are harmed or potentially harmed, villains who are the source of the harm, and heroes who offer solutions to the harm.

Plot
Organizes the characters relative to each other, the setting, and situates the narrative in time and space.
Moral of the Story Typically a call to action or a policy solution.

Beliefs
The content of narrative varies substantially, but people make sense of it in systematic ways despite that variation. In the NPF this is captured by robust belief system theories and measures which identify how groups of people interacting with narrative content interpret said content. Common belief system approaches in the NPF include ideology and cultural theory, among other possibilities.

Strategies
Consist of the goal seeking patterns of narrative organization, which typically focus on the use of narrative form elements as well as belief system content.
Source: Jones 2018 However, even though the NPF shares a familial resemblance to the ACF in its methodological orientation, the framework also leverages social constructivism as one of its foundational assumptions, which has facilitated the NPF's use of interpretive approaches to the study of public policy (e.g., Dupuis, 2018;Jones and Radaelli, 2015;Gray and Jones, 2016). The scientifically oriented approach is adept at assessing the observable, while critical theory and interpretive approaches are adept at capturing the unobserved. There have also been explicit integrations of the NPF and Social Construction and Policy Design, using both quantitative and qualitative methods, demonstrating the capabilities of merging the two approaches (e.g., Husmann, 2015;Lybecker et al., 2018). In short, the NPF has already demonstrated an epistemological and methodological flexibility that regularly pairs scientific and critical policy approaches at multiple levels of analysis, using a variety of methodologies. Consequently, we argue that the NPF presents a rather unique opportunity among existing policy process approaches to generate a rigorous conceptualization of power that captures multiple dimensions of the concept. Our argument builds off of Peterson and Jones (Peterson and Jones, 2016), who have taken the NPF steps in this direction by examining how elites use the power of narrative to shape opinions and preferences to control and manipulate agendas, which we discuss more below.

Theoretical Implications
The above exploration has shown that the examined policy process approaches all reference some form of power: power as a shared resource, divided among interest groups in an openly competitive policy arena (ACF), power as an unequally distributed resource, which in connection with beliefs and social norms has a lasting impact on policy design, and power as a policymaker's tool for manipulation of opinions and policy preferences (NPF). However, none of the approaches currently use a clear concept of power. This section aims to show that the absence of a clear theoretical power concept, paired with dominant pluralist roots in positivist policy process approaches, runs the risk of falsely portraying the policymaking process as an open field-type competition, in which every affected group is potentially and perhaps fairly represented. Furthermore, we argue that the ever-present divide between scientifically oriented and critical policy theory within the field of public policy adds to the difficulty of discussing power inequality as an important aspect in the empirical analyses of the policy process.
For some time now (e.g., Dahl, 1961;Truman,1981), pluralism has been one of the most dominant theoretical approaches for explaining the U.S. political system. Its critics have been manifold (e.g., Kelso, 1978;McConnell, 1970;Olson, 1965), and the gist of those criticisms have stayed remarkably consistent. From the initial criticisms (e.g., Lowi, 1979Lowi, , 1972Schattschneider, 1960), to the more recent (e.g. Jorgensen, 2017;Schneider and Ingram, 1997), critics observe that competition is not open, participation is contingent, and the results of policy systemically favor those with resources. One of the more salient criticisms to the argument we present here is voiced by Schneider and Ingram (1997) and Jorgensen (2017), who argue that utilizing pluralism as a theoretical explanation amounts to conflating a normative a priori assumption with empirical observations-that is, because a researcher starts with a pluralist assumption, what they see is a pluralist reality. Simply put, these critics argue that pluralism constitutes an "idealized way of policymaking that no longer exists or has never been realized " (Schneider and Ingram, 1997: 18). If this claim is correct, then it renders theoretical frameworks with pluralist underpinnings normatively pluralist rather than objectively so. We think this to be the case. Extant policy process theory is simply ill-equipped to see otherwise.
Even though the above identified shortcomings of a pluralist view on politics in the U.S. context may seem dated or well-known, they are relevant when analyzing power conceptions within current policy process approaches. For instance, despite the fact that the ACF takes precise steps to outline its underlying assumptions, it is silent on pluralism as one of the guiding principles, even though parallels between classic pluralist theory and the ACF's perspective on policymaking are easily identified (e.g., Jorgensen, 2017). Not making transparent that one of the underlying premises of the framework is disputed is problematic when one's framework aims to describe and explain the policy process in both an empirical and generalizable fashion.
What is left out due to the ACF's-or any process framework's-pluralist assumptions? For starters, only non-trivially cooperating actors are counted in the ACF's coalitions. By definition, this leaves 'trivial' actors invisible. Trivial actors include the general public, which is treated as a resource by the ACF (Jenkins-Smith et al., 2017; although see Jones and Jenkins-Smith, 2009), as well as weakened groups such as single mothers or sex workers, who might find themselves the target groups of subsystem policies. As a result, the ACF researcher might never ask questions such as: how do dominant belief systems within a subsystem affect populations -both internal and external to the subsystem -who are not meaningful parts of coalitions? When these beliefs dominate, do they drive policy outcomes over long periods of time (see Crenson, 1971: 173)? Which decisions get made because they fit a belief system and, conversely, which decisions are never made because they do not (Bachrach and Baratz, 1962)? In such a scenario, policy learning may only occur within a restricted set of options, guided by a narrow spectrum of values, and that may in turn limit or even precisely direct the possibilities of policy change or stasis. 6 At the moment the ACF does not offer the theoretical tools to examine these more complex expressions of power and their interaction with belief systems. As a consequence, we believe more established power theories may offer a way forward.
According to Steven Lukes (2004, 1974 a pluralist understanding of power is one-dimensional 7 because it only focuses on observable behavior. This one-dimensional view of power was most notably contested by Peter Bachrach and Morton Baratz, who argued that conceptions of power also need to account for who has control over the agenda. According to Bachrach and Baratz it is possible to identify certain "rules of the game" (Bachrach and Baratz, 1962: 950), which are based on the beliefs of those in power, and which lead the powerful to exclude non-conforming issues and parties from the political agenda. As a consequence, dominant beliefs in policymaking are re-enforced and lead to the "mobilization of bias" (ibid). While Lukes agrees with Bachrach and Baratz on the importance of including agenda setting as an expression of power (he refers to this as the second dimension of power), he posits that the exercise of power goes beyond decision-making and non-decision making. He argues that a comprehensive conception of power would capture the processes through which preferences in a society are shaped, or manipulated, by means of domination, which he calls the third dimension of power. In this extended view, Lukes defines domination as "the ability to constrain the choices of others, coercing them or securing their compliance, by impeding them from living as their own nature and judgment dictate" (Lukes, 2004: 85). In this context, it is important to note that Lukes acknowledges that his work focuses on 'power over' rather than 'power to', or what Wartenberg terms transformative power, meaning the use of power to enable and empower others (Wartenberg, 1990). Consequently, Lukes does not claim to comprehensively define the concept of 'power' but rather "addresses the question: 'How do the powerful secure the compliance of those they dominate?'" (Lukes, 2004: 110).
While decision-making and agenda setting power are rather concrete phenomena, the third dimension of power is comparatively abstract because it manifests itself through dominant beliefs and social norms, which are challenging to observe at all, let alone measure. Moreover, the notion of false consciousness, which describes individuals acting against their 'real interests' because their preferences have been manipulated, has further been rightly criticized as disregarding the agency of those who are dominated (e.g., Benton, 1981;Bradshaw, 1976;Clegg, 1989;Hayward and Lukes, 2008;Isaac, 1987). However, we would like to note that we follow John Gaventa's line of argument regarding the term false consciousness. Gaventa, a student of Lukes, rejects the notion of 'false' consciousness on the grounds that an individual's consciousness constitutes reality for the person in question. However, he emphasizes that it is important to analyze the "content, source or nature of the consciousness" (Gaventa, 1980: 29) as it may become apparent that an individual is only aware of a restricted spectrum of possible interests. The question asked by Gaventa is why this might be the case, and whether a limited scope of perceived possibilities may be the result of power being exercised as domination. Following this argument, moving away from a discussion of false consciousness, the shaping of preferences as an expression of power must be seen as important when attempting to understand the interaction of beliefs and power (Lukes, 1974), less one be left with only observable behavior and nonagendas. For instance, as previously touched upon, the NPF refers to the use of power to shape narratives with the aim of manipulating individuals' opinions on issues such as climate change, which could ostensibly be considered contrary to their 'real' scientific interests (Shanahan et al., 2017). To get a full view of power, Gaventa stresses that the three dimensions of power interact and that they must be understood by examining all three in tandem (Gaventa, 1980: 256). For instance, when issues are kept off the policy agenda in a second-dimension expression of power, this may occur because policy preferences have been shaped by specific beliefs or norms, which constitutes a third-dimension expression of power.
Even though no policy process approach has fully embraced Lukes' three-dimensional power concept, Social Construction and Policy Design comes closest to a multidimensional power understanding. While it does not offer a detailed conceptualization of what constitutes power and how it can be empirically tested, Anne Schneider and Helen Ingram's earlier work does refer to some of the theoretical foundations of power research in the social sciences. Schneider's and Ingram's 1993 article on Social Construction and Policy Design portrays the approach as an extension of Theodore Lowi's concept of Arenas of Power, which criticizes limiting the focus of policy research to a pluralist form of democracy and argues that elitism plays a significant role in U.S. policymaking (Lowi, 2009(Lowi, , 1979. Furthermore, Schneider and Ingram reference Gaventa, who uses Lukes' third dimension of power to understand the (absence of) resistance among citizens in Central Appalachia (Gaventa, 1980). Gaventa and his application of the second (Bachrach and Baratz, 1962) and third dimensions of power are further referred to in Schneider and Ingram's book Policy Design for Democracy (Schneider and Ingram, 1997).
However, while the earlier iterations of Social Construction and Policy Design allude to some of the intellectual roots of its concept of power, the most recent outlines of the approach are less specific. In the third edition of Theories of the Policy Process, the political power of a target group is defined as "the extent of its political resources, such as whether it is large, united, easy to mobilize, wealthy, skilled, well positioned, focused on issues of concern to it, accustomed to voting and contacting public officials, and so on" (Ingram et al., 2014: 109 f.). It should be noted that, as previously outlined, the authors qualify the use of this definition by stating that it needs to be considered hand-in-hand with the positive or negative social constructions or images of the respective group, i.e. whether elites and/or the general public approve of the target population receiving a benefit and are willing to frame the benefit allocation accordingly (Schneider and Ingram, 1993). Nonetheless, in spite of this integration of overt political power and interpreted deservedness, the concept of power is not expanded upon, which makes it difficult to see what a fully operationalized theory of power in policymaking would actually look like within this approach. While Pierce et al. (2014) argue that all three dimensions of power are present in Social Construction and Policy Design, this terminology is not used by the authors themselves, even though it may be implied. Additionally, while likely interacting, political power and social constructions are treated as orthogonal elements within the approach.
However, if following Lukes, both should be seen as different dimensions of the same general power concept and the consolidation of social constructions through policy can be considered second and third dimension expressions of power. In this context it is noteworthy that Lukes' work is not explicitly referenced by Schneider and Ingram. As previously shown, the most recent overview of the NPF only makes limited references to the concept of power and its theoretical foundations. However, in a 2016 book chapter on the NPF and agenda setting, Peterson and Jones argue that policy actors and coalitions operating at what the NPF calls the meso level (e.g., policy subsystem) "strategically manipulate policy narratives in order to obtain, alter, or prevent agenda access and content" (Peterson and Jones, 2016: 115). The text references Lukes, ' Schattschneider's (1960), and Riker's (1986) work on agenda setting as theoretical jumping off points. Given that the meso level is at the heart of a number of policy process approaches, including the ACF, this angle potentially offers the opportunity to examine the manipulation of narratives as a function of the second and third dimensions of power within policy subsystems, as well as at the NPF's macro level, which we argue is also suited for third dimension power analysis.
The above discussion has shown how different facets of power theory permeate current policy process theory and how different understandings of power can lead to diverging explanations of policymaking, roughly demarcated by what are commonly considered scientifically oriented policy process approaches (e.g., ACF) and critical policy theory (Social Construction and Policy Design). On the one hand, classic pluralist thinking lives on relatively unchanged in the ACF (and other process theories not addressed here), which leads to blind spots regarding more complex understandings of power inequality. We do not think that these blind spots should be regarded as conventional research efforts to maintain objective neutrality. Rather, it is an omission of important normative aspects of public policy that are the core of democratic concerns-or, why it is that so many care in the first place. On the other hand, Social Construction and Policy Design is rooted in the very normative concerns we reference-or more formally, in critical policy theory-and puts great emphasis on the interaction of power and value in producing or overcoming inequality, albeit without explaining how power operates in a way that is readily operationalized and measured. Finally, the NPF raises the issue of preference manipulation, but does not elaborate on this notion in detail. However, as referred to above, Peterson and Jones (2016) raise some aspects which can potentially be helpful to make a multidimensional power model empirically viable. Furthermore, as previously mentioned, the NPF takes a 'hybrid' position between scientifically oriented and critical policy theory, which potentially opens the door for incorporating a power concept that is both aware of structural power inequalities and the need for empirical rigor in operationalizing them, which is the focus of the sections below.

Making a Multidimensional Power Concept Empirically Testable -Pipedream or the Way Forward?
While a multidimensional power concept is theoretically intriguing and allows for more complexity in the discussion of power in policymaking, the question remains how it can be used in scientifically oriented policy research. In the following, we argue that power can in fact be operationalized using Lukes' power concept and why we think that this is a valid approach. For this purpose, we draw on the work of Matthew Crenson and John Gaventa, who have done empirical work on the second and third dimension of power respectively. Upon publication, Bachrach and Baratz as well as Lukes faced considerable pushback, with critics doubting that non-decisions and domination through preference manipulation can be empirically observed, let alone tested (Merelman, 1968;Polsby, 1963;Wolfinger, 1971). These reservations are not unfounded, as it is certainly challenging for researchers to pinpoint the role of social norms and beliefs in policymaking and determine whether they act as causal drivers when policy issues are left off the agenda or when people acquiesce to detrimental policy with no resistance. Nonetheless, we argue that even though an empirical treatment of complex power dynamics presents a challenge, it should not be disregarded by policy process scholars. Put in Crenson's words: "It is hardly more objective and scientific to ignore non-decisions and non-issues than to account for these phenomena" (Crenson, 1971: 4).
It is important to stress that we do not claim that Lukes' power concept fully defines power, and nor does it capture all its facets. Put differently, we are not trying to settle the age-old question regarding what power truly is and fully recognize that power has been and can be conceptualized differently. In particular, we acknowledge that as a result of using Lukes' conceptualization of power our focus is on 'power over' rather than the potentially transformative or empowering aspects of power identified in the many works of Michel Foucault. While we recognize that there are important differences between these two conceptions, we consciously choose to set aside this debate-and the associated disagreements about ontology and epistemology-in the interest of pursuing our central aim, which is to present an empirically viable concept amenable to scientific operationalization. 8 In line with what Adcock and Collier (2001) term a systematized concept, we argue that Lukes' three-dimensional power conceptualization offers a theoretical frame amenable to the operationalization of power using the NPF. In the pursuit of making power empirically testable, we find that a power conceptualization, which distinguishes between decision-making power (i.e. observable political power), agenda setting power, and preference-shaping power allows policy process scholars to capture power in greater complexity than has been the case thus far, while still enabling operationalization and allowing for further refinement of the concept of power down the road. In fact, existing in-depth studies by Crenson and Gaventa applying aspects of Lukes' conception of power have already described approaches to measure the second and third dimensions of power and recently a similar method has been suggested by Peterson and Jones (2016) in the context of the NPF, all of which indicate possible ways to integrate a power concept into policy theory. Here we revisit some of this classic power literature to help illuminate a path.
In The Un-Politics of Air Pollution (1971), Crenson identifies key aspects of the second dimension of power by exploring why air pollution is a 'non-issue' in some U.S. cities, but of concern to others. He does so by comparing the "local political climate" (Crenson, 1971:22) in cities which neglect the issue and those that have taken political action. Criticizing pluralism, Crenson posits that focusing on overt political action shows the researcher a limited range of involved actors, and uses survey data to illuminate the stance of political elites on air pollution to show that the dominance of some actors (perhaps identified in an "overt" power observation) likely played a role in which issues are represented. He found that the issue of pollution was likely not included on the policy agenda if elite groups 'behind the scenes' opposed action, even 8 -Michel Foucault characterizes power as being omnipresent, i.e. it is not limited to certain actors who exercise power over others (Foucault, 1980). Foucault's power understanding also finds expression in the literature on governmentality, which discusses the 'conduct of conduct', i.e. how individual behavior in a society is shaped through different means of power, such as coercion, persuasion and discourse (e.g., see Foucault, 1991). Lukes referred to this position as an "ultra-radical" constructivist lens (2004:12), viewing it at odds with his conceptualization of power. We concur, finding Foucault's approach less amenable to scientific operationalization than that of Lukes' three-dimensional approach.
if the general public considered it important, thereby limiting "the scope of decision-making" (Crenson, 1971: 178).
While Crenson's quantitative methodology constitutes a possible research strategy for investigating the second dimension of power, it does not take into account the third dimension of power, in which individuals could have been subjected to power in the form of preference shaping. Gaventa's ethnographic study Power and Powerlessness (1980) examines why citizens in Central Appalachia remain passive in the face of inequality, or conversely, why they rebel against it. A student of Lukes, Gaventa puts emphasis on the importance of understanding the interaction of all three dimensions of power. In order to counter criticism that the third dimension of power cannot be empirically specified, Gaventa details what he argues is a falsifiable methodological approach (Gaventa, 1980: 28). Similar to Crenson, Gaventa argues that non-issues usually occur under circumstances in which individuals experience inequality and as a consequence cannot make their dissent heard. To be able to show a relationship between inequality and apathy, Gaventa borrows Lukes' idea of counterfactuals, offering strategies for how these can be observed. In particular, Gaventa argues that it is not sufficient to limit research to the policy arena or subsystem, but that researchers interested in complex power dynamics should inquire into the experiences of affected target populations, who are "non-actors and non-leaders" (Gaventa, 1980: 27), and compare those with how the issues are framed on the policy-agenda. Gaventa's assigned importance to impacted target groups outside of the decision-making sphere stands in stark contrast to the ACF, which posits that research should focus on policy subsystems precisely because they "demarcate the integrated and non-integrated actors on a given policy topic" (Jenkins-Smith et al., 2017: 139). Gaventa suggests that the researcher should examine whether the ideology of those in power and the actions or nonactions of groups outside of the decision-making realm can be put in relation to each other. He further proposes comparing cases where the power of a dominant group weakens and to observe the actions of populations, which were previously not in power. Gaventa argues that when a group acts differently in a situation of changing power dynamics, compared to a group which faces similar inequalities but has no room for intervention, it can be determined that the third dimension of power has played a significant role in suppressing or even manipulating opposition. If, however, none of these methodological steps are successful in identifying variation, one can conclude that there is 'true' agreement with policy-decisions.
In summary, these two examples make clear that social scientists have explored how a multidimensional power concept could be operationalized and tested, both quantitatively and qualitatively. While it is ostensibly possible to make the argument that both Crenson's and Gaventa's in-depth case studies are dated and therefore do not pertain to the current policy process, it is interesting to note that Peterson and Jones come to a similar conclusion in their 2016 chapter on the NPF and agenda setting. Even though it may often not be feasible to conduct in-depth research on the scale of Gaventa's case study, the presented research nonetheless provides us with a strong point of reference that outlines avenues for testing complex power dynamics in policymaking and the interaction of power and belief systems. In the following, we make a case for how Lukes' three-dimensional concept can be leveraged in conjunction with the NPF to provide an empirically robust depiction of power in the policy process. In order to illustrate how the NPF could be used to achieve such an operationalization of power, we first examine and compare two policy process studies, one using ACF and the other Social Construction and Policy Design.

The NPF and Lukes' Three Dimensions of Power: Operationalizing a Multidimensional Power Concept
In this section, we analyze two applications, one of the ACF (Dougherty et al., 2010) and the other of Social Construction and Policy Design (Reich and Barth, 2010), with a focus on expressions of power therein. Both applications discuss in-state tuition eligibility of undocumented students within the U.S. An analysis of these applications allows us to see how different policy process approaches address the same policy area in diverging ways and how they (implicitly) treat power. We then transition into an illustration of what an analysis of the same policy issue and its inherent power dynamics could look like when the NPF is employed as a theoretical lens. In doing so, we formulate specific but generally applicable power propositions for the second and third dimension. Following Adcock's and Collier's seminal work on measurement validity of complex concepts (Adcock and Collier, 2001), Table 2 highlights how each dimension of power can be integrated with the NPF at the micro, meso and macro levels of analysis and provides suggestions on which methods would be appropriate to collect data, enabling researchers to operationalize the indicators, primarily through the NPF's well-established protocols for operationalizing policy narratives (i.e. form and content), discussed more below.  Dougherty et al. (2010) and Reich and Barth (2010) compare a U.S. state in which undocumented students were granted in-state tuition with one where a similar policy proposal failed and ask how this difference in outcome can be explained. Reich and Barth argue that the implementation of a policy granting in-state tuition eligibility in Kansas and the failure of the bill to pass in Arkansas can be explained by the social construction of the target population as well as by examining the political authority to frame the policy issue. In Kansas, undocumented students were portrayed as deserving young people striving for education, with the potential to positively contribute to Kansas' economy. In Arkansas, on the other hand, the debate was framed in terms of legality, meaning whether in-state tuition for illegal immigrants is compatible with the state constitution. As a consequence, Reich and Barth argue that it is important to recognize how the target group in question is socially constructed by policy actors, but that such an analysis is not sufficient without taking into account which actors have the legal authority-or power-to frame the issue and to thereby create a "perceived reality" (Reich and Barth, 2010: 425).
Dougherty et al. apply the ACF (as well as policy entrepreneurship theory) to Texas and Arizona and find that the Hispanic population in Texas was politically more powerful than in Arizona and hence able to form a successful advocacy coalition. They further note that this finding partly reflects a relatively accepting attitude towards immigration in Texas, i.e. it illuminates one of the belief systems which allowed the coalition in favor of in-state tuition eligibility to be successful. On the contrary, the general attitude towards immigration in Arizona is found to be more negative and this belief has been expressed through the action of policy makers, who have mostly taken action to restrict the rights of undocumented immigrants. In this context it is interesting to note that the first bill, which would have given undocumented students the right to in-state tuition, died in committee and hence never even reached the floor. Later Arizona passed a house bill which actively prohibits the granting of in-state tuition to undocumented students, instead of merely rejecting one that would allow it.
While it should be noted that an in-depth analysis of power dynamics is not the aim of these studies, it is notable that both applications do not address the kinds of questions that could emerge when leveraging a more nuanced understanding of power such as that provided by Lukes (1974), which would allow an analysis of why a policy issue does not make it onto the policy agenda-as initially happened in Arizona-or how structural inequalities, in particular concerning race and ethnicity, potentially shape the dominant belief system in a state. In fact, the authors of the two illustrative studies themselves note the limitations regarding an analysis of power dynamics within their respective studies. Dougherty et al. critique both the ACF and policy entrepreneurship theory for their pluralist bent, which sees power "as dispersed and, even if not equally distributed, not concentrated" (Dougherty et al., 2010: 164). Reich and Barth, on the other hand, point out that Schneider and Ingram do not define the complex concept of power in their framework and end up making assumptions as to its meaning for the purpose of their research. Consequently, both applications focus on the most visible expressions of power, i.e. its first dimension: in one case from a policy content perspective and in the other from a process-oriented perspective. In the case of Reich and Barth, this can potentially be attributed to the under defined nature of power in Social Construction and Policy Design, while ACF's pluralist assumptions do not allow Dougherty et al. to dig deeper into unequal power relations without consulting other theories. As a result, both applications are essentially forced by their theoretical lenses to neglect issues of non-decisions and preference-shaping, which could very well play a significant role in immigration policy.
For instance, a focus on agenda setting, or the second dimension of power, would allow a more meaningful analysis of why the non-documented immigrants were not successful in advocating for in-state tuition eligibility in Arizona, going beyond elite belief systems. Put differently, it can be hypothesized that the omission of the policy issue from the agenda and the subsequent restrictions of rights for undocumented immigrants might have cemented the status quo by 'mobilizing bias', meaning that policy elites were able to shape the policy agenda in line with their own beliefs to the exclusion of others. In addition, a focus on the third dimension of power would potentially make it possible to not only examine if mobilization of bias occurred, but to explore why a certain belief system, or constellations of social constructions, remain dominant in one state, but not in another, by considering cultural norms; or, in power parlance, the shaping of preferences over time.
According to John Gaventa (1980), the full extent to which power dynamics play a role in policymaking can only be understood if the different dimensions of power are analyzed as a whole. For instance, if we posit that non-decisions occur when policy-makers cement the status quo at the expense of disadvantaged groups, it is necessary to identify whether these non-decisions in fact go against the preferences or interests of the target (or untargeted) population. Furthermore, it is important to be able to show whether an identified power imbalance continues to shape beliefs inside and outside of a policy subsystem over time or whether shifts in power dynamics and belief systems can be observed. In the following, we aim to illustrate how these 'blind spots' could potentially be addressed in an analysis which uses the NPF as a theoretical lens. We argue that the epistemological flexibility of the NPF, together with its structural clarity (i.e. the distinction between micro, meso and macro narratives in policymaking), make it well equipped to explore these dimensions. In order to make our illustration easy to follow, we have formulated what we term power propositions for both the second and third dimension of power.
Current applications of the NPF already capture the first dimension of power (i.e. the decisionmaking process) quite well. At the policy-subsystem meso level, the NPF has a long standing tradition of specifying policy narratives through examinations of interest group narratives (e.g., Shanahan et al., 2013), legislative hearings (e.g., Smith-Walter, 2018, media (e.g., Crow et al., 2017) and other "non-trivially" cooperating elite actors, to use the language of the ACF. Embracing the language of the ACF is appropriate, as the NPF was very early on shown to be compatible with the ACF's coalitional approach at the meso level, especially in terms of identifying coalitional beliefs (see Shanahan et al., 2011), and has since regularly implemented research designs mirroring the ACF (see Shanahan et al., 2017). Such designs inherit the ACF's pluralist assumptive baggage. To see a multi-dimensional conception of power the NPF must engage questions that explicitly move beyond observable decision-making processes and the elite focus they entail. It is important to note that the primary aim of our illustration is not to endorse specific methods or research designs. Rather, we seek to illustrate how the NPF's concepts and constituent theories can be leveraged abstractly to study Lukes's three dimensions of power. For those interested in discussions of appropriate NPF methods and research designs, we would steer the reader toward Shanahan et al. (2018) and Gray and Jones (2015). The former addresses issues of research design and methodology from a scientific perspective, while the latter does so from an interpretive perspective.

The NPF and the Second Dimension of Power
We summarize our above review of the relevant policy and power literatures to indicate that if policymakers wish to control the policy outcome on a policy issue and have the capacity to sup-press opposition from groups and individuals, without their opposition becoming visible, nondecisions occur. This is likely to happen at the expense of disadvantaged (target) groups. Thus: NPF second dimension power proposition: It is likely to be able to observe non-decisions when policy elites' and target groups' beliefs, as represented by their policy narratives on the meso and micro level, are incongruent.
With regard to the second dimension of power, or agenda setting power, Gaventa suggests inquiring into the perspectives of "non-leaders" (Gaventa, 1980(Gaventa, : 1927 to identify potential rifts between their interests and the actions of those in positions of power. For the NPF, this translates into examining and comparing the policy narratives of elite and non-leaders, either from a scientific perspective (see Shanahan et al., 2018) or an interpretive perspective (see Gray and Jones, 2015). 9 In practice, the application of these kinds of comparisons would map out the structural narrative elements (i.e., setting, characters, plot, and moral) and the content elements (e.g., belief systems, causal mechanism and strategies) of elite and non-leader narratives, and then compare how the two differ and along what dimensions. We posit that said comparisons would help illuminate the extent to which non-leader perspectives are expressed by elites. Comparisons of these policy narratives to actual policy debates in authoritative forums (which may already be the case, dependent on the data sources of the elite narratives), as well as to actual policies (see James and Jorgenson, 2009), could further show to what extent mobilization of bias occurs, or the extent to which "some issues are organized into politics and others are organized out" (Schattschneider, 1960). Classic NPF micro-level methodological techniques such as interviews (e.g., Gray and Jones, 2015), focus groups (e.g., Jones et al. 2015;Smith-Walter et al., 2019) and/or surveys (e.g., Jones, 2014;Jorgensen et al. 2018) are all applicable for these kinds of research designs.
In the context of the immigration issue illuminated by our two illustrative studies above, the NPF can be used to examine the extent to which elite narratives on in-state tuition for undocumented immigrants at the meso-subsystem level differ from those of non-leader target populations at the micro level. However, it needs to be emphasized that in order to capture these potential 'non-agenda narratives' in the form of detailed first-hand accounts of undocumented immigrants, it is necessary for NPF researchers to step out of the policy subsystem and engage those not immediately involved in the policy process. Subsequently, if the researcher can identify substantial differences in terms of the structure and content of the narrative elements (e.g., characters, plot, moral, strategies, etc.) between the elite and non-leader narratives it is possible to conclude that only 'one side of the story' received attention in the policy subsystem. Put differently, we may witness the phenomenon of "narrative closure [which] results in a routinization of problem formulation and resolution" (Bennett and Edelman, 1985: 169) whereby policy elites ignore narratives that are not in line with their own preferences and beliefs, thus creating a non-decision. If the NPF is applied from a scientific perspective, all of the traditional methods mentioned above are likely applicable. However, we recognize that in some instances obtaining data from vulnerable populations can be quite difficult. In these instances, methodologies and data will be more limited, perhaps best served by interviews or participant observations and research conducted from an interpretive perspective.

The NPF and the Third Dimension of Power
Recall that preference-shaping power is the ability of a dominant group or entity to shape the beliefs and preferences of a target population, potentially against their own interests. Here the NPF can be used to identify policy narratives reflective of beliefs and preferences at the micro (individual), meso (group or coalition) and macro (institutional or cultural) level. The micro and meso level can be examined much as we describe above for the second dimension of power. Of particular interest for a third dimension of power analyses is the macro level, where the NPF has sought to examine enduring policy narratives emanating from cultural and institutional sources, the rationale being that historical macro policy narratives are proxies for underlying societal belief structures, which slowly change over time, if at all. 10 These macro-narratives can serve as a baseline of societal beliefs, how contested these beliefs are, and to what extent they manifest in public policy. Stated differently, macro-narratives become proxies for the social and institutional mobilization of bias: i.e., the extent to which certain narratives are embedded in social and institutional norms (Bachrach and Baratz, 1962).
Once the policy narratives at their respective levels have been identified, the next step is to determine what the target population's "true" interests are. In our estimation, this can be accomplished in two ways. The first possibility is that the target population provides information that indicates that they feel as if they are unable to express their true preferences, most likely due to some negative consequence or threat of reprisal when queried about their respective narratives regarding the policy issue under examination. However, when queried this population may reveal that their preferences and beliefs are in line with societal and/or elite preferences and beliefs. In this situation, the researcher may wish to compare these expressed preferences with information from an exogenous authority to identify potential discrepancies. This second approach is traditionally the domain of some critical theory approaches and controversial in that it assumes away individual agency and makes the researcher the authority on whether stated preferences equal a population's true interests or not. For example, a population may be found to hold a belief or set of preferences that run counter to their economic interests, which could be shown by citing research and literature that shows the population would benefit or suffer hardship from a particular economic policy, as was argued in the case of Republican voters in the U.S. who vote against their economic interests (Frank, 2007). We expect that this third-dimension expression of power is most likely to occur when macro and meso narratives are both congruent and homogenous, meaning the dominant macro narrative, which is taken up by policy elites, is (largely) not questioned by target populations and therefore not opposed Thus, we offer our first third dimension power proposition: NPF third dimension power proposition 1: When macro (cultural and institutional) and meso level (elite) policy narratives are both congruent and homogenous, target populations are more likely to express preferences and beliefs for policies that are harmful to the target population.
Regarding our ACF and Social Construction and Policy Design studies examining in-state tuition eligibility of undocumented students, such an approach could be illuminating. While we expect that it is highly unlikely that undocumented immigrants would not support in-state tuition eligibility, our conjecture is that this kind of analysis could, however, reveal a great deal about the preferences and beliefs of documented immigrants who may have come to their legal status via similar routes to those currently undocumented. In such a scenario, it is possible that the expressed preferences and beliefs of a disadvantaged population are shaped through 'standard policy narratives' which support commonly accepted cultural beliefs and as a consequence prevent individuals from realizing that a policy may be harmful to them longer-term (Bennett and Edelman, 1985). After performing a three-level NPF power analysis, the researcher may find that many of these legally documented immigrants genuinely support in-state tuition but feel they must be against it, less they suffer negative consequences in terms of their social standing, employment, or the like. Or it might be the case that they simply do not acknowledge that their situation was similar or analogous to the undocumented immigrants and thus hold preferences that are detrimental-if not to them directly, at least to people like them. In such a case, the researcher would need to complement their micro and meso policy narrative analyses with sufficient evidence showing that the policy in question actually does harm the population.
We now turn to our second third dimension power proposition. While we may see ongoing conflict in the policy subsystem, macro-level narratives, which often serve to maintain or 'normalize' structural power inequalities, tend to persist and be less visible, which impedes change in the overall power dynamics (Bennett and Edelman, 1985). Comparing and contrasting historical institutional and cultural policy narratives from the macro with the meso and micro level could allow researchers to highlight whether beliefs expressed in narratives of target groups and policy elites are reflected in overarching macro narratives and to track under which circumstances changes on the macro level occur.
Assuming that narratives are attached to dominant societal (and policy) beliefs, we offer the following second proposition on the third dimension of power and the NPF: NPF third dimension power proposition 2: If a previously disadvantaged target group gains access to decision-making and agenda setting, this can indicate a shift in macro narratives and therefore a shift in the dominant (policy) belief system. For instance, it is possible to imagine a situation where the undocumented immigrant populations in Texas and Arizona start out from a similarly weak position of political influence, but the population in Texas (target group A) becomes able to overtly exert influence on the decision-making process (first dimension) and the placement of issues on the policy agenda (second dimension), whereas the equivalent target group in Arizona (target group B) continues to be unable to make its voice heard. In such a scenario, we argue that it is possible to determine a change in the power of target group A by examining whether the micro narratives of said target group, which were not congruent with elite policy narratives at time 1, are represented in elite policy narratives and consequently appear on the policy agenda at time 2, meaning we no longer witness "narrative closure" (Bennett and Edelman, 1985: 169).
In order to analyze whether such change can potentially be associated with a long-term shift in dominant societal beliefs, which the NPF assumes to be reflected by (historical/institutional) macro narratives, it is then helpful to examine the individual elements of narrative form through which narrative content is expressed. For instance, the researcher can trace whether the predominant portrayal of characters (e.g., villain, victim, hero), the plot (i.e., how the story is organized), as well as the moral of the story (e.g., call to (policy) action), have shifted longterm in a way that has given voice to target group A and has thus enabled a transfer of power to said group (Gaventa, 1980). While such an approach will not definitively establish a causal link between the increased power of a target group and changing macro narratives, identifying a correlational long-term shift in these narrative elements can indicate a connection between the changed power distribution and shifts in dominant societal beliefs. Conversely, in the case of target group B, which did not see an increase in its decision-making and agenda-setting power, a researcher may not witness a shift in the relevant macro narratives should they examine the elements of narrative form over time.
While macro-level studies within the NPF are not as ubiquitous as micro and meso, the NPF has shown it can be used to track this kind of variation in preferences over the course of many years by describing the macro policy narratives of the prevalent culture (e.g., Ney, 2014;Tuohy, 2019) and/or institutions with jurisdiction over the target populations (e.g., Veselková and Beblavý, 2014;Peterson, 2019). In a next step, identified macro level narratives can then be compared to both meso (e.g., subsystem) and micro (i.e., target populations) level policy narratives. This could, for example, be done by a longitudinal analysis of narrative texts by those advocating for the rights of undocumented immigrants and the narrative texts of policy elites in hearings, press statements, etc. on immigration and to compare the respective beliefs or preferences to overarching accounts, such as State of the State addresses. This could be accomplished by means of content analysis and retrospective interviews, but ostensibly it is also feasible to apply quantitative methods, perhaps time series approaches (e.g., Peterson, 2019) or event history analyses (Büthe, 2002) to test whether changes in macro narratives are statistically significant.

Conclusion
Existing policy process theory does not pay much attention to the concept of power and sometimes even dismisses it as too abstract or too complex to play a role in empirically anchored public policy research. However, while operationalizing a multidimensional power concept may be challenging, working towards an empirically viable power theory should not be dismissed out of hand based on its complexity. Even though they may be dated, the in-depth case studies by Matthew Crenson and John Gaventa show that it is possible for social science scholars to find evidence for processes such as selective agenda setting, non-decisions, and preference shaping by policy elites through means of empirical research. In this article, we have aimed to strengthen this argument by illustrating the drawbacks of existing policy process theory, which either does not clearly define its understanding of power (as domination) and thereby makes itself vulnerable to criticism, or which focuses on visible expressions of power in a pluralist subsystem. We have further taken a first step towards making a complex theoretical understanding of power applicable to empirical public policy research. In particular, and following Lukes, we have developed suggestions for operationalizing the three dimensions of power as domination or power over, which leverage the epistemological flexibility of the NPF through its differentiation between micro, meso and macro narratives to tease out the second and third dimension of power in the policy process. More specifically, we argue that an analysis of cultural or institutional (macro), and elite (meso) narratives, can help to understand how dominant narratives shape individual-level preferences of target populations-possibly to their disadvantage-and how shifts in these narratives can potentially give previously excluded target groups access to the agenda setting arena.
While it is beyond the scope of this article to realize the suggested approaches, which admittedly relegates our argument to remaining theoretical in nature, it is pertinent to recognize that disadvantaged and minority populations continue to face structural power inequalities in different policy contexts. However, we would like to note this article is by no means equipped to provide a final definition of power, but instead uses an existing definition to enable an operationalization of power that is amenable to scientifically oriented policy research. We further acknowledge that by focusing on 'power over' or power as domination we do not discuss the transformative potential of power. While we do not wish to dismiss the potential of analyzing power as a potentially productive, agency-enhancing force, analyzing these facets of power is not the main aim of this article. That said, even though power remains a complex and multifaceted concept, policy process theory needs to be able to analytically dissect power dynamics if it wishes to remain relevant, and we see this article as an initial step in this direction. Future research is needed to put our suggestions to an empirical test and to further develop the empirically viable power concept we envisage.