Beyond powering and puzzling: the political dimensions of policy learning

ABSTRACT

Over the years, policy learning has emerged as a theoretical lens to provide an alternative understanding of powering-based approaches to the policy process and policy outcomes. This has been a success story, with policy learning gaining presence among theoretical frameworks on the policy process. However, the puzzling-powering dichotomy is reductive. In real-world policy processes, we find both puzzling and powering. Also, we cannot reduce the value of learning to what is not explained by power alone. This Introduction shows how puzzling and powering interact and with what consequences on policy processes, moving beyond the dichotomy. We present how the four contributions of this Special Issue explore this territory and suggest how to approach the design of governance architectures for learning.

KEYWORDS:

• Ambiguity
• policy learning; policy process theory
• political power
• uncertainty

1. From powering to puzzling: a success story

One of the reasons behind the success of policy learning as a conceptual-analytical lens on the policy process is that it has offered a more comprehensive and realistic understanding of policymaking. This went beyond the traditional view of policymaking as an outcome of power struggles underlying who gets what, how, when, and with what effect (Lasswell, 1950). Laswell’s foundational approach is consistent with the classic understanding of political science as the creation and use of constrained social power (Goodin & Klingemann, 1996, p. 7). Unconstrained power is pure force. In real-world political systems, behaviour is constrained by resource interdependence, institutions, and ideational constructs like social identity and collective understandings of notions like welfare and distribution. Constrained as it may be, in the classic approach to public policy, the creation and deployment of political power is fundamental to the explanation of the outcomes of policy processes. What goes in ‘the black box’ of the policy process is a set of resources, owned by actors with different endowments, strategies, and objectives. What comes out of the box is a public policy outcome that crystallizes, for some time at least, not only technical solutions to policy problems but also winners and losers of the policy game.

Policy learning sees policies and policy processes in a fundamentally different way. For a policy learning thinker, actors do what they do, also as a function of what, how, and why they learn. The policy learning’s perspective on the policy process is ontologically and epistemologically different from that of classic political science models. For the latter, resources are decisive. For a policy learning thinker, resources do matter of course, yet the focus shifts to cognitive structures, layers of belief systems, and the psychological-cognitive mechanisms that actors leverage to solve policy problems and even deploy those resources. Instead of being fuelled solely by a blunt application of power, the policy process is also fuelled by various manifestations of how actors seek to learn. This means that the key identifier of a process is not necessarily ‘who gets what’. Rather, it is whether learning appeared or did not appear in a given process (Radaelli, 2009), or was a matter of individuals and groups in organisations (Heikkila et al., 2023). And further, a policy process can be characterised in terms of learning about policy instruments, goals and paradigms (Zaki et al., 2023), or, at the macro-level, social-collective learning (Hall, 1993).

So, where does this alternative approach to the policy process stem from? At the risk of oversimplification, we argue that it originates from the fact that actors cannot always calculate the pay-offs of alternative courses of action. They cannot attribute a value to making decision A or decision B, C, down to N. Often uncertainty is such that actors in the policy process can only formulate vague expectations about the outcomes of a few possible decisions that appear on the cognitive radar. Only a few possible courses of action appear on the cognitive radar, and not with the necessary information set to calculate pay-offs.

Ambiguity adds to the problem of making strategic calculations about the consequences of this or that behaviour (Kingdon, 1984). Ambiguity implies that policy problems are increasingly prone to multiple and even conflicting interpretations and, thus, can be understood and approached in fundamentally different ways (Cairney et al., 2016). Furthermore, actors at the policymaking table change as the policy process is set in motion (Kingdon, 1984). For example, a minister may kick off a discussion about a policy innovation, but then for months the technical bureaucrats of three or four departments take over and discuss with the regulators and pressure groups, only for the decision-making process to open up in a further stage with open contestation and social pressure, experts taking on advocacy positions, and interventions from international organisations.

The venues or institutional fora where the search for solutions for already ambiguous problems is carried out change more than once – a process may be very open at the beginning, with a range of economic and social pressure groups involved, yet then it may move behind closed doors in cabinet decision-making and then open again, but in an institutional forum like the parliament. Furthermore, the very social or economic problem that a policy solution is supposed to handle is constantly re-defined by new events, like elections, scandals, wars, financial crises, and pandemics.

In short, the presuppositions of policy learning thinking seem much more realistic than the ones of the classic approach to public policy as the pursuit and use of power. On the one hand, we find an ontology of actors that puzzle in a world of ambiguity and uncertainty. This puzzling is not only about technical solutions to policy problems but also about how political power is employed. On the other, we find combinations of actors-strategies that move like power-propelled billiard balls on a smooth baize. In the classic ‘who gets what’ approach, one could predict policy outcomes as the results of the parallelogram of forces, substituting the physical value of a force with a power index. Not so with the learning billiard, where balls zig-zag and change direction after being hit.

With its close approximation to reality and the acknowledgement of ambiguity, policy learning has gained good currency among applied and conceptual researchers. This success story has evolved over decades of research. At the time of Hugh Heclo (1974, p. 305), the case of learning was made by observing that, in the end, policy actors not only engage in powering. Puzzling (the gerund that defined the beginning of the success story) can then be conceived as a residual or an ‘error term’ denoting what cannot be explained by an analytical lens of powering. Moving from this ambitious intellectual objective, policy learning scholarship flourished over the past three decades. This growth was heralded by Bennett and Howlett’s (1992) contribution which synthesised the state of a (then already!) emerging field and created one of the earliest learning typologies, mostly drawing on existing work carried out by Etheredge & Short (1983), Rose (1991), May (1992), Heclo (1974) and Hall (1993). There, learning types were assessed, appraised and finally categorised: instrumental learning, focused on updating understandings of policy instruments – that is, how instruments work, and their calibration and viability (Lee & Van de Meene, 2012); social learning, about the evolving social construction of policy issues through discursive and reflexive processes (Hall, 1993); and political learning, concerned with political strategies and an interest in knowledge about policies that gives more chances to be re-elected (Gilardi, 2010).

Since then, the field has remarkably expanded (Dunlop et al., 2018). As this happened, and since there was never a single root or intellectual champion for the family tree (Dunlop et al., 2018), policy learning conquered terrain by embracing a range of ontologies and epistemologies, ranging from mechanistic, positivist, interpretivist, to social constructivist, and discursive institutionalist (Grin & Loeber, 2007; McCann & Ward, 2012; Oliver & Pemberton, 2004; Zaki & Radaelli, 2024). The constructivist interpretivist approach views learning from the perspective of social processes of negotiated problem definition, communication, interpretation, understanding, and solution assessment (Grin & Van de Graaf, 1996). Others focus on how actors process new information through the lens of pre-existing, relatively coherent belief systems (Sabatier, 1988; see the debate in this journal over twenty years ago: Dudley, Parsons, Radaelli, & Sabatier, 2000).

Within each of those ontological-epistemological approaches, the details of what policy learning is have become more finely textured. For example, we find the understanding of learning as a relatively enduring change in beliefs (Sabatier, 1988); a process in which knowledge about policies, administrative arrangements, institutions in one time and/or place being used in the development of policies, administrative arrangements, and institutions in another time and/or place (Rose, 1991); attempts to adjust the goals or techniques of policy in response to past experience and new information (Kerber & Eckardt, 2007); deliberate and critical reflection on priors (Rietig & Perkins, 2018); or, more broadly, a pursuit and consumption of policy issue-related information and knowledge by policy actors aiming to address policy problems (Zaki et al., 2022).

This has led to substantive theoretical developments where learning has found very comfortable homes. For example, the Advocacy Coalition Framework (ACF) approaches policy learning as an outcome of exogenous shocks that imply value trade-offs, leading to enduring changes in some belief systems, that, in turn, affect policy behaviour (Sabatier, 1988). By unpacking the causal mechanisms of learning, Heikkila and Gerlak’s Collective Learning Framework addresses the sub-processes by which learning aggregates from individuals to collectives through discursive processes of knowledge and information acquisition, translation, and dissemination are described (Heikkila & Gerlak, 2013). Dunlop and Radaelli (2017) explain causal processes by viewing learning’s aggregation across the micro-meso-macro levels. Extending the claims of organisational learning theories (especially Argyris & Schön, 1978), Lee, Hwang, and Moon (2020) construct a framework for quadruple loop learning showcasing how policy learning effectively takes place under evolving conditions through ongoing interactions between objectives and context. In a theory of policy learning governance Zaki (2024) finally recently connected agency, specifically the agency of governing actors, to policy learning processes that are strategized, designed, and continuously adjusted.

2. Reimagining the debate: going back from puzzling to powering

The aforementioned theoretical and conceptual developments helped advance policy learning to the rank of a policymaking ontology, making it a coherent and rather independent lens to viewing policy processes and outcomes (see Kamkhaji, 2022; Zaki, 2024). Given policy learning’s well-established ontological and epistemological position, it is no longer viewed as a residual explanatory lens covering what power cannot explain. A byproduct of this success is that learning scholars predominantly do not feel the need to theoretically justify why they are not exclusively looking at the role of power anymore. Gradually, the question of ‘where is the role of power in our learning-oriented explanations of policy processes and outcomes?’ fades away or takes a backseat. In other words, in many instances, the field’s analytical purview is often inadvertently caught in a dichotomy between puzzling and powering. Yet, this dichotomous view is not entirely (if at all) convincing. Powering and puzzling exist in each other’s territories, and policy actors clearly engage in both, powering, and puzzling simultaneously (Zahariadis, 2016, p. 464).

Policy learning’s political dimension was laid bare a long time ago of course. Weiss’s pioneering work on knowledge utilisation pushed social scientists to consider not only the ‘enlightenment’ functions of ideas but also their symbolic and political uses in the policy process (1979). Yet this has been a lesson that has taken a while to stick. Too often we think of policy learning as a truth-seeking activity that reveals information and holds the promise of better policy design (think for example about some discussions of ‘evidence-based policymaking’). But everything we know about policymaking tells us that learning is imbued with politics. Learning is about the very power to produce knowledge, set standards and create the commonsense. At its heart then, policy learning is an activity that faces these two directions. This complexity is one of the reasons it remains an enduring and compelling area of study. Thus, puzzling and political powering are entwined. Political power and resources affect learning, and learning also shapes how political power is accumulated and wielded at and across multiple levels micro, meso, and macro (Zaki et al., 2022).

When we destabilise the powering/puzzling dyad, we see actors exercising their power and agency to shape how learning takes place by strategising learning objectives, designing learning processes, and adjusting them to ensure their technical and political objectives are being met. Hence, to truly harness the full potential of policy learning in understanding the causal mechanisms in policymaking, we must account for the role of political powering therein.

The political power aspect is omnipresent across different dimensions pertinent to learning from the context and institutional designs surrounding learning, to teachers and learners involved, and the lessons being learned. Puzzling takes place in a powering context, or, as others have argued, reflection may need powering because actors use their power, authority, and control to forge the political discourses that change beliefs (Stock et al., 2021). The very fact that someone learns (whilst others do not) at a certain time (t) is a formidable resource in terms of exercising political power at a future time (t + 1). Also, if the exercise of political power is constrained, as per the classic definition, then policy learning too is constrained. Yet, constrained by what? Most likely and most immediately, by institutions (Moe, 1990), procedures (Moe & Wilson, 1994) and policy design legacies that stack the deck in favour of some interests and actors (Ingram & Schneider, 1997). As Heclo noted some five decades ago: ‘puzzling over problems … occurs in the networks of informed activists and policy middlemen as well as through the learning pathways opened and closed by previous policy decisions’ (1974: xviii).

Empirically, the distribution of power across actors and trust across key interests in a network facilitate social learning (Levesque, Calhoun, Bell, & Johnson, 2016). Such collective learning has been captured in the idea of ‘learning through bargaining’ where interdependent actors learn about their own and each other’s preferences, what strategies work in that context and point at which to compromise (Dunlop & Radaelli, 2013, 2018). Alternatively, institutional power may dominate the boundaries of the possible in learning. For example, Zahariadis (2016) shows that even under external conditionality during the financial and economic crisis (2009–2013), the Greek government was able to learn when making reductions in public personnel. Some of the policy changes he observed cannot be explained by power alone. This chimes with the claim about the presence of learning under conditionality and hierarchical power structures (Dunlop & Radaelli, 2013). Intense conflict can actually force policymakers to learn (May, 1992; Zahariadis, 2016). Taken one step further, the learning pathways (to borrow from Heclo), may be severely narrowed by the presence of rigid, strong power structures, as shown by Fenger and Quaglia (2016) in their comparative analysis of the responses to the global financial crisis. In the same vein, new opportunities to learn can be deliberately created through the design of policy instruments (Dunlop & Radaelli, 2019 on impact assessment in the EU).

Moving beyond how learning takes place, the political contours of the learning zone are fundamentally shaped by actors and specifically who gets to teach on an issue. For sure the existence of ‘teachers’ is a presupposition. Yet, in the policy process, teachers do not appear out of thin air. Those teachers owe their position to socially constructed understandings of what knowledge is, and to the powering political constellations of authority often ‘congealed’ in national and international institutions like the International Monetary Fund or the European Commission that create forums for teachers to exist. Teachers can also be in competition – seeking to overpower each other’s schools of thought or epistemic preferences (e.g., Dunlop, 2017).

The presence of cognitive heuristics makes this anthropomorphic dimension of policy learning crucial. The power of teachers is mediated by the beliefs already held by policymakers. The messages of authoritative experts are filtered. In that process, lessons can become distorted, repurposed, misunderstood and so on. In short, politicised. In such cases, the only way to shift the dial may be to seek a new audience. For example, when European Commission policymakers cherry-picked evidence on the safety of milk produced by cows treated with a biotech yield enhancer, the epistemic community convened to teach it exited the advisory process and engaged in venue shopping lobbying the UK government and more specialist scientific international organisations (Dunlop, 2009). Meanwhile, the European Commission as ‘learner’ assembled a new group of scientists from different disciplines willing to teach alternative and more politically aligned lessons. The point this story tells is that the role of the teacher can be highly contingent as powerful learners work out what lessons ‘fit’ their political context and worldviews (Zaki & Wayenberg, 2021).

Turning to what is learned – the lessons – resource asymmetry can explain why societies may learn the ‘wrong lessons’: they may be puzzling in the wrong way because power operates as the social anchor of collective learning processes. The competition for power explains why policymakers do not seek or accumulate lessons about what is useful for policy improvement, but rather learn only what helps them to win elections (Gilardi, 2010). These socio-political forces shape the contours of prevailing knowledge around an issue. Frequently, these politics see policymakers and expert advisors coming into conflict over the types of evidence to rely on and the extent to which policy action should wait for science to catch up. The policy logic of ‘satisficing’ (Simon, 1957) means learning invariably happens in partial ways where a mix of sources from science, interests, society and beyond are pulled together into the lesson (see Freeman, 2007 on learning as bricolage).

Finally, there is the opposite of problem-driven learning (Gherardi, 1999). References to learning can be deployed as the symbolic veneer of reasoned argumentation. Politicians, bureaucracies, governments, and organisations ‘talk’ the language of learning to obfuscate crude power dynamics. To ‘talk learning’ is how an organisation seeks to erect semantic barriers to critiques, the argument being, more or less, that in the past for sure an organisation made mistakes, but since lessons were learned, today that organisation can no longer be criticised. A learning organisation (which in this case means an organisation that self-defines its behaviour as learning) is therefore an organisation that is harder to contest than a powering organisation (Alam, 2007; Radaelli, 2009). Even when there is no intent to obfuscate power, learning can be just trivial: an organisation that has existed for a long time must have learned something along the way; otherwise, it would not have survived in its environment (Radaelli, 2009).

3. Bringing together powering and puzzling

Building on the above state of the art concerning powering and puzzling in learning literature, this special issue serves two main purposes. The first is to reconcile the perspectives of powering and puzzling across multiple levels of analysis and within several theorisations and empirical settings. This ranges from the individual cognitive level to collectives, the processes of learning and lessons learned, and the overall quality of learning. The second is to explore the conceptual and empirical territory that comes after disposing of the dichotomous approach of ‘powering versus puzzling’. To do so, this special issue presents four contributions.

In Heikkila, Gerlak, and Smith (this issue), we start at the individual level by diagnosing individual barriers to collective learning. Learning is political. That puzzling and powering are intertwined makes ‘good’ policy learning hard. We might ask who doesn’t want to learn? But such a question fails to grasp that the blocks to functional learning start in the human mind. Heikkila, Gerlak, and Smith (this issue) take us into that mind exposing the forms and implications of seven of the big cognitive heuristics that shape us all. Drawing on cognitive and social psychology, they lay bare the impact of emotionally rooted biases and how they perform in different governance contexts. Importantly, this conceptual analysis reveals the biases that distort learning can be both enabled and mitigated by differences in structural conditions, social dynamics among policy actors, issue complexity and broader exogenous forces. In revealing the learning barriers that arise when, for example, adversarial policy settings work to reinforce in-group/out-group dynamics, practical design solutions emerge. The authors’ blueprint offers a variety of ways to put powering at the service of puzzling which accepts the limitations of the human mind, urgency of complex policy problems and emancipatory possibilities of institutional design to remake governance worlds.

Moving to the collective, in the second contribution to this special issue, Zaki and Dupont (2024) focus on the nexus of political and technical learning being undertaken by commonly overlooked students of political learning, that is scientific experts working in a fluid and highly politicised environment. The authors theorise and trace the process of expertise displacement due to shifting political conditions in the case of EU climate policy development between 1990 and 2022. They illustrate how these shifts undermine scientific experts’ access to and influence on policymaking, nudging them to manoeuvre the vagaries of a rapidly shifting political landscape, and speak truth to power. Their analysis identifies three sets of political advocacy strategies aimed at enhancing the role of scientific expertise in policymaking, being: Narrative and semantic strategies (policy issue-oriented), Socialisation strategies (actors-oriented), and Governance strategies (systems and structures-oriented). In doing so, the authors enter a new conceptual and empirical terrain by moving beyond a focus on scientific experts as teachers in the policy process where their expertise is used (and abused) by those with the power to make decisions. Rather, they develop a novel understanding of how politics and political learning carry technical and epistemic lessons into policymaking. These findings help us better understand how experts actively advocate for science’s role in public policy and employ powering within puzzling.

In the third contribution, Stark and van der Arend (2024) start from the process of learning a lesson. Puzzling is also how a policy lesson travels across time and space. Triggered by the process of learning, the authors undertook a series of interviews within the disaster management community in Queensland/Australia. They thus identified barriers that restrict a lesson from moving as well as strategies that practitioners can use to push it forward. In their view, success in policy learning heavily depends upon the so-called dynamic capacity that stirs a lesson through the institutional fabric of a(ny) policy system. Besides making this valuable conceptual contribution that better links the micro and meso – levels of learning, Stark and van der Arend also created a traveller’s guide to policy learning. After all, a lesson’s dynamic capacity thrives under a series of conditions such as cognisance of the implementers required to act and/or when the latter makes translation efforts to reconcile varying meanings along its way. And, ultimately, such good lessons empower anyone interested in policy learning, from theory as well as practice.

In the special issue’s fourth contribution, we start with the quality of the learning process. The quality of lessons and specifically their suitability depends on the learning ‘mode’. We know, for example, that it is possible to learn in the ‘wrong mode’ – a mode of learning that is dysfunctional for an organisation. Entire political systems can learn how to use more hierarchy (more rules, more constraints imposed from the top, as arguably was the case of austerity) when, instead, learning in more reflexive ways is more appropriate and effective. Möck and Feindt’s article (this issue) takes us deep into these ‘misfit’ dynamics. They advance a typology to advance our understanding of the distinct biases found in different learning environments and how these link to particular types of suboptimal learning processes and outcomes. The case of innovations in grassland framing stands as an exemplar of how modes can be shifted – in this instance from narrow epistemic learning to reflexive forms – through conscious institutional design. The use of participatory co-productive fora moved the dial away from science-heavy learning forms which dominate socio-technical transitions to more inclusive messier learning processes. The case demonstrates the value of early recalibration to avert policy failures down the line.

4. Where do we go from here?

The complexity of policy settings and politicised nature of knowledge production frustrates drawing direct links between learning and change. This may be unsatisfactory, but it is reality. And it is important. Integrating puzzling and powering surfaces the political realities of policy learning (disappointing as they may be). This analytical approach reminds us that we cannot (and should not) assume policies will get better (whatever that may mean) because learning has happened (Moyson et al., 2017). Not only is learning normatively neutral (Heikkila and Gerlak, 2013), but it is also frequently associated with policy failure (Dunlop, 2017; Newman & Bird, 2017).

Yet this is not a message of fatalism. This reality check means we can proceed with our eyes wide open. We may proceed by pinpointing the change we want to understand, and perhaps, promote. Heikkila and Gerlak (2013) distinguish between two outcomes – the cognitive where beliefs and ideas are refreshed, rejected, or renewed and the behavioural where rules and instruments are adjusted. We can get deeper into both using classic studies. Most obviously, Hall (1993) and Argyris and Schön (1978) provide ways of thinking about how deep into policy these outcomes cut.

Moving to another dimension, what are the implications for policy design? By moving on to consider learning-sensitive governance architectures we echo the ambitions of the earliest work on policy learning. Dewey, Simon, Deutsch and Lindblom all, in their own ways, kept in their sights the need to take seriously the practical implications of policy learning (Dunlop & Radaelli, 2018). When we ask how to design for the politics of policy learning, we need to be clear about (at least) two things: what type we have in mind and what does it need to perform? Learning is often unplanned, but it is not random. Learning takes different forms and involves specific actors depending on the political context. These different forms or modes perform well against some measures and less so against others. Take, for example, issues characterised by high levels of uncertainty where experts dominate knowledge production. Here we will likely see epistemic learning where policymakers seek out or gather experts into advisory groups. These settings perform well in basic enlightenment terms – making complex issues more accessible to time-pressured policymakers such that they can start to identify their preferences (Haas, 1992). Moreover, policymakers can use these narrow learning sites to depoliticise divisive and emotive issues. Epistemic learning modes fall down however on, among other things, wider social legitimacy, and diversity of voices. Baekkeskov and Öberg’s (2017) comparative study of policymaking in Sweden and Denmark during the 2009 H1N1 pandemic amply shows how the mechanisms of epistemic learning that drove consensus-building worked well because they froze out wider societal discussions.

Thinking through these dynamics of mode suitability provides direction on design as they enable some level of prediction about likely pitfalls of learning in a single mode for too long. Policymakers can choose to accept these, plan and mitigate for them or engage in re-design while being aware that it would involve a new set of trade-offs. So, in the epistemic scenario, policymakers could choose to change the mode of learning by phasing in social representatives to the advisory arena. This is precisely the territory that Möck and Feindt (2023, this issue) explore in their analysis of socio-technical developments in German grassland agriculture. Their case shows the practical power of understanding the mode of learning and how good a ‘fit’ it is with the prevailing socio-political rhythms. Learning research can be at the service of policy actors where what is politically important at a given time – say identifying preferences in a crisis or winning back wider social trust – can be advanced by institutional designs sensitive to the politics of an issue that alter the preeminent learning mode.

Acknowledgements

Early versions of the Special Issue manuscripts were delivered to the International Workshops on Public Policy of the International Public Policy Association, Budapest, June 28–30, 2022. We are grateful to the participants and discussants for their comments.

Disclosure statement

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

Additional information

Notes on contributors

Claire A. Dunlop

Claire A. Dunlop is a Professor of Politics and Public Policy at the University of Exeter.

Claudio M. Radaelli

Claudio M. Radaelli is a Professor of Comparative Public Policy at the European University Institute, Florence, on leave from University College London.

Ellen Wayenberg

Ellen Wayenberg is a Professor in the Department of Public Governance and Management at the University of Ghent.

Bishoy L. Zaki

Bishoy L. Zaki is a Professor in the Department of Public Governance and Management at the University of Ghent.