• policy learning;
  • typologies;
  • concept formation


  1. Top of page
  2. Abstract
  3. Conceptualising Policy Learning: Four Learning Genera
  4. Expanding the Property Space
  5. Discussion
  6. Conclusions
  7. About the Authors
  8. References

The field of policy learning is characterised by concept stretching and a lack of systematic findings. To systematise them, we combine the classic Sartorian approach to classification with the more recent insights on explanatory typologies, distinguishing between the genus and the different species within it. By drawing on the technique of explanatory typologies to introduce a basic model of policy learning, we identify four major genera in the literature. We then generate variation within each cell by using rigorous concepts drawn from adult education research. By looking at learning through the lenses of knowledge utilisation, we show that the basic model can be expanded to reveal sixteen different species. These types are all conceptually possible, but are not all empirically established in the literature. Our reconstruction of the field sheds light on mechanisms and relations associated with alternative operationalisations of learning and the role of actors in the process of knowledge construction and utilisation. By providing a comprehensive typology, we mitigate concept-stretching problems and lay the foundations for the systematic comparison across and within cases of policy learning.

In this article, we critically reflect on the state of policy learning: there has been a proliferation of concepts and models, with the result that different strands of the literature, both in international relations (IR) and in political science, tend to talk past each other. We suggest a way forward by blending the seminal Sartorian approach to classification with recent work on typologies in qualitative research (Collier et al., 2012; Elman, 2005; George and Bennett, 2005, ch.11; Sartori, 1970). We build an explanatory typology that shows the connections and scope conditions for different forms of learning. Debates in this field resemble the classic ships crossing in the night because scholars have failed to sort out and clarify concepts and the relationships between them. Our proposal revolves around the distinction between four major genera (primary or basic categories) of learning and the ‘species’ (that is, secondary categories) within each genus.

At the outset, however, we have to define learning. At a general level, most accounts depict learning as the updating of beliefs based on lived or witnessed experiences, analysis or social interaction (Hall, 1993; Heclo, 1974; Meseguer, 2006; Checkel, 2001, respectively). For some scholars this updating process is intentional and conscious (Hall, 1993); for others it is more organic (Heclo, 1974); and for yet another group of scholars it is ‘un-intended’ (Liberatore, 1999). Such epistemic updates also vary in their depth. David Dolowitz distinguishes between ‘hard’ and ‘soft’ learning where deep, analytical processes are contrasted with superficial, limited analysis (Dolowitz, 2009, p. 323). In IR, Jack Levy treats learning in similar terms. Echoing the seminal work of management theorists Chris Argyris and Donald Schön (1978) on ‘single-’ and ‘double-loop’ learning, learning is ‘simple’ where political actors alter their strategies, and is ‘complex’ where fundamental preferences and objectives are questioned and revised (Levy, 1994, pp. 285–92). Jeffrey Checkel adds that the second complex understanding of learning represents the area where constructivist analysis adds value with its focus on the transformation of identities as well as preferences (Checkel, 2001, p. 561).

Learning as a process is often left undefined entirely, however, with analysts preferring to identify the products of learning. For example, learning has been described by Peter May (1992, p. 332) as ‘lessons’ that are drawn (sometimes cross-nationally –Rose, 1991) about ‘the viability of policy instruments’, ‘the social construction of policy problems’ or ‘the political feasibility of policy proposals’.

Given our aim to bring some order to this ‘conceptual minefield’ (Levy, 1994), our argument is not contingent on any specific definition of learning. Accordingly, we treat learning as the updating of beliefs at its most general level. Using this minimal definition we will be able to handle the entire literature we refer to in the remainder of the article.

There is renewed interest in policy learning, as shown by three special issues published in 2009 – two on diffusion and transfer (Dolowitz, 2009; Evans, 2009, respectively) and the third on the EU as learning organisation (Zito and Schout, 2009a) – and a collection of papers revisiting policy transfer in Political Studies Review in 2012. Further, some of the most popular frameworks developed since the early 1990s, such as policy transfer (Dolowitz and Marsh, 1996), the epistemic communities approach (Haas, 1992) and the advocacy coalitions framework (Sabatier and Jenkins-Smith, 1993; Sabatier and Weible, 2007), have emerged from the attempt to generate theoretical propositions about learning in domestic and international policy domains.

Yet even the most casual of observers would note that the field is struggling to produce systematic and cumulative knowledge on this topic. The focus on learning marked an important reorientation of the political science literature (Bennett and Howlett, 1992) away from economic accounts but, arguably, its full potential has not been reached; the field has not been marked by studies building on each other. A lack of communication within and between politics' sub-disciplines has resulted in conceptual stretching (Sartori, 1970) and has diluted policy learning's analytical purchase.1 This bleak verdict simply echoes that first registered twenty years ago by Colin Bennett and Michael Howlett (1992).

The analysis of learning – a classic topic since Karl Deutsch (1966), Herbert Simon (1947; 1957) and, soon after, Hugh Heclo (1974) and Charles Lindblom (1965, but see also the seminal work on muddling through –Lindblom, 1959) – has taken place within several self-contained sub-fields where there is empirical progress but relatively little interest in conversations across the discipline. At the same time, theoretical models (for example, Volden et al., 2008) have not as yet fully connected with the work of empirical analysts, perhaps with the exception of the field of policy diffusion (Dobbin et al., 2007; Graham et al., 2008; Meseguer, 2008; Weyland, 2005). The literature on governance has also added to the interest of the last fifteen years or so in rediscovering the analytical properties of learning, with emphasis on the organisational-institutional properties that generate learning outcomes (Eising, 2002; Olsen and Peters, 1996), the notion of experimentalist-democratic polyarchies (Gerstenberg and Sabel, 2002), varieties of principal–agent models (Dunlop, 2010; Dunlop and James, 2007; Waterman and Meier, 1998), epistemic structures in international governance regimes (Dunlop, 2009; Haas, 1992) and models of bargaining and mutual adjustment (Elgström and Jönsson, 2000). At the micro level, researchers have worked on models of the mind, rationality and emotions, and how individuals get locked in by persuasion, majority opinions and other characteristics of group behaviour (Denzau and North, 1994).

Finally, international organisations are engaged in a debate with strong normative assumptions about how governments should learn. They have launched different instruments and architectures for cross-national and transnational learning, such as benchmarking, peer review, checklists and ‘facilitated coordination’ (Borrás and Radaelli, 2011). Such normative concerns are refracted by academic studies on lesson drawing (Rose, 1991), extrapolation (Barzelay, 2007) and benchmarking (Schäfer, 2006).

This intellectual activity is still very fluid (Radaelli, 2009). Recent reviews of the state of the art (Dobbin et al., 2007; Freeman, 2006; Grin and Loeber, 2007) contain only a handful of empirical studies, making it difficult to assess what we may collectively learn about this topic. This perhaps explains the general feeling of disappointment (Egan, 2009; James and Lodge, 2003; Volden et al., 2008). We still know little about how communities of policy makers learn in real-world settings (Freeman, 2006; 2007).

Thus, we need to step back a bit in order to sort out and systematise this intellectual activity and create the preconditions for cross-case research. Interestingly, both Dolowitz (2009) and Christina Boswell (2008) say we should focus on knowledge utilisation to overcome some of the problems. We shall follow up on this suggestion. After all, the fundamental issue at stake, in all the work on learning, concerns how knowledge is used and deployed by political actors to facilitate learning. By using the tools of concept formation and explanatory typologies (Elman, 2005; Sartori, 1970), we separate the individual from the species within the genus. The article is organised step by step. In the first section, we conceptualise the policy learning literature. We build four genera by combining two salient variables of contextual or macro-level variation – problem tractability and actor certification. Then, in the second section, we expand the property space within each genus by considering the fundamental dimension of control over knowledge production (see Dunlop, 2009). These dimensions refer to the actor-related part of learning. Such ‘expansion of the property space’ (Elman, 2005) leads us to sixteen species of learning, and enables us to show the relationships between species and the conditions for one genus of learning and another. We conclude with a discussion of the utility of the typological analysis in the third section and suggestions for future research in the conclusion.

Conceptualising Policy Learning: Four Learning Genera

  1. Top of page
  2. Abstract
  3. Conceptualising Policy Learning: Four Learning Genera
  4. Expanding the Property Space
  5. Discussion
  6. Conclusions
  7. About the Authors
  8. References

Learning has been examined in many different ways in public policy. By using different search options in the ISI social science citation index we identified some 833 articles from the main sub-fields of political science.2 After sifting these down to 86, we then added books and articles that are concerned with learning but do not show up in the social science citation index.

The first step in systematising the field is to plot the articles on an n-dimensional space, where the n-dimensions are theoretically justified. This is not possible for a large majority of the articles, which belong to the ‘null dimension’ of not having any theoretical depth. True, some sources of theoretical inspiration are mentioned in the introduction – it is for example almost expected to find ‘evocations’ of Peter Hall's three types of learning (Hall, 1993) or of Peter Haas' epistemic communities (Haas, 1992). But essentially the study remains empirical, and evidence is not used for theory testing or theory building. This is not surprising since the whole landscape of policy analysis is still best described by a few ‘mountain islands of theoretical structure, intermingled with, and occasionally attached together by foothills of shared methods and concepts, and empirical work, all of which is surrounded by oceans of descriptive work not attached to any mountain of theory’ (Schlager, 1997, p. 14).

What about the papers with some theoretical depth, then? It is obvious that not all dimensions are on the same level in the proverbial ladder of abstraction (Sartori, 1970). Some are abstract categories; others are more concrete specifications of these categories. Put differently, we have to classify per genus et differentiam in the tradition of Aristotle and (in political science) Giovanni Sartori (1970, p. 1036). We have to identify the genus or basic or primary category first (Collier and Mahon, 1993). Then we look at the variations (the differentia) that really matter in order to create the species or secondary categories. Once we understand this pattern, we can correctly classify two studies of the same genus in different species. This is equivalent to building an explanatory typology first, and then expanding the property space (Elman, 2005).

We argue that there are two dimensions with which to construct a mutually exclusive and jointly exhaustive explanatory typology of policy learning. These dimensions have been selected because they are prominent in the literature on social mechanisms and mechanisms of learning and theories of the policy process (Hedström, 2005; Liberatore, 1999; Sabatier and Jenkins-Smith, 1993).

The first concerns the level of uncertainty or problem tractability. There is a consistent body of literature in international relations, risk assessment and public policy that points to uncertainty as the main discriminatory factor between ‘thick’ and ‘thin’ learning, and between processes that can be handled with technical or technocratic approaches and ‘contested boundaries’ that become ultimately political (Checkel, 1998; Jasanoff, 1987). Think of epistemic communities (Haas, 1992). Without radical uncertainty, elected policy makers and their bureaucracies can calculate the pay-offs of different courses of action (Moravcsik, 1999; Radaelli, 1999). Here, the epistemic actors' role is less educative and more legitimating, as they reinforce policy makers' positions (Dunlop, 2010; Verdun, 1999).

To find the second, we observe that the genera vary in relation to the authority and legitimacy of some key actors or venues. The key question lies in mechanisms of actors' certification. Doug McAdam et al. (2001) have shown that this is a key mechanism of social action. If we ‘translate’ this key mechanism in the language of learning theory, the issue is whether there is a sort of ‘teacher’ that can be easily identified by the learners and enjoys some social legitimacy. To illustrate, teachers and narrators can be trusted because they have institutional roles or are endorsed by elected policy makers (Jones and McBeth, 2010). Or we can conjecture that institutional structures provide legitimacy and make a specific teacher heard – or withdraw legitimacy and silence those who speak from outside institutional fora. What was said, and how this is received by the ‘pupils’, depends on the institutional position of the teacher. Organisational roles and institutional rules and fora are the classic place to look for the mechanisms of an actor's certification. Of course, learning also takes place in ‘structureless’ environments. For example, the literature on reflexivity does not recognise any ex ante certification of types of actor – often major innovations come from tapping into the benefits of local knowledge, not ‘from the top’. We can also conjecture that there are pluralistic polyarchic settings where no-one is a pre-defined teacher and all actors are virtually able to play the role of learner. In these settings, learning has a relational quality (Fischer, 2003).

Taken together, the two dimensions provide our basic model of policy learning and four primary categories or genera (see Figure 1). The four genera identified are differentiated by the manner in which knowledge is used by actors for learning.


Figure 1. Mapping the Four Genera of the Policy Learning Literature

Download figure to PowerPoint

  • (1)
    In reflexive learning, knowledge is utilised with the aim of deepening discussion and facilitating argument. Learning in this cell is often characterised as ‘deep’ or ‘complex’ because it is the major mechanism through which actors not only adjust their strategies, but also explore their fundamental preferences and identities (Checkel, 1998; 2001). In turn, such explorations bring about the outcomes described by the literature on experimental governance (Lenoble and De Schutter, 2010; Sabel and Zeitlin, 2008), social and deliberative learning (Sanderson, 2002; 2009) and the discursive-argumentative turn in public policy that emphasises the transformative potential of framing (Daviter, 2007; Fischer, 2003; Schmidt, 2002). Most of this literature has normative assumptions or, at least, shares an interest in identifying the conditions for ‘good’ or ‘participatory’ governance. However, scholars working within the advocacy coalitions approach have pinned down the conditions for reflexive learning across coalitions without relying on normative assumptions (Sabatier and Weible, 2007).
  • (2)
    Let us now consider another of the most important lenses on learning, the epistemic communities tradition. In this genus, knowledge is deployed by a limited set of expert actors to narrow discussion with the aim of reaching a technical policy solution. Taken in its broadest sense, this lens covers experts in governments and international organisations, the political role of lawyers and economists in public policy and ‘palace wars’ (Delazay and Bryant, 2002) and the politics of knowledge utilisation carried out by social and natural scientists for agencies and departments (Boswell, 2008; Dunlop, 2007; 2009; 2010; Haas, 1992; Schrefler, 2010; Stolfi, 2010; Stone, 2005). Most of this literature revolves around the question of whether rationality, science and experts bring about change in public policy, and if so via what types of instrument, organisational setting or institutional device.3 There are different variations within this tradition, but there is no doubt that they have a common core grounded in the notion of communities of experts with shared causal policy beliefs and a paradigm of public policy. It follows that we can start the reconstruction of the field with a typology containing epistemic communities as one genus. We also have some expectations that there are species within this genus – as shown by the different declinations of the research on learning and epistemic communities (Dunlop and James, 2007; Lindquist, 1992; 1993; Radaelli, 1999; Stone et al., 1998; Youde, 2007).
  • (3)
    Third, we identify the genus of learning as a product of bargaining and social interaction. Three species in the literature fit this profile: interaction in loosely coupled organisations and political systems where learning is often the unintended product of dense systems of interaction between politicians and bureaucrats (Liberatore, 1999), strategic forms of learning (Cram, 1993; Dolowitz, 1997; Grossback et al., 2004) where knowledge is conceived of as a political resource and, arguably, learning ‘under conditionality requirements’ (Jacoby, 2001; Schimmelfennig and Sedelmeier, 2004).
  • (4)
    When reflexivity is constrained and bargaining limited by strong hierarchical mechanisms we exit the rather pluralistic world of reflexivity and bargaining. Here knowledge is utilised by actors or organisations to exert control. We enter a fourth genus characterised by learning in the shadow of hierarchy. There are several characterisations of the shadow (Börzel, 2010; Falkner, 2011; for the original approach see Scharpf, 1988; 1997). But most of these studies focus on the shadow of international organisations, domestic institutions and networks creating pressure to learn (Dobbin et al., 2007; Mahon and McBride, 2009; on coercion as a specific type of pressure, see DiMaggio and Powell, 1983). The shadow of hierarchy can alternatively be found in processes where learning is part of a delegation chain (Thatcher and Stone-Sweet, 2002; Waterman and Meier, 1998).

Expanding the Property Space

  1. Top of page
  2. Abstract
  3. Conceptualising Policy Learning: Four Learning Genera
  4. Expanding the Property Space
  5. Discussion
  6. Conclusions
  7. About the Authors
  8. References

Typological theory is particularly useful in qualitative studies, but its role is wider as it assists with concept formation, a step that is logically prior to measurement and therefore affects quantitative as well as qualitative studies (George and Bennett, 2005, ch. 11). There is also a strong connection between the classification steps across the ladder of abstraction and the notion of ‘expanding the property space’ in typological theory (Elman, 2005), although this connection has not yet been fully exploited. We use the expansion step to highlight the differentia among species. To be ‘explanatory’, however, a typological expansion has to be grounded in theory. We expand each cell using the findings of education studies of learning as control over knowledge production (Mocker and Spear, 1982; for a political science application of this adult learning typology, see Dunlop, 2009).

Donald Mocker and George Spear point to two dimensions: (1) control over the objectives of learning; and (2) control over the means and specific content of learning, that is, why one learns and what one learns. Mocker and Spear's typology of adult learning has four types. We use them consistently when we expand the property space. The four types are self-directed learning, informal learning, non-formal learning and formal learning (see Figure 2). Because their typology is constructed upon empirical reality, as well as theoretical assumptions of intended rationality, Mocker and Spear's descriptions of learning are well grounded and involve a sufficiently low degree of abstraction that captures specific learning dynamics whose occurrence is ‘objectively probable’ as opposed to the ‘objective possibility’ associated with single ideal types.4


Figure 2. Mocker and Spear's Lifelong Learning Typology

Source: Mocker and Spear, 1982 , p. 4.

Download figure to PowerPoint

This framework is transferable to the political world by dint of the assumptions that are made about the desire of individual actors – both learners and teachers – to control their environment. In their work, Mocker and Spear (1982) argue that, to understand the interactions between teachers and learners, we must view adult learners as intentional actors who choose to learn and aim to control the learning processes in which they become engaged. Just as decision makers’ control is usually mediated, it is not assumed that learners have it all their own way all of the time. Learning is heavily conditioned. While they are ‘intendedly rational’ (Simon, 1947), learners’ ability to control knowledge production and take ownership of their learning is mediated by variables that determine the extent to which their rationality is bounded: for example, their pre-existing ‘mental maps’ (Denzau and North, 1994), values and knowledge (Brown, 1995; Tolman, 1948) and perception of the socio-political and institutional ‘lifespaces’ they inhabit (Argyris and Schön, 1978; John-Steiner, 1997; Lave and Wenger, 1991; Lewin, 1951).

We must also explain what is meant here by ‘learners’ and ‘teachers’. In essence the defining feature of these actors concerns their role in any decision-making instance. So learners are those decision makers, policy makers and public organisations that hold the decision-making power at any moment in time. Teachers are those actors that seek to influence the decision-making process. Most obviously, teachers will be authoritative sources of knowledge – experts – but the possibilities go beyond those who can credibly claim expertise. Any actor, group or organisation engaged in producing or communicating information or knowledge has the potential to ‘teach’ decision makers about the issue at hand. Most commonly, these teachers are interest groups, think tanks or other public administrations or departments. The feature shared by these teaching actors is that they fuel and maintain debates around issues but do not have the final decision-making power on the matter at hand. Rather, they aim to influence the learners that do.

Self-directed learning is individualised and experiential. Here, learning is unstructured and driven by the learner. With their learning unrestricted by any disciplinary silos or paradigms and pre-determined goals, learners enjoy control over all aspects of learning, seeking out knowledge from a variety of sources, constructing the problem and establishing their own solutions in their own time; ‘learning what they want to learn, when they want to learn it and for as long as they want to learn it’ (Rogers, 2005, p. 253). In its most extreme form, self-directed learning can result in learners both adjudicating and creating evidence, rejecting the possibility that any expertise is superior to their own (Rogers, 2002, p. 275). More usually, knowledge creation here is not entirely autodidactic; notably, learners in the self-directed mode may take advice from a range of teachers on the veracity of the information they find (Hiemstra, 1994). They do not, however, identify with a single actor to inform the content and direction of policy. There are no single paradigms or knowledge hierarchies to structure what is learned by policy makers.

Informal learning treats learners as task-conscious. Here learning is not enlightenment for its own sake but rather revolves around assembling the means to dispatch a specific task that has been effectively set for them (Rogers, 2003, pp. 18–21). While the learner directs the selection and production of substantive resources, the presence of externally determined policy goals limits this scope for choice; the development of ‘know-how’ requires that learners are conscious of extrinsic evaluation where the substantive arguments they amass will be assessed in terms of goals that are determined by other actors.

Formal learning refers to externally imposed learning where the learner's control over both the substantive content of knowledge and the ends to which it is applied is severely constrained (Coombs and Ahmed, 1974). Learning here takes the form of guided episodes from teacher to learner where there is acceptance on both sides that learning needs to occur (Rogers, 2002, p. 279).

Non-formal learning refers to situations where information is moulded to learners’ own circumstances and the teacher's role is that of facilitator. Here, learners’ awareness of what they want to do with what they learn ensures that their engagement with codified knowledge is mediated by pre-existing expectations for determining the use or success of that knowledge (Heimlich, 1993; Tough, 1971). For non-formal learning to be identified, evidence is required of decision makers’ dependence on, for example, epistemic communities for the delivery, legitimisation or justification of policy preferences that have been formed independently of their relationship with these experts.

By using control over the contents of learning and control over the goals of learning – what is learned and why it is learned – we can then generate four cells out of each genus of the basic model of Figure 1. This is the step of ‘expanding the property space’ (Elman, 2005). By way of expansion, we create sixteen species out of the four genera. There is yet another benefit to the expansion of the property space. The expansion shows the variability within the four genera. Instead of treating reflexivity, epistemic communities, bargaining and ‘the shadow’ as monolithic, we explore the scope conditions for them to occur and the nuances possible within a single genus.

Reflexive Learning

We first expand the cells of the reflexive learning genus, using Figure 3. Recall that, for reflexive learning to occur, problem tractability has to be low and the certification of specific actors low. There is no pre-identified hierarchical role in terms of learning and no presupposition about who should learn from whom. Learning is the outcome of a social relation within a community of actors or a network (see Freeman, 2006, on the difference between the two). Institutions do not set hierarchical rules for the production and utilisation of knowledge.


Figure 3. Expanding the Property Space: Reflexivity

Download figure to PowerPoint

The distribution of power is polyarchic, and there has to be room for force-free deliberation, at least in the pure type discussed by the literature on reflective social governance. Preferences can change as a result of communicative rationality. There is no ex ante hypothesis about where the seeds of learning may germinate (lay knowledge is as important as professional and expert knowledge for reflexive dynamics to occur). The major problem is how to diffuse innovation across the community or network that is engaged with learning. This is why reflexive learning is often accompanied by governance architectures that facilitate the exploitation of innovation (Sabel and Zeitlin, 2008).

‘Deliberative’ is arguably the most pure or ideal-typical form of reflexivity (it is bold and underlined in Figure 3); learning is not deduction, but the outcome of a process of communication, persuasion and invention. The constellation of actors is, as it were, its own principal. It can pursue enlightenment for its own sake. It can reflexively modify preferences without exogenous constraints. Since preferences can change, the objectives of learning are dynamic and endogenous to the process of social interaction. In this specification, ‘policy’ is not something ‘finished’ out there that has to be learned by the constellation of actors (Freeman, 2006). Policy, instead, is finished and even produced in the act of learning.

Reflexivity is constrained in ‘experimental’ learning. There cannot be enlightenment for its own sake but rather it is bound up with the ideal of democratic polyarchies (Sabel and Zeitlin, 2008). Reflexivity is anchored to the task set for the community of learners – the task or end is exogenous. However, the constellation of actors has control over the means and content of what is being learned. We can think of experimental processes of trial and error with different forms of know-how and policy instruments, where Bayesian learning leads to the type of content that best suits the exogenous learning goal.

In ‘framing’ the learning experience is contingent on how the learners frame their problem. Since they have no control over the specific content of what is learned, the learning experience will operate through issue framing in the context of a given overarching goal (Daviter, 2007; Grin and van de Graaf, 1996). For those groups and actors whose evidence and information is learned by decision makers, their role can be one of the legitimation or justification of political objectives. Of course we should not assume that the evidential resources exist to deliver the favoured frame. Reflexivity can fail, and produce zero learning, if the mechanisms do not fit with the pre-existing objectives – put differently, there may be mismatch.

Although we did not identify it in our trawl of the literature on policy learning, a fourth conceptually possible species is highlighted by the expansion. Where learning is ‘evolutionary’, it takes place in loose issue networks where there are no pre-defined teachers, and the participants involved change over time. Following evolutionary theory, this mutation in the group of learners reflects the process of policy selection external to the network that they cannot control (see John, 2003, for an exposition of evolutionary accounts in the social sciences). The content of what is learned is a random process, triggered by exogenously determined learning objectives and policies. Thus, reflexivity is extremely limited in this species because neither the objectives nor content of learning are controlled by the constellation of actors. Can such networks survive? Although constrained, the continual mutation and turnover of learners ensures that some level of reflexivity is possible and preferences may be modified to avoid a doomsday scenario.

Epistemic Communities

Let us move to the species in the epistemic community genus in Figure 4. Recall that we need high actor certification and low problem tractability for this type to operate. By expanding the property space, we find the ‘expert as teacher’ category, which is our pure type (called ‘expert as teacher’, and bold and underlined in Figure 4) and three other species.


Figure 4. Expanding the Property Space: Epistemic Communities

Download figure to PowerPoint

Expert as ‘teacher’ is the purest form in that the literature on epistemic communities has implicitly identified this as the ideal-typical form of epistemic learning (Haas, 1992). Since epistemic actors are socially certified teachers, or there are characteristics of the venue that produce certification (such as mandatory consultation of experts before decisions are taken or delegation of power to expert committees), we have a clear teacher–learner relation. The epistemic community approach assumes the role of teacher and provides both the broad ends of the learning exercise and the substantive means (Dunlop, 2009, p. 292; Radaelli, 1999). In this way, epistemic communities reduce uncertainty and contribute to the definition of decision-makers' interests.

At the opposite end, we find the ‘contributor’ as the weakest species of epistemic learning. Here, epistemic communities may be treated as one teacher among many, ignored or their evidence contested by decision makers. In this specification, policy makers direct their own learning, choosing the interpretations and objectives that suit them (see Youde, 2007, on the rise of ‘counter epistemic communities’ in policy making on AIDS in South Africa). Empirical research suggests that the ‘epistemic community as contributor’ scenario has a temporal dimension. Decision makers' confidence in their capacity to adjudicate between knowledge claims and manage all aspects of their learning would only be high enough once they had learned from critical reflection of their own or others' experiences (see Dunlop, 2009, pp. 302–4). Thus, advisory relationships may start out with the epistemic community in the strong role of teacher only to have their influence diluted, and perhaps challenged over time, as decision makers' learning and awareness of alternative knowledge products and goals increases.

Next we have the ‘facilitator’. In this category, learners use their teachers to facilitate their process of learning. The epistemic actors are facilitators; they do not contribute to the definition of the interests of the learner. Rather, they are agents whose involvement is effectively controlled by their decision-maker principals (see Dunlop, 2010). The learner engages with the teacher and learns, but knowledge utilisation is subordinate to a context of pre-determined expectations set by the learner. Think of a pre-determined goal to use knowledge about public expenditure produced by a new monitoring system of public accounts to reduce the overall public budget for the next year. The role of facilitator may be more political, where an epistemic community is effectively crafted by decision makers to justify a pre-determined policy objective. Amy Verdun's (1999) account of the creation of the ‘Delors committee’ to advise the European Commission on the technical requirements of an accepted goal – Economic and Monetary Union (EMU) – is an example of how epistemic communities can facilitate policy action. We should be clear about why epistemic actors would become involved in such advisory relationships. Epistemic communities have socio-political beliefs, as well as substantive and methodological stances (Haas, 1992, p. 3); these actors are not politically neutral. The presence of these convictions ensures that the knowledge they produce is policy relevant and enables decision makers, who know their policy preferences, to select like-minded experts.

The last quadrant we examine in Figure 3 is the ‘producer of standards’. This has been empirically unexplored so far. Here, experts set policy targets or standards which decision makers must meet. Demonstrating informal learning within epistemic community–decision-maker exchanges requires evidence that decision makers recognise the goals anchoring their learning and are engaged in creating or gathering evidence of ‘what works’ in terms of delivering them (Dunlop, 2009, p. 297). The essence of the learning experience is about decision makers amassing the technical know-how that can meet the pre-existing exogenous standards or objectives set by experts. For example, governments frequently need to produce knowledge to satisfy standards set by international expert communities (for examples from international risk regulation, see Demortain, 2011).


Turning to bargaining (see Figure 5), this genus has high problem tractability and low or no actor certification. Bargaining produces learning as the unintended product of political competition and negotiations, as in Lindblom's partisan mutual adjustment. Yet again, we do not have neat separation between the teacher and the learner (low actors' certification). The tractability of the problem (low uncertainty) facilitates bargaining and strategic interaction since the pay-offs of different moves can be calculated by actors whose preferences do not change.


Figure 5. Expanding the Property Space: Bargaining

Download figure to PowerPoint

We mention that it is conventional to think of the Lindblomian notion of the ‘intelligence of democracy’ (Lindblom, 1965) as the ideal-typical manifestation of bargaining approaches to learning (though empirically we could find no studies where partisan mutual adjustment is the key mechanism of learning). This underlines that what is conceptually possible in the typology may not be empirically probable.

Moving to the next species –‘conditionality’– we find a degree of coercion to meet certain objectives of learning – which are determined exogenously. Actors bargain to meet standards and targets and, in doing so, they informally encounter learning in the shape of know-how and possibly discover smarter policy-making procedures or instruments. One way this may happen is when actors are pressured by conditionality requirements – the literature has shown that ‘smart’ and ‘clever’ solutions are facilitated by the imperatives of conditionality (Jacoby, 2001; see Weyland, 2005, on the latitude that policy makers possess when facing external pressure). Learning under conditionality can also help actors to get smarter in another way, that is, to get around the formal conditionality requirements and ‘use’ standards and targets to their benefit.

The next cell is ‘loosely coupled’ learning. There are no pre-defined teachers and learners – the model is contingent on a pluralist vision of the policy process where ‘n’ actors interact with given preferences. The actors do not control the objectives of learning. Neither do they control the contents and means of learning. They interact and encounter learning in a very unstructured or loosely coupled format. It is difficult to predict how learning can occur, or when it takes place. Here, instances of failed learning (failure to use bargaining to learn) will be common, as well as episodes of unintended learning (Liberatore, 1999).

The fourth bargaining species involves a ‘strategic’ approach to learning. The constellation of actors does not exercise control over the means of learning, thus there will be a certain degree of improvisation in this learning process. Bargaining can show how to acquire know-how, or it can produce failed learning. This, however, does not mean that the trajectory of learning is confused. The actors involved in the process can set their own goals – they are strategic about where they want to go with their learning process (see Cram, 1993, on the ‘purposive opportunism’ of the European Commission).

The ‘Shadow’

The ‘shadow’ is our last genus to be expanded into four cells (Figure 6). High problem tractability (low uncertainty) and high certification of actors and/or venues characterise the shadow. As mentioned, the original intuition was linked to the problem-solving bottlenecks of German federalism and the European Community of the 1970s, as examined by Fritz Scharpf (1988). The idea is that interaction in complex systems with multiple veto players creates ‘joint-decision trap’ (JDT) blockages. The causal argumentation of the JDT is a straightforward application of Coasian thinking about transaction costs. Sub-optimal decision-making outcomes are caused by the direct representation of lower units (such as states in a federal government) into key federal decisions (or member states in the decision-making process of the European Union). Direct representation is the first condition for the JDT. Unanimity de facto or de jure makes it impossible to override the concentrated interests of individual units that object to the decision – this is the second element or condition. The third condition is that participation of the individual units is compulsory. However, the presence of default conditions (that is, what happens if agreement is not found) generates agreement under certain conditions. To illustrate, within a given system of interaction, the actors may learn that their default condition is the intervention of a hierarchical, irreversible decision by a court (Schmidt, 2000). This kind of Damocles' sword alters the attitude of governments in the EU Council to the point of overcoming joint-decision traps, especially if the threat of the court is brought about by purposeful bureaucratic actors like the European Commission (Schmidt, 2000).


Figure 6. Expanding the Property Space: The Shadow

Download figure to PowerPoint

The ‘shadow’ is not necessarily produced by a court. We can think more generally of political institutions that, when present, can either provide ‘rules of the game’ (in the sense of the rational-choice version of the shadow articulated by Scharpf) or codes, identities, collective memories, lock-in mechanisms and roles that also alter the default condition, but through the logic of appropriateness of organisational sociology. Although research on the JDT is mostly based on rational-choice assumptions (see the different variations in Falkner, 2011), conceptually there is also room for a sociological version. The key question is what happens to learners and teachers under the shadow? There is a kind of teacher–learner relationship but this has less to do with specific actors (such as the experts, the scientists, etc.) and more with the content of institutional rules. Institutions ‘teach’ roles via socialisation and/or are channels through which rules are taught to the actors in the system of interaction.

The expansion of the shadow requires imagination: indeed, two of the four ‘shadows’ are conceptually possible but, as far as we can tell from our sample of articles, not empirically demonstrable. Let us start with a species we have found. Where learning is ‘instrumental’ it is focused on substantive policy improvement (Gilardi, 2010). Here, learners operate within exogenous objectives of learning, but they do control the means and content of learning. The shadow provides clues on what to do with learning. It structures interaction among learners. There is pressure to learn (in turn, pressure can come from the organisation, political competition in government, elections, and so on) but the logic of discovery is relatively free in relation to the means and content. This freedom should facilitate improvement at the level of the tools of government and the instruments of public policy.

The second species we found is that of ‘delegation’. Here, learners set their learning objectives autonomously, but are constrained as to the contents and means of learning. Here the shadow can be a resource used to organise the production and dissemination of knowledge. The vast literature on principal–agent models has shown different ways in which knowledge production, utilisation and diffusion can be organised along a chain of delegation of tasks. The learners design the tasks by remaining in control of the objectives of learning (Waterman and Meier, 1998).

And now, to the dark sides of this genus. We propose the idea of ‘hetero-directed’ learning as the ideal-typical or classic way of thinking about the shadow (bottom-right quadrant in Figure 6) since it typifies the essence of steering. Learners are entirely driven by the shadow since they cannot control either the means of learning or the objectives. This type of severely bound learning takes place in highly institutionalised environments. Learning is reduced to coping with instructions: a type of learning by rote about procedures, roles and memorising of doing things.

‘Autonomy’ is almost antithetical to the essence of hierarchy represented by the ‘hetero-directed’ cell. Indeed, ‘autonomy’ is the weakest specification ever possible of any shadow. True, we assume that the learners still operate within the shadow – think of domestic policy makers implementing EU directives in the shadow of Community law. But they can decide autonomously what they want to learn and how. To continue with the example: they can try to learn how not to gold-plate directives and which regulatory instruments to use in implementation. An ‘escape from hierarchy’ scenario may be most possible where the structure of authority is spurious.5 Learners' autonomy can increase in instances where authority is fragmented – for example in fragile coalition governments – or where a proliferation of authorities and agencies enables actors to play one ‘boss’ against the other – for example, in crisis coordination efforts. The essence of learning here is about carving out spaces for experimentation and genuine discovery.


  1. Top of page
  2. Abstract
  3. Conceptualising Policy Learning: Four Learning Genera
  4. Expanding the Property Space
  5. Discussion
  6. Conclusions
  7. About the Authors
  8. References

What then do we gain from the typological exercise? What does it really ‘show’? How can it be used? To begin with, let us think of the typology in the same way as we approach hypertexts on the internet.6 At the most abstract level, we have four concepts that typify how political scientists have approached the study of learning: they are reflexivity, the epistemic school, the bargaining-pluralist approach and institutional or rule-based modes of analysis. This is a broad map that sheds light on how the various schools of thought approach learning from radically different assumptions concerning preference formation, hierarchy and the ontological question of whether learning is an asymmetric relation between teachers of pupils or a highly interactive process. Further, it is obvious that while some look at macro processes involving negotiations among states or broad trends in policy convergence, others deal with micro processes, organisations and communities of policy makers (Zito and Schout, 2009b, p. 1103).

Thus, at this abstract level, the map can be used to generate meta-theories of policy learning. Meta-theory is concerned with the sociology of knowledge, the ontological dimension of theories and the core epistemological propositions. Applied to policy learning, a meta-theoretical approach exposes the different ontological assumptions and the core epistemological beliefs of different social scientists engaged with learning in public policy. It shows why and how different areas of the discipline are genuinely after ‘different things’ because of their understandings of the social and political worlds. Within the higher-level map, propositions such as ‘learning contrasts with rational policy theories in which optimal policy conclusions are derived from static analysis’ (Zito and Schout, 2009b, p. 1104) can be questioned, since we can produce rational-choice learning theories derived from an ontology that is social (meaning that the policy objects are socially constructed via meanings attached to events and problems) and an epistemology that is objective, but accounts for information that is costly.

To proceed with the metaphor of the hypertext, we can then metaphorically ‘click’ on any of the four concepts and find more detail when we expand the property space. Here our major finding is to show bias in what has been done until now. Most of the research has gone in some cells but others have been neglected. There are areas where it would be instructive to drill more, such as the cells of the ‘producer of standards’ and ‘evolutionary patterns of learning’. We should guard against investing too much time in some of these gaps in analysis, however. The intriguing but odd cells of ‘intelligence of democracy’, ‘autonomy’ and ‘hetero-directed’ learning may represent analytic culs-de-sac. Recalling Patrick Dunleavy's warning about the quest for the ‘gap-filling’ PhD thesis (Dunleavy, 2003, p. 21), uninhabited regions most probably exist for good reasons – they are too difficult to study or are of little interest in the first place.7

At the opposite end of the spectrum, we find cells such as ‘delegation’ and ‘deliberative’ learning that are never short of scholarly attention. In brief, the typological exercise shows where high-risk but potentially rewarding investment of research time and energy should be made. It also shows how some fields in the discipline of learning navigate pretty well across the cells. Arguably the best example is the literature on policy diffusion, which, depending on the emphasis on herding, pressure or strategic behaviour, can be classified in one cell or another. For diffusion scholars, our exercise is relevant because it makes clear how different studies have refracted the phenomenon of diffusion by using alternative conceptualisations of learning, and ultimately have generated different results.


  1. Top of page
  2. Abstract
  3. Conceptualising Policy Learning: Four Learning Genera
  4. Expanding the Property Space
  5. Discussion
  6. Conclusions
  7. About the Authors
  8. References

There is no point in denying that we have ‘learned about learning’ in public policy analysis. But to show exactly what we have learned is difficult because a massive conceptual stretching has occurred, with the resulting lack of communication between different strands of the discipline and loss of analytical and theoretical leverage.

We have therefore engaged in an attempt to reconstruct the field, by presenting a basic typology and suggesting how the primary categories can be subdivided into secondary categories. We have found it useful to blend Sartori's original intuitions about classification with more recent work on explanatory typologies. By doing that, we have demonstrated that concepts such as ‘reflexive learning’ and ‘epistemic communities’ are not monolithic: they can be disaggregated to make them amenable to fine-grained empirical analysis. Classificatory analysis provides maps and toolboxes, and clarifies concepts that are often confused. Our argument throughout the article has been that scholars working on learning tend to talk past each other because they do not realise the difference between genus and species; because they do not see the variables that connect one type to others; and because they measure learning with different types of bias (Radaelli, 2009). The purpose of our classification is to lay the foundations for systematic comparison across and within cases.

Another result was to make explicit the theoretical assumptions we make when we move from one genus to another. Our preference is for theories of knowledge utilisation, and for a blend of politics and sociology on one hand, and the education literature on the other. We have said that the choice of looking at knowledge utilisation is an asset, but for other researchers this might be a liability since it narrows down learning to specific processes involving knowledge.

Future research could usefully go in three directions: conceptual-theoretical, empirical and normative.8 Theoretically, there is the option of shedding more light on the ontological and epistemological assumptions of different approaches, and showing if and how these assumptions reverberate in methodological choices. We have hinted at meta-theory, but not pursued this option in this article (see Jupille, 2006, for an excellent application of meta-theory to the field of European studies). Another theoretical path is to connect our types to the results achieved by the so-called cognitive-psychological approach – Kurt Weyland (2005; 2008) has shown how learning theories benefit from the encounter with this approach. Conceptually, one may also question our choice to reconstitute policy learning by using knowledge utilisation. Certainly, there are other frameworks to map learning. Our choice has limits: by looking at who learns (teachers or pupils), where (in arenas where preferences change or do not change, for example), how (via the shadow of hierarchy or experimentally) and with what mechanisms, we have said little about content of learning, that is, the distinction between thin and thick learning. Thus, there is plenty of room for theory development. Equally considerable is the potential for theory testing: now that we have a classification built around explanatory variables, we can test better than in the past.

Empirically, explanatory typologies allow for systematic cross-case research. As we said, we have ‘learned about learning’ but individual empirical studies have not cumulated findings with previous work. With detailed empirical studies sensitive to the temporal dimension, we could also show how a constellation of actors and policy problems moves from one cell to another, and via what types of mechanism and strategy. This sort of travelling across the cells should lead to more robust results, given that the different steps on the ladder of abstraction (between species and genera) have been now clarified.

Finally, there are possible normative explorations of our types. Given a certain policy stalemate, where actors are stuck in a sub-optimal position, learning is certainly a way out. Learning, however, is often boundedly rational, and distorted by heuristics. For this reason, real-world learning falls short of the expectations of policy makers. Another way of looking at learning fiascos and making policy recommendations is to argue that policy makers are not searching for the appropriate type of learning. But exactly what type of learning does a given constellation of actors need? What is the most appropriate type given a certain level of tractability and actors' certification? When should we recommend a particular type? What are the preconditions that make a learning type particularly useful for problem solving, and which if violated would make it pointless to search for that particular learning process? Policy designers may use our explanatory typologies to try to move a given constellation from ‘experimental’ to ‘evolutionary’, or insert doses of ‘shadow of hierarchy’ to obtain certain outcomes.9 Research on learning ought to be connected to the applied-normative category of problem-solving approaches in public policy and public management, thus establishing a link between policy analysis and the so-called design sciences.

About the Authors

  1. Top of page
  2. Abstract
  3. Conceptualising Policy Learning: Four Learning Genera
  4. Expanding the Property Space
  5. Discussion
  6. Conclusions
  7. About the Authors
  8. References

Claire A. Dunlop is Senior Lecturer in Political Science in the Department of Politics at the University of Exeter. Her research interests centre upon the politics of expertise, epistemic communities, risk governance and the science–policy interface. Most recently, she has published research articles in Regulation and Governance, Journal of European Public Policy, Policy Studies, Policy Sciences and Science and Public Policy. Claire A. Dunlop, Department of Politics, University of Exeter, Amory Building, Rennes Drive, Exeter, Devon EX4 4RJ; email:

Claudio M. Radaelli is Professor of Political Science at the University of Exeter, where he directs the Centre for European Governance. He is co-editor of the European Journal of Political Research. He is co-editor of Research Design in European Studies: Establishing Causality in Europeanization Research (Palgrave, 2012) and co-author of Research Design in the Social Sciences (Sage, 2012). Claudio M. Radaelli, Department of Politics, University of Exeter, Amory Building, Rennes Drive, Exeter, Devon EX4 4RJ; email:

  • This article is based on research carried out with the support of the European Research Council grant on Analysis of Learning in Regulatory Governance, ALREG, directed by Claudio Radaelli ( The authors wish to express their gratitude to the Norwegian Political Science Association Annual Conference, 6 January 2010, University of Agder, Kristiansand; participants of the ‘Establishing Causality in Policy Learning’ panel at the American Political Science Association (APSA) annual meeting, 2–5 September 2010, Washington DC; and the European Consortium of Political Research (ECPR) Joint Sessions, St Gallen, 12–17 April 2011, workshop 2. In addition, we wish to thank Michael Barzelay, Peter Biegelbauer, Christina Boswell, Oliver Fritsch, Fabrizio Gilardi, Johan Olsen and Jeremy Rayner for their helpful comments on earlier drafts and our anonymous referees for insightful comments.

  • 1

    We would like to thank one of our referees for pointing this out to us.

  • 2

    We searched in the ISI Social Science Citation index on 16 May 2011 with the following criteria: Topic = (policy learning) OR Topic = (organisational learning) OR Topic = (social learning) AND TOPIC = (public policy) AND Topic = (learning). We then refined by subject areas: International Relations OR Public Administration OR Political Science OR Law OR Sociology OR Urban Studies. This produced an initial sample of 833 articles. Seven hundred and thirty articles were rejected from this sample through a search for duplicates (26), low citation counts (5 and under, 430) and two abstract sifts (270 and 31 articles were rejected, respectively). The main reasons for the rejection of non-duplicated papers were papers that focused on learning between non-state actors or actors outwith the political sphere; those which used the term learning in a purely descriptive way; those for whom ‘more’ or ‘better’ learning was a prescriptive punchline; and, finally, those studies which reviewed the literature without attempting to conceptualise or systematise it. Of the remaining 103 articles a further 8 were rejected and 9 were not available in their full form. This left us with a sample of 86 articles that could be explored.

  • 3

    The advocacy coalition scholars (Sabatier and Weible, 2007) refer to some of the core propositions on this literature. Yet they examine experts and scientists within coalitions and processes of knowledge utilisation.

  • 4

    See McKinney (1966, ch. 2) for a comparison of ideal types and constructed typologies, and Smith (2002) for a comparison of empirically-based and conceptually-inspired typologies.

  • 5

    We would like to thank one of our referees for pointing this out to us.

  • 6

    Hypertext uses hyperlinks to other text and web pages that can be accessed by the click of a mouse.

  • 7

    We would like to thank one of our referees for pointing this out to us.

  • 8

    We would like to thank one of our referees for pointing this out to us.

  • 9

    We thank Jeremy Rayner for making this point to us.


  1. Top of page
  2. Abstract
  3. Conceptualising Policy Learning: Four Learning Genera
  4. Expanding the Property Space
  5. Discussion
  6. Conclusions
  7. About the Authors
  8. References
  • Argyris, C. and Schön, D. (1978) Organizational Learning: A Theory of Action Perspective. Reading MA: Addison-Wesley.
  • Barzelay, M. (2007) ‘Learning from Second-Hand Experience: Methodology for Extrapolation-Oriented Case Research’, Governance, 20 (3), 52143.
  • Bennett, C. J. and Howlett, M. (1992) ‘The Lessons of Learning: Reconciling Theories of Policy Learning and Policy Change’, Policy Sciences, 25 (3), 27594.
  • Borrás, S. and Radaelli, C. M. (2011) ‘The Politics of Governance Architectures: Creation, Change and Effects of the EU Lisbon Strategy’, Journal of European Public Policy, 18 (4), 46384.
  • Börzel, T. (2010) ‘European Governance: Negotiation and Competition in the Shadow of Hierarchy’, Journal of Common Market Studies, 48 (2), 191219.
  • Boswell, C. (2008) ‘The Political Functions of Expert Knowledge: Knowledge and Legitimation in European Union Immigration Policy’, Journal of European Public Policy, 15 (4), 47188.
  • Brown, G. (1995) ‘What is Involved in Learning?’, in C. Desforges (ed.), An Introduction to Teaching: Psychological Perspectives. Oxford: Blackwell, pp. 1133.
  • Checkel, J. T. (1998) ‘The Constructivist Turn in International Relations Theory’, World Politics, 50 (2), 32448.
  • Checkel, J. T. (2001) ‘Why Comply? Social Learning and European Identity Change’, International Organization, 55 (3), 55388.
  • Collier, D. and Mahon, J. E. Jr (1993) ‘Conceptual “Stretching” Revisited: Adapting Categories in Comparative Analysis’, American Political Science Review, 87 (4), 84555.
  • Collier, D., Laporte, J. and Seawright, J. (2012) ‘Putting Typologies to Work: Concept-Formation, Measurement and Analytic Rigor’, Political Research Quarterly, 65 (2), 21732.
  • Coombs, P. H. and Ahmed, M. (1974) Attacking Rural Poverty. Baltimore MD: John Hopkins University Press.
  • Cram, L. (1993) ‘Calling the Tune without Paying the Piper? Social Policy Regulation: The Role of the Commission in European Community Social Policy’, Policy and Politics, 21 (1), 13546.
  • Daviter, F. (2007) ‘Policy Framing in the European Union’, Journal of European Public Policy, 14 (4), 65466.
  • Delazay, Y. and Bryant, G. G. (2002) The Internationalization of Palace Wars: Lawyers, Economists, and the Contest to Transform Latin American States. Chicago IL: University of Chicago Press.
  • Demortain, D. (2011) Scientists and the Regulation of Risk. Cheltenham: Edward Elgar.
  • Denzau, A. D. and North, D. C. (1994) ‘Shared Mental Models: Ideologies and Institutions’, Kyklos, 47 (1), 331.
  • Deutsch, K. W. (1966) The Nerves of Government. New York: The Free Press.
  • DiMaggio, P. J. and Powell, W. W. (1983) ‘The Iron Cage Revisited: Institutional Isomorphism and Collective Rationality in Organizational Fields’, American Sociological Review, 48 (2), 14760.
  • Dobbin, F., Simmons, B. A. and Garrett, G. (2007) ‘The Global Diffusion of Public Policies: Social Construction, Coercion, Competition, or Learning?’, Annual Review of Sociology, 33, 44972.
  • Dolowitz, D. (1997) ‘British Employment Policy in the 1980s: Learning from the American Experience’, Governance, 10 (1), 2342.
  • Dolowitz, D. (2009) ‘Learning by Observing’, Policy and Politics, 37 (3), 31734.
  • Dolowitz, D. and Marsh, D. (1996) ‘Who Learns What from Whom: A Review of the Policy Transfer Literature’, Political Studies, 44 (2), 34357.
  • Dunleavy, P. (2003) Authoring a PhD. Basingstoke: Palgrave.
  • Dunlop, C. A. (2007) ‘Up and Down the Pecking Order, What Matters and When in Issue Definition: The Case of rbST in the EU’, Journal of European Public Policy, 14 (1), 3958.
  • Dunlop, C. A. (2009) ‘Policy Transfer as Learning: Capturing Variation in What Decision-Makers Learn from Epistemic Communities’, Policy Studies, 30 (3), 291313.
  • Dunlop, C. A. (2010) ‘Epistemic Communities and Two Goals of Delegation: Hormone Growth Promoters in the European Union’, Science and Public Policy, 37 (3), 20517.
  • Dunlop, C. A. and James, O. (2007) ‘Principal–Agent Modelling and Learning: The European Commission, Experts and Agricultural Hormone Growth Promoters’, Public Policy and Administration, 22 (4), 40322.
  • Egan, M. (2009) ‘Governance and Learning in the Post-Maastricht Era?’, Journal of European Public Policy, 16 (8), 124453.
  • Eising, R. (2002) ‘Policy Learning in Embedded Negotiations: Explaining EU Electricity Liberalization’, International Organization, 56 (1), 85120.
  • Elgström, O. and Jönsson, C. (2000) ‘Negotiation in the European Union: Bargaining or Problem-Solving?’, Journal of European Public Policy, 7 (5), 684704.
  • Elman, C. (2005) ‘Explanatory Typologies in Qualitative Studies of International Politics’, International Organization, 59 (2), 293326.
  • Evans, M. (2009) ‘Policy Transfer in Critical Perspective’, Policy Studies, 30 (3), 24368.
  • Falkner, G. (ed.) (2011) The EU's Decision Traps: Comparing Policies. Oxford: Oxford University Press.
  • Fischer, F. (2003) Reframing Public Policy: Discursive Politics and Deliberative Practices. Oxford: Oxford University Press.
  • Freeman, R. (2006) ‘Learning in Public Policy’, in M. Moran, M. Rein and R. E. Goodin (eds), Oxford Handbook of Public Policy. Oxford: Oxford University Press, pp. 36787.
  • Freeman, R. (2007) ‘Epistemological Bricolage’, Administration and Society, 39 (4), 47696.
  • George, A. L. and Bennett, A. (2005) Case Studies and Theory Development in the Social Sciences. Cambridge: MIT Press.
  • Gerstenberg, O. and Sabel, C. F. (2002) ‘Directly-Deliberative Polyarchy: An Institutional Ideal for Europe?’, in C. Joerges and R. Dehousse (eds), Good Governance in Europe's Integrated Markets. Oxford: Oxford University Press, pp. 289341.
  • Gilardi, F. (2010) ‘Who Learns from What in Policy Diffusion Processes?’, American Journal of Political Science, 54 (3), 65066.
  • Graham, E., Shipan, C. and Volden, C. (2008) ‘The Diffusion of Policy Diffusion Research’. Paper presented at American Political Science Association annual meeting, Boston, 28 August.
  • Grin, J. and Loeber, A. (2007) ‘Theories of Policy Learning: Agency, Structure and Change’, in F. Fischer, G. J. Miller and M. S. Sidney (eds), Handbook of Public Policy Analysis. Boca Raton FL: CRC Press, pp. 20119.
  • Grin, J. and van de Graaf, H. (1996) ‘Implementation as Communicative Action: An Interpretive Understanding of Interactions between Policy Actors and Target Groups’, Policy Sciences, 29 (4), 291319.
  • Grossback, L. J., Nicholson-Crotty, S. and Peterson, D. A. M. (2004) ‘Ideology and Learning in Policy Diffusion’, American Politics Research, 32 (5), 52145.
  • Haas, P. M. (1992) ‘Introduction: Epistemic Communities and International Policy Coordination’, International Organization, 46 (1), 136.
  • Hall, P. A. (1993) ‘Policy Paradigms, Social Learning and the State: The Case of Economic Policymaking in Britain’, Comparative Politics, 25 (3), 27596.
  • Heclo, H. (1974) Modern Social Politics in Britain and Sweden. New Haven CT: Yale University Press.
  • Hedström, P. (2005) Dissecting the Social: On the Principles of Analytical Sociology. Cambridge: Cambridge University Press.
  • Heimlich, J. (ed.) (1993) Nonformal Environmental Education: Toward a Working Definition. Information Bulletin 502E. Columbus OH: ERIC.
  • Hiemstra, R. (1994) ‘Self-Directed Learning’, in T. Husén and T. N. Postlethwaite (eds), The International Encyclopedia of Education, Vol. 12, 2nd Edition. Oxford: Pergamon Press, pp. 539499.
  • Jacoby, W. (2001) ‘Tutors and Pupils: International Organizations, Central European Elites, and Western Models’, Governance, 14 (2), 169200.
  • James, O. and Lodge, M. (2003) ‘The Limitations of “Policy Transfer” and “Lesson Drawing” for Public Policy Research’, Political Studies, 1 (2), 17993.
  • Jasanoff, S. (1987) ‘Contested Boundaries in Policy-Relevant Science’, Social Studies of Science, 17 (2), 195230.
  • John, P. C. (2003) ‘Is There Life after Policy Streams, Advocacy Coalitions, and Punctuations: Using Evolutionary Theory to Explain Policy Change’, Policy Studies Journal, 31 (2), 48198.
  • John-Steiner, V. (1997) Notebooks of the Mind: Explorations of Thinking. New York: Oxford University Press.
  • Jones, M. D. and McBeth, M. K. (2010) ‘A Narrative Policy Framework: Clear Enough to be Wrong?’, Policy Studies Journal, 38 (2), 32953.
  • Jupille, J. (2006) ‘Knowing Europe: Metatheory and Methodology in European Union Studies’, in M. Cini and A. Bourne (eds), Palgrave Advances in European Union Studies. Basingstoke: Palgrave, pp. 20932.
  • Lave, J. and Wenger, E. (1991) Situated Learning. Cambridge: Cambridge University Press.
  • Lenoble, J. and De Schutter, O. (eds) (2010) Reflexive Governance: Redefining the Public Interest in a Pluralistic World. Oxford: Hart.
  • Levy, J. S. (1994) ‘Learning and Foreign Policy: Sweeping the Conceptual Minefield’, International Organization, 48 (2), 279312.
  • Lewin, K. (1951) Field Theory in Social Science. New York: Harper and Row.
  • Liberatore, A. (1999) The Management of Uncertainty: Learning from Chernobyl. Amsterdam: Gordon and Breach Publishers.
  • Lindblom, C. E. (1959) ‘The Science of Muddling Through’, Public Administration, 19 (2), 7888.
  • Lindblom, C. E. (1965) The Intelligence of Democracy. New York: The Free Press.
  • Lindquist, E. A. (1992) ‘Public Managers and Policy Communities: Learning to Meet New Challenges’, Canadian Public Administration, 35 (2), 12759.
  • Lindquist, E. A. (1993) ‘Think Tanks or Clubs? Assessing the Influence and Roles of Canadian Policy Institutes’, Canadian Public Administration, 36 (4), 54779.
  • McAdam, D., Tarrow, S. and Tilly, C. (2001) Dynamics of Contention. Cambridge: Cambridge University Press.
  • McKinney, J. C. (1966) Constructive Typology and Social Theory. New York: Meredith.
  • Mahon, R. and McBride, S. (2009) ‘Standardising and Disseminating Knowledge: The Role of the OECD in Global Governance’, European Political Science Review, 1 (1), 83101.
  • May, P. J. (1992) ‘Policy Learning and Failure’, Journal of Public Policy, 12 (4), 33154.
  • Meseguer, C. (2006) ‘Rational Learning and Bounded Learning in the Diffusion of Policy Innovations’, Rationality and Society, 18 (1), 3566.
  • Meseguer, C. (2008) Learning, Policy-Making and Market Reforms. Cambridge: Cambridge University Press.
  • Mocker, D. W. and Spear, G. E. (1982) Lifelong Learning: Formal, Nonformal, Informal, and Self-Directed. Information Series No. 241. Columbus OH: ERIC.
  • Moravcsik, A. (1999) ‘A New Statecraft? Supranational Entrepreneurs and International Cooperation’, International Organization, 53 (2), 267306.
  • Olsen, J. P. and Peters, G. B. (eds) (1996) Lessons from Experience: Experiential Learning in Administrative Reforms in Eight Democracies. Oslo: Scandinavian University Press.
  • Radaelli, C. M. (1999) ‘The Public Policy of the European Union: Whither Politics of Expertise?’, Journal of European Public Policy, 6 (5), 75774.
  • Radaelli, C. M. (2009) ‘Measuring Policy Learning: Regulatory Impact Assessment in Europe’, Journal of European Public Policy, 16 (8), 114564.
  • Rogers, A. (2002) Teaching Adults, third edition. Maidenhead: Open University Press.
  • Rogers, A. (2003) What is the Difference? Leicester: NIACE.
  • Rogers, A. (2005) Non-formal Education: Flexible Schooling or Participatory Education? Dortrecht: Kluwer.
  • Rose, R. (1991) ‘What is Lesson-Drawing?’, Journal of Public Policy, 11 (1), 330.
  • Sabatier, P. A. and Jenkins-Smith, H. C. (1993) Policy Change and Learning: An Advocacy Coalition Approach. Boulder CO: Westview Press.
  • Sabatier, P. A. and Weible, C. M. (2007) ‘The Advocacy Coalition Framework: Innovations and Clarifications’, in P. A. Sabatier (ed.), Theories of the Policy Process. Boulder CO: Westview Press, pp. 189220.
  • Sabel, C. and Zeitlin, J. (2008) ‘Learning from Difference: The New Architecture of Experimentalist Governance in the European Union’, European Law Journal, 14 (3), 271327.
  • Sanderson, I. (2002) ‘Evaluation, Policy Learning and Evidence-Based Policy Making’, Public Administration, 80 (1), 122.
  • Sanderson, I. (2009) ‘Intelligent Policy Making for a Complex World: Pragmatism, Evidence and Learning’, Political Studies, 57 (4), 699719.
  • Sartori, G. (1970) ‘Concept Misformation in Comparative Politics’, American Political Science Review, 64 (4), 103353.
  • Schäfer, A. (2006) ‘A New Form of Governance? Comparing the Open Method of Co-ordination to Multilateral Surveillance by the IMF and the OECD’, Journal of European Public Policy, 13 (1), 7088.
  • Scharpf, F. (1988) ‘The Joint Decision Trap’, Public Administration, 66 (3), 23978.
  • Scharpf, F. (1997) Games Real Actors Play: Actor-Centered Institutionalism in Policy Research. Boulder CO: Westview.
  • Schimmelfennig, F. and Sedelmeier, U. (2004) ‘Governance by Conditionality: EU Rule Transfer to the Candidate Countries of Central and Eastern Europe’, Journal of European Public Policy, 11 (4), 66179.
  • Schlager, E. (1997) ‘A Response to Kim Quaile Hill's In Search of Policy Theory, Policy Currents, 7 (2), 145.
  • Schmidt, S. K. (2000) ‘Only an Agenda Setter? The European Commission's Power over the Council of Ministers’, European Union Politics, 1 (1), 3761.
  • Schmidt, V. A. (2002) ‘Does Discourse Matter in the Politics of Welfare State Adjustment?’, Comparative Political Studies, 35 (2), 16893.
  • Schrefler, L. (2010) ‘The Usage of Scientific Knowledge by Independent Regulatory Agencies’, Governance, 23 (2), 30930.
  • Simon, H. (1947) Administrative Behavior: A Study of Decision-Making Processes in Administrative Organization. New York: The Free Press.
  • Simon, H. (1957) Models of Man. New York: Wiley.
  • Smith, K. B. (2002) ‘Typologies, Taxonomies, and the Benefits of Policy Classification’, Policy Studies Journal, 30 (3), 37995.
  • Stolfi, F. (2010) ‘Testing Structuralist and Interpretative Explanations of Policy Change: The Case of Italy's Budget Reform’, Governance, 23 (1), 10932.
  • Stone, D. (2005) ‘Knowledge Networks and Global Policy’, in D. Stone and S. Maxwell (eds), Global Knowledge Networks and International Development. London: Routledge, pp. 89105.
  • Stone, D., Garnett, M. and Denham, A. (eds) (1998) Think Tanks across the World: A Comparative Perspective. Manchester: Manchester University Press.
  • Thatcher, M. and Stone-Sweet, A. S. (2002) ‘Theory and Practice of Delegation to Non-majoritarian Institutions’, West European Politics, 25 (1), 122.
  • Tolman, E. (1948) ‘Cognitive Maps in Rats and Men’, Psychological Review, 55 (4), 189208.
  • Tough, A. (1971) The Adult's Learning Projects. Toronto: Ontario Institute for Studies in Education.
  • Verdun, A. (1999) ‘The Role of the Delors Committee and EMU’, Journal of European Public Policy, 6 (2), 30829.
  • Volden, C., Ting, M. M. and Carpenter, D. P. (2008) ‘A Formal Model of Learning and Policy Diffusion’, American Political Science Review, 102 (3), 31932.
  • Waterman, R. H. and Meier, K. J. (1998) ‘Principal–Agent Models: An Expansion?’, Journal of Public Administration Research and Theory, 8 (2), 173202.
  • Weyland, K. (2005) ‘Theories of Policy Diffusion: Lessons from Latin American Pension Reform’, World Politics, 57 (2), 26295.
  • Weyland, K. (2008) ‘Toward a New Theory of Institutional Change’, World Politics, 60 (1), 281314.
  • Youde, J. (2007) Aids, South Africa and the Politics of Knowledge. Aldershot: Ashgate.
  • Zito, A. R. and Schout, A. (eds) (2009a) ‘Learning and Governance in the EU Policy-Making Process’, Journal of European Public Policy, 16 (8), 1103253.
  • Zito, A. R. and Schout, A. (2009b) ‘Learning Theories Reconsidered: EU Integration Theories and Learning’, Journal of European Public Policy, 16 (8), 110323.