What happens on the ground: Persistent issues in implementation research

A ‘yesterday’s issue’?

As a broad term, ‘implementation’ may refer to anything meant to happen after an intention or aspiration has been expressed. The theoretical core of what is called implementation theory and research as a scholarly field in the narrow sense is usually addressed as a subdiscipline of Public Administration and Political Science. In this article, the object of that field is defined as examining what happens within governmental agencies as well as in their relationships with other actors while fulfilling public tasks; and asking how that does matter (for alternative definitions, see Barrett, 2004: 251; DeLeon, 1999: 314, 315; Ferman, 1990: 39; O’Toole, 2000: 266 and 273, also 2005; Pressman and Wildavsky, 1973 [1984]: xxi).

Implementation studies, in the indicated sense, have lost their fashionable character. In core journals on public administration, like Public Administration Review and the Journal of Public Administration Research and Theory, attention to the subject of implementation has considerably declined, as Moynihan (2008) has evidenced for the period 1998–2008. Implementation is ranked among the ‘traditional’ (p. 481) or ‘classic’ (p. 486) themes of public administration, like personnel management, red tape and political influence on bureaucracy. For the majority of academics, the study of implementation has become a ‘yesterday’s issue’ (for a critical analysis of that claim, see Hill, 1997). Nevertheless, the number of articles reporting on some kind of implementation research – often published in journals dedicated to specific policy sectors – is still substantial, as Sætren (2005) has shown.

It is not immediately obvious how to measure the development of implementation research as a scholarly field. Particularly, the operational question is how to formulate a ‘yard stick’ with which the present state of affairs can be assessed. As possibilities we see mapping a list of diagnostic observations drawn from existing reviews of the field; using a specific selection of issues made elsewhere (Hill and Hupe, 2014) or adopting a diagnosis-cum-agenda as formulated in a report of a specific implementation study (Goggin et al., 1990).

Measuring ‘progress’ in implementation research

Since Pressman and Wildavsky (1973) the development of the field has been reviewed by a range of scholars. Next to Sætren’s bibliometric study already mentioned, other overviews are, in chronological order, Yin (1982), Alexander (1985), Sabatier (1986), O’Toole (1986, 2000, 2004, 2005), Lester et al. (1987), Linder and Peters (1987), McLaughlin (1987), Goggin et al. (1990), Ingram (1990), Palumbo and Calista (1990), Matland (1995), Ryan (1996), Hill (1997), Lester and Goggin (1998), DeLeon (1999), Winter (1999, 2006, 2012), Lester (2000), Sinclair (2001), Schofield (2001), DeLeon and DeLeon (2002), Barrett (2004), Schofield and Sausman (2004), Pülzl and Treib (2006), Robichau and Lynn (2009), May (2012).

Several authors use straightforwardly dismissive qualifications when characterizing the state of the scholarly field: ‘(C)omplex, without much cumulation or convergence (..)’ (O’Toole, 1986: 181); ‘(y)et to reach conceptual clarity’ (Ingram, 1990: 462); ‘misery research’ (Rothstein, 1998: 62); ‘an intellectual dead end’ (DeLeon, 1999: 313); ‘we know surprisingly little’ (Sætren, 2005: 559). Winter states (2006: 163): ‘(N)o general implementation theory has emerged (…). The implementation subdiscipline has been characterized by many different approaches representing different research strategies, evaluation standards, methodologies, concepts, and focal areas for research’. (…) ‘The highly fragmented character of implementation research is not very conducive to theory accumulation’ (Winter, 2006: 164, referring to Sætren, 2005).

Some scholars, however, come to an assessment that is, at least, more mixed: ‘There is more than meets the eye’ (O’Toole, 2000: 265). He observes: ‘One has to look, sometimes, in unusual places and be informed by a broader logic of intellectual development to make sense of the relevant scholarship. (..). (S)ome of the discourse has shifted, the questions have been broadened, and the agenda has become complicated’ (O’Toole, 2000: 263). Meier (1999: 6, 7) comes to a similar conclusion. ‘My biased survey of literature suggests that a wide range of journals publish articles that inform the study of policy implementation – the mainstream sociology journals, most of the public administration journals, the professions journals (public health, social work, sometimes law or medicine), many of the economics journals, and on rare occasion a political science journal. Much of this literature is not intended to directly answer questions of policy implementation, but it addresses concerns that are central to policy implementation’. Having found substantial numbers of articles in journals devoted to education, health, law, economics and the environment, Sætren (2005: 562) concludes on the basis of an extensive bibliometric study of the field: ‘(T)he data overwhelmingly refute the decline proposition’ (italics from the original). Pülzl and Treib (2006) arrive at a similarly mixed judgement on the state of the field.

These overviews provide rather overall assessments than explorations of specific theoretical and methodological issues. An alternative would be to use the selection made in the most recent state of the art. In the third edition of their monograph, Hill and Hupe (2014, chapter 7) mention several issues researchers of implementation have to address: defining studies of implementation, explaining what needs explanation, isolating implementation as a part of the policy process, dealing with layers, specifying inter-organizational relationships, differentiating agency responses, identifying stakes, recognizing relevant macro factors and making appropriate decisions as far as quantitative and qualitative research designs are concerned. It is clear that when looking at how contemporary implementation research deal with these nine issues, a proper treatment of the whole range would go beyond the scope of this article.

A third option to measure the development of the field would be to adopt the diagnosis-cum-agenda put forward by Goggin et al. (1990), before applying it in an analysis of the present state of knowledge. It seems justified to do so, as we will, because their book combines a review of the implementation literature up to that date with a report of empirical research. The book is particularly relevant because they also make concrete suggestions for ‘third generation research’, entailing ‘conceptualizing, operationalizing, measuring and testing’ (Goggin et al., 1990: 205). Hence, Goggin et al. (1990) formulated a programmatic agenda for implementation research, which can be used to assess how the field has developed more than two decades later.

Four questions

By ‘joining a quest for better theory with a search for improved practices’, Goggin et al.‘(..) synthesize and assess two decades of theoretical and empirical research and chart the way for the next generation – a third generation – of research’ (1990: iv). Aiming ‘to make sense of the causal complexities of policy implementation’, they argue for the adoption of a ‘more “scientific” search for the patterned regularities as well as the idiosyncrasies of implementation decisions and actions’ (Goggin et al., 1990: 8). They identify some ‘major impediments’ to the achievement of a satisfactory explanatory theory: ‘too many variables, but too few cases, or “overdetermination”; a lack of attention to measures, and a missing orientation of ‘empirical research that is truly comparative, longitudinal, and synthetic’ (Goggin et al., 1990: 11). They elaborate these topics in subsequent paragraphs, headed as, respectively, the cases/variables problem, the lack of estimates, the neglect of the role of the states and ‘treating implementers as autonomous rational actors’ (Goggin et al., 1990: 10–13).

Goggin et al. (1990) observe that the nature of policy implementation invites case studies. At the same time, it is difficult ‘either to introduce the element of control for the effects of extraneous independent variables or to generalize from their findings’ (Goggin et al., 1990: 10). They refer to Goggin’s (1986) framing of this issue earlier as the ‘too many cases/too many variables problem in implementation research’. The question here is then: Have solutions been found to deal with the variety of potential explanatory variables?

Second, instead of ‘proverbs’ (O’Toole, 1986) and checklists, Goggin et al. (1990) identify a need for operational definitions of concepts. Precise measures of variables are needed, as well as specified expectations about relationships among the various relationships, Goggin et al. state. One could add that a prescriptive (normative) orientation needs to be replaced by, or at least distinguished from, a descriptive (empirical) one: How do researchers address their relationship with the object of their research?

Third, Goggin et al. (1990) observe that neither ‘top-down’ nor ‘bottom-up’ approaches to implementation research give sufficient attention to what they call the ‘levels of analysis’ issue. Therefore, in the research project reported on in their book, they use the United States as the primary unit of analysis. After all, the intergovernmental context of implementation influencing policy outputs includes federal and local, as well as state-level variables. The question then is: How do researchers handle the fact that policy implementation, almost by definition, implies the involvement of multiple actors on different layers?

Fourth, Goggin et al. (1990) recommend the treatment of implementers as ‘autonomous rational actors’. In fact, the political character of implementation is acknowledged in their work. Hence, the question here is: How do researchers deal with the consequences of the complex, non-technical nature of policy making?

Goggin et al. (1990: 15) summarize their programmatic agenda for the field as follows: ‘(T)he challenge for the next generation of scholarship is to develop and test explanatory and predictive implementation theories of the middle range’. They add that this ambition implies – as ‘the unique trait of third-generation research’ – a research design characterized by an adequate treatment of the four issues mentioned above. Before addressing to what extent this is the case (in section ‘Comparing approaches’), we first have to specify the present state of the study of implementation.

The policy-implementation paradigm and beyond

Adapting Kuhn’s (1970) definition of a ‘paradigm’, Hill and Hupe (2014) historically locate the ‘policy-implementation paradigm’ as prevalent in the 1970s and closely connected to large-scale policy programmes of interventionist central governments. They identify the following characteristics of studies representing that paradigm. First, there is the core assumption of implementation as a ‘later’ part in the policy process consisting of a range of ‘stages’. After the particular public policy has been formulated and decided upon, its implementation follows subsequently. A second characteristic is the assumption that once one knows the objectives of a policy, one knows what to expect about its results. The official documents in which the policy has been laid down are seen as setting out these objectives. A final characteristic is the orientation ‘from the top’, in the geographical sense; it is at the ministries or other places in the centres of authority where policies are being made.

Hill and Hupe (2014) stress that also before the publication of Pressman and Wildavsky’s (1973) classic, there were implementation studies, although under different headings. In his bibliometric study, Sætren (2005) underlines this observation. He particularly mentions the early evaluation studies in the mid-1960s of social reform programmes in the US as triggering the focus on policy implementation a decade later. In the same vein Sætren shows that substantial numbers of implementation studies are still reported on in journals on particular policy sectors, like education and health.

The sheer volume of studies that are, from one perspective or another, about implementation makes any effort to develop theory through bibliometric studies inapplicable. What can be done instead is to map the major streams of work in terms of categories. We do so below, distinguishing between ‘mainstream’, ‘neo-’ and ‘advanced’ implementation studies.

Categories of contemporary implementation research

Mainstream implementation studies. Studies representing the policy-implementation paradigm most of the time will have a design of a single case study, a usage of methods of qualitative research and a ‘downward’ theoretical approach. The latter means that intentions and achievements are compared, while the policy goals as formulated are used as a starting point. Next the compliance with those goals is being documented, before assessing deviance from them.

On the agenda of the journals with a policy studies, public administration or public management focus, direct attention to implementation has been replaced in prominence by studies on subjects like performance, public sector reform and e-government (Moynihan, 2008). Nevertheless, the interest in what happens with Great Expectations in the daily practice of public administration remains, sometimes driven by more or less obvious ‘policy failures’. It seems worthwhile to notice that also in much contemporary implementation research, a straightforwardly prescriptive stance is adopted; see, for instance, Johansson (2010). Most of the time studies in this category involve a case study of a single policy programme, offering a ‘thick description’ (Geertz, 1973), and adopting a ‘view from the top’. Staying within the policy-implementation paradigm – in fact, still representing it – this category justifiably can be called mainstream implementation studies.

Neo-implementation studies. Under the heading of ‘multi-level governance’, since the 1990s various studies concerning policy making in the European Union (EU) have appeared (Peters and Pierre, 2001). The term multi-level governance is used to refer to policy making across more than one administrative layer, but with the addition of an extra layer, the one of the EU. The fact that in most of such studies ‘the top’ in a particular policy process is not located in a national capital but in Brussels as the ‘capital’ of the EU leads some researchers to address the activities of EU-member states, national governments, as ‘implementation’. At the same time, at all ‘levels’ potentially a multiplicity of actors are active – inasmuch as EU-policy is likely to be in itself a product of multi-actor interaction. Among this multiplicity, non-state actors like companies and organizations of interests, may become involved in the implementation process. Pertinent here is the debate within the EU literature between those who see ‘implementation deficits’ and those who focus on inter-governmental relations (cf. for instance Héritier and Lehmkuhl, 2008; Treib, 2008). The research design in studies of this kind often concerns a particular policy as a single case, the methods used are mostly qualitative, while the theoretical approach is downward. Although under a different heading, that of ‘multi-level governance’, therefore actually the policy-implementation paradigm seems to prevail here. The combination of a contemporary locus looked at from a traditional focus makes Hill and Hupe (2014) speak of ‘neo-implementation studies’.

Advanced implementation studies. As noted above, also nowadays implementation is researched in varying disciplines, such as management studies (cf. Hickson et al., 2003), and from different angles. Among those angles are decision making (cf. Torenvlied, 2000), delegation (cf. Bendor et al., 2001), political control (cf. Meier and O’Toole, 2006) and (inter-) organizational conflict (cf. Oosterwaal and Torenvlied, 2009). In the SAGE Handbook of Public Administration edited by Peters and Pierre (2012), the section ‘Implementation’ has chapters expressing attention to the central subject from the focus of policy design, inter-organizational relations and street-level bureaucracy. In his overview of the field, Winter (2006) gives an exposition of ‘third generation’ implementation studies. What these studies have in common, and what makes them different from mainstream studies, is the elaborated character of their research design. When scientific rigour justifies calling these implementation studies advanced, a distinction can be made between studies in which a ‘downward’ or an ‘upward’ theoretical approach is being used.

The studies mentioned in the beginning of this paragraph adopt a downward approach, as do ones on policy design and mandating (for instance, May 1993, 2006). Although implementation studies of this kind have a sophisticated research design, their rationale is a two-layer comparison. They start from policy intentions formulated at the top, while confronting them – ‘downward looking’ – with goal achievement at the bottom. On the other hand, the studies of implementation looked at in terms of inter-organizational relations and street-level bureaucracy can be seen as embodying a different sub-category. They do have a research design going beyond a single case study (larger n), aiming at rigour also when qualitative research methods are used. These characteristics indicate therefore a research mode going beyond the policy-implementation paradigm, making them advanced. However, an ‘upward’ theoretical approach is employed (cf. Winter, 2006; UNICEF Innocenti Research Centre, 2012). This means the research objective is explaining empirical variation in respect of policy outputs or sometimes outcomes. Candidate explanatory factors are primarily sought at the micro- and meso-level of analysis, on the basis of ‘local’ theory formation. Rather than making a two-layer comparison, upward studies start by looking at implementation variation, while treating policy intentions as among a range of independent variables.

Distinguishing between these categories of contemporary implementation studies, with the third divided in a downward and an upward variant, can be justified by the fact of the joint characteristics shared within each category. Of course there will be outliers, but the studies of each sort have a certain degree of similarity, particularly when the characteristics of the policy-implementation paradigm are used as a point of reference. We go on from this then, to address how these various categories of implementation studies deal with the issues of implementation research identified in the previous section.

The problem of ‘too many variables’

How do researchers deal with the variety of potential explanatory variables? Johansson (2010) recommends both implementers and implementation researchers interested in a specific policy to start with an ‘in-depth analysis of the nature of the policy and the nature of the implementation context’ (122). To detect potentially problematic issues in the practice of the implementation of the policy, Johansson (2010) recommends seeking answers from the clusters of independent variables identified by Hill and Hupe (2009) as potentially explaining implementation results. Since Johansson’s (2010) object was to be helpful to policy makers these questions clearly imply a top-down perspective. This example is mentioned here because it shows the lasting relevance of the concern with goal achievement as characteristic of mainstream implementation studies. Rather than analytically redressing the number of potential causal variables, the selection of practically manageable factors is upfront here. Not so much seeking explanation of empirical variation but, in fact, establishing and possibly enhancing policy effectiveness is at stake. That the differences between implementation research and evaluation research hence may become confused is a price easily paid.

Putting together two or three case comparisons, neo-implementation studies are essentially case studies. A conceptual framework consisting of (more than) a few key variables is applied in the analysis, hence consolidating rather than tackling the problem of ‘too many variables’. In advanced implementation studies, this problem is handled by a deliberate focus on specifying a limited number of hypotheses. They differ with their mainstream counterpart in having a quantitative research design, often with a large n, while formal modelling may be featuring. Also in qualitative studies, the aim at scientific rigour implies a systematic conceptualization around a limited set of variables.

The theory/practice relationship

How do researchers address their relationship with the object of their research? Implementation research often, even currently, is characterized by an adoption of the goals of a public policy as legitimately decided upon and laid down in a formal document. As a consequence, the dependent variable in mainstream implementation studies tends to be goal achievement. Pressman and Wildavsky’s (1973) subtitle of their classic book catches what since then paradigmatically has become known as an ‘implementation gap’. If the observed implementation results in Oakland differ from the policy intentions as officially expressed in Washington, something on the way between them seems to have gone wrong. This type of approach has led to a lot of ‘misery research’ (Rothstein, 1998: 62).

Also in neo-implementation studies, the adoption of the policy goals by the researcher may be joined by an identification with what has been articulated ‘from the top’. When things go wrong – and it is clear they sometimes seriously do – the standard reaction implies a blaming of the implementers. Although in advanced implementation studies the stance of the researchers will remain an analytical one, the implication of studies in which a downward approach is pursued is similar. As the terms suggest, ‘deviant’ or ‘non-compliant’ behaviour is incorrect behaviour. However, Winter (2012) suggests a focus primarily on policy outputs, as the performance of implementers, thus recommending an upward approach. It is empirical variation in performance (horizontally) that needs explanation, rather than goal achievement (vertically). If there is variation in outputs within a sample of implementers (for example local governments), the explanation of that may lie in local activities, in ambiguity of the specification of objectives from the centre, or in other factors.

The multi-layer problem

How do researchers handle the fact that policy implementation, almost by definition, implies the involvement of multiple actors on different layers? Authors of mainstream implementation studies seem to deal with this fact each in their own way. Some authors acknowledge that policy goals may be deliberately not specified precisely, while extensive license is given to discretionary action. In the EU literature, for instance, a ‘hard law’/’soft law’ distinction is made and the term ‘framework legislation’ is used (cf. Knill and Lenschow, 2003; Treib, 2008). Implementation hence may be mandated in various ways. Multi-level governance studies and related literature particularly bring this out in the light of problems of enforcement and compliance by nation states, while encountering the ‘problem of the many layers’. The proposition can equally be applied to issues within nation states about the mandating of implementation towards officials on lower layers (for example, Torenvlied and Akkerman, 2004). In the latter and similar advanced implementation studies, the multi-layer problem usually is being addressed. Scholars like Torenvlied (2000) analyze policy implementation by documenting the congruence between the results of what happens on two layers, that of political decision making and the one of implementation. In studies with an upward approach, the locus of investigation remains limited to one layer, usually the ‘street-level’. Factors potentially explaining local variation may be sought in policy design, but not exclusively.

The policy/politics nexus

How do researchers deal with the consequences of the complex, non-technical nature of policy making? Within mainstream implementation studies, formal policy documents are used as point of reference. The policy goals have been formulated and decided upon, while their implementation is expected to follow accordingly. Hence the measure stick is fixed; the goals have to be achieved as, more or less concretely, prescribed in the policy-on-paper. In neo-implementation studies, the view is not fundamentally different. In fact, also in advanced implementation studies in which a downward approach is used congruence is being presupposed, as an implementation compliant with the stated policy goals (cf. Hupe, 2011). Actual action of implementers is looked at in corresponding terms, with ‘divergence’ or even ‘deviation’ as labels for what is seen as non-compliant behaviour.

In studies showing an upward approach, in contrast, the primary focus is on observed rather than presupposed behaviour. It is acknowledged that stated policies may be the result of social interaction between various actors, therefore may have an ambiguous character, and hence less function as a technical instruction. Important political intentions behind a policy often remain unwritten. While implementing public policies, street-level bureaucrats always have to act, often in unforeseen circumstances. Because there is an eye for these facts, in this kind of studies formal policy goals theoretically are treated as one input among a range of variables explaining actual policy performance.

Issues resolved?

At the end of this section now the question can be answered, in what way the distinguished categories of implementation studies deal with the four problems Goggin et al. (1990) deemed pertinent for the development of the scholarly field. To what extent have the four issues been resolved?

As Table 1 shows, the answer is mixed. In advanced implementation studies, explicit steps towards the formation of low middle-range theories have been taken. Therefore, the problem of ‘too many variables’ can be considered to have been resolved. This is to say, ways have been invented to deal with this issue – even if they are not equally being adopted in all categories of studies.

OPEN IN VIEWER

Table 1.

Modes of dealing with issues of implementation research

Category of studies	Too many variables	Theory/practice relationship	Multi-layer problem	Policy/politics nexus
Mainstream	+/−	−	−	−
Neo	+/−	−	+/−	+/−
AdvancedD	++	+/−	+/−	−
AdvancedU	++	+	+	+

The theory/practice relationship concerns what Goggin et al. (1990) addressed in terms of a lack of a clinical view. Neo-implementation studies share with mainstream implementation studies an acceptance of policy goals. The latter are used as a standard implying formal ‘intentions’ actual ‘achievements’ are to be compared with. Advanced implementation studies with an upward approach are the only ones in which an a priori or implicit adoption of policy goals is absent.

In mainstream implementation studies the ‘problem of the many layers’ is not really addressed. Rather more or less explicitly formal hierarchical relations are considered as prevailing. In fact, similar observations can be made about multi-level governance studies. Also in advanced implementation studies with a downward approach, usually two layers are being looked at.

As far as the policy/politics nexus is concerned, only in advanced implementation studies with an upward approach the possibility of legitimate action beyond formal mandates is being acknowledged and analyzed accordingly.

The problem of ‘too many variables’

The problem concerns one of supposed causality. In the public domain, numerous factors can be expected to influence the behaviour of actors, which are hard to isolate. Bowen (1982) investigated the implications of the chain metaphor used by Pressman and Wildavsky (1973). Winter (2006: 157) observed that the research of the first and second generations of implementation studies ‘had not succeeded in sorting out the relative importance of the explanatory variables. A substantial part of the studies could be criticized as merely presenting – often long – checklists of variables that might affect implementation’. Speaking about ‘forty-seven variables that completely explain five case studies’ Meier (1999: 5) suggested: ‘Any policy implementation scholar who adds a new variable or a new interaction should be required to eliminate two existing variables’ (6). ‘A literature with three hundred critical variables doesn’t need more variables; it needs structure’ (Matland, 1995: 146).

In their attempt to formulate a programmatic agenda for a ‘third generation’ of implementation studies, Goggin et al. (1990) pleaded for a ‘synthesis’ approach. With its characteristics as outlined in section ‘Characterizing a ‘synthesis’ approach’ of this article, it was meant to counter an ongoing promotion of isolated factors. Rigour and getting away from a predominantly normative orientation were important objectives. One of Goggin’s co-authors later acknowledged that their level of aspiration was rather unrealistic (O’Toole, 2000: 272).

In dealing with the problem of ‘too many variables’, advanced implementation studies show that it is possible to formulate theoretical arguments in ways which allow them to be empirically tested. While qualitative studies can handle greater complexity, quantitative studies offer more scope for comparison and generalization. Since the latter are often data-driven, it seems feasible to look beyond available data sets, seeking connections with other bodies of knowledge and research partners, across borders and across research communities.

The theory/practice relationship

The research problem involved here concerns the dilemma between seeking influence and accumulating knowledge. The problem regards the relationship between knowledge and action. As one of the most fundamental issues in social science it gets scholarly attention from various angles. In policy analysis, a continuum of stances currently can be observed ranging from ‘speaking truth to power’ (Wildavsky, 1979) to ‘making sense together’ (Hoppe, 1999). It is commonly accepted that both mere instrumentality and overall societal irrelevance are to be avoided. This being so, in much implementation research almost by definition a ‘top-down’ view is being adopted, particularly when a case study of a single public policy is involved. The assumption is then that once one knows the good intentions as laid down in a law or official policy document, one can infer the results-to-be-realized from it. In other words: the policy inputs determine the policy outputs, in what is seen as a one-to-one relationship. However, premature judgements on ‘implementation failures’ and ‘policy fiascos’ may be the result. As far as this implicitly happens, this may lead to confusing the normative and the empirical.

Winter’s (2012) plea can be interpreted as one ‘to start at the other end’. It can be adopted as a way to deal with the issue of the theory/practice relationship. Rigorous implementation research then begins with formulating a theory-based question, specifying what empirical variation in governmental outputs (and if possible outcomes) needs explanation, documenting what is known already and what not, and following a pragmatic research strategy. One could call this a bottom-up perspective that does not reject the case for systematic analysis.

The multi-layer problem

This research problem concerns dealing with complexity. Government acts at different geographical scales. These include that of the national ministries, local administration and even hospitals, but also, for instance, the European Union. Looking at neighbourhoods, Somerville (2011: 82) defines multiscalarity as ‘a property of trans-scalar action, or action across more than one geographical scale’. Earlier Goggin et al. (1990) argued for attention to the consequences of the ‘multi-level’ character of government appearances for implementation theory and research. In such ‘multi-level’ research, there is a chance of confounding actors and activities, a phenomenon Hill and Hupe (2003) call the ‘fallacy of the wrong layer’. The ‘stage’ of implementation is seen as a range of activities (focus) inherently connected with a certain category of actors working at a specific spot (locus).

‘Opening up the black box’ of street-level interaction may be advised as a way to deal with the multi-layer problem – with a reference to the subtitle of Palumbo and Calista ‘s edited book (1990). Looking at what happens and why in social interaction in micro-networks, however far from ‘Washington’, ‘Whitehall’ or ‘Brussels’, may imply a mode of analyzing implementation in a both scientifically advanced and societally highly relevant way.

The policy/politics nexus

This research problem concerns how to conceive policy analysis as social science analysis. Lowi (1972) has tried to distinguish between types of policy while looking at the substance of the latter. Many public policies – certainly in the area of human services, but not only there – have more than a technical character. Their substance may indeed be expected to matter. Implementing such policies means making value-loaded judgements.

In a study of Danish employment policy, for instance, May and Winter (2009: 466) point to (legitimate) differences between local and national officials in the way they address the same policy – differences related to its substance. Looking at the means and ends of a public policy as technical devices, and approaching their relations accordingly, may imply a pitfall. In fact, it is seldom obvious when ‘policy formation’ ends and ‘implementation’ starts. Sometimes these two ‘stages’ in the policy process seem integrated, like in the management of the economy where a politically driven steering process is going on all the while.

In most cases there are institutional boundaries, which may obscure the possibility that actual ‘policy formation’ takes place where (only) ‘implementation’ is presumed. Observed behaviour of policy implementers then may be too readily judged as deviant from the norms laid down in the legal provisions of the policy. Policy goals but also policy instruments express a degree of conflict and consensus and embody certain values. While May (2012: 283) speaks of ‘policy politics’ it can be stressed that all public policy is fundamentally political. Public policies have a political character, not in the least in the sense that questions are involved about ‘what has to be implemented’.

Rather than hiding or ignoring political dimensions of public action, ‘bringing politics back in’ can be suggested to address the policy/politics nexus (cf. Evans et al., 1985). This means the incorporation within research of attention to power mechanisms, stakes, values, symbolic aspects and other non-technical dimensions of public policies. All these elements, however ‘political’, can be treated in a clinical way by looking at ‘what happens’ rather than ‘what should happen’.