Does agency age matter in administrative reform?: Policy autonomy and public management in Swedish agencies

Transcription

1 Policy and Society ISSN: (Print) (Online) Journal homepage: Does agency age matter in administrative reform?: Policy autonomy and public management in Swedish agencies Birgitta Niklasson & Jon Pierre To cite this article: Birgitta Niklasson & Jon Pierre (2012) Does agency age matter in administrative reform?: Policy autonomy and public management in Swedish agencies, Policy and Society, 31:3, , DOI: /j.polsoc To link to this article: Policy and Society Associates (APSS) Published online: 03 Mar Submit your article to this journal Article views: 49 View related articles Citing articles: 3 View citing articles Full Terms & Conditions of access and use can be found at Download by: [ ] Date: 05 December 2017, At: 06:27

2 Available online at Policy and Society 31 (2012) Does agency age matter in administrative reform?: Policy autonomy and public management in Swedish agencies Birgitta Niklasson *, Jon Pierre University of Gothenburg, Department of Political Science, Sweden Abstract This paper looks at the policy autonomy of different generations of agencies in Sweden. Taking a historical institutional perspective, we argue that the policy autonomy of public agencies is related to the dominant political and administrative trends at the time when the agency was first created, i.e. that institutional relations created at that formative moment tend to persist. In the Swedish context, one such trend is of particular importance; the managerial trend that started in 1963 with the Program Budgeting (PB) reform. This was an institutional reform that presented a paradigm shift in Swedish public administration. This reform was reinforced in the 1990s by the New Public Management (NPM) reform, which introduced new instruments and measures consistent with the managerial paradigm previously introduced by the PB reform. This managerial trend aimed at limiting agencies policy autonomy. Thus, we hypothesise that (1) agencies founded before the PB reform will enjoy more extensive policy autonomy than agencies founded during or after this reform and (2) there will be no such difference between agencies founded before and after the NPM reform. These hypotheses are tested on empirical data consisting of a web survey in which 157 Swedish agencies have stated to what extent they perceive that they enjoy policy autonomy. The results confirm both our hypotheses, thus demonstrating the importance of considering the scope of institutional change when analysing and comparing the effects of administrative reforms in different countries. # 2012 Policy and Society Associates (APSS). Elsevier Ltd. All rights reserved. 1. Introduction The relationship between public agencies and government departments has been a key issue in administrative reform for the past couple of decades. Indeed, the policy-operations institutional split and the creation of autonomous public agencies has been one of the most conspicuous manifestations of public management reform in a large number of countries (Aucoin, 1998; Christensen & Laegreid, 2006; Pollitt & Talbot, 2004). Executive agencies have been institutionalised features of government for some time now and there is a growing interest in the policy autonomy of these agencies. The issue of policy autonomy is important not least since it relates to wider issues concerning the legitimacy of the political system. In formal terms, and leaving the growing debate on emerging forms of output-based legitimacy aside, citizens can only hold politicians not civil servants to account for political decisions and agency performance. Thus, not least from a democratic perspective we need to know to what extent public agencies act independently of their political masters, how much influence agencies exercise in policy processes, and what factors affect these patterns. The purpose of this paper is to contribute new knowledge to this field of research. * Corresponding author. addresses: birgitta.niklasson@pol.gu.se (B. Niklasson), jon.pierre@pol.gu.se (J. Pierre) /$ see front matter # 2012 Policy and Society Associates (APSS). Elsevier Ltd. All rights reserved.

3 196 B. Niklasson, J. Pierre / Policy and Society 31 (2012) Many accounts of agencies policy autonomy overlook the fact that such autonomy is essentially a two-way street. There is now growing evidence of the significance of executive agencies in shaping public policy. Still, the typical approach to agency autonomy is to investigate the extent to which departments can control the policy activities or the internal management of the agencies, thus ignoring the more complex issues of agency influence over policy. However, there are studies, for example that by Verhoest, Verschurere, Rubecksen, and MacCarthaigh (2010), that take the opposite perspective by studying the level of autonomy as reported by the agencies themselves. Following their example, we look at the level of policy autonomy perceived by different generations of Swedish agencies. In most countries, the rise of autonomous or semi-autonomous public agencies is closely associated with the rise of New Public Management (NPM) (Aucoin, 1998; Christensen & Laegreid, 2006; Pollitt, Bathgate, Caulfield, Smullen, & Talbot, 2001; Pollitt & Bouckaert, 2004; Pollitt & Talbot, 2004). Starting in a few Anglo-Saxon countries in the 1980s, this reform spread quickly across the Western world and beyond. One of the key points of this reform was to reduce the policy autonomy of agencies and instead increase their operational autonomy. Drawing on a historical institutional perspective on NPM reform, Verhoest et al. (2010: ) show that agencies formed before the introduction of NPM enjoy higher levels of policy autonomy compared to agencies that were created later. There are several reasons why the time of foundation of an agency should have an impact on its policy autonomy. The historical institutional explanation offered by previous scholars (Lægreid, Roness, & Rubecksen, 2006; Verhoest et al., 2010:35) is that, analytically speaking, age is a proxy for agency culture: a public sector organization is established at a specific time, and is therefore shaped by particular cultural contexts or norms and values that leave a permanent impression on it (Verhoest et al., 2010:35). This paper conducts an empirical test of these findings by measuring policy autonomy in a different way and in a different cultural and institutional context. The empirical data is collected in Sweden where core public management ideas were introduced in 1963 during the Program Budgeting (PB) reform, i.e. about two decades before the global diffusion of NPM. Our analysis substantiates the importance of considering the types of change triggered by administrative reform when using age as an explanatory variable. Specifically, we investigate what degree of change is necessary before we should expect a difference in policy autonomy between agencies that were founded before and after an administrative reform. Our results indicate that this requires what Hall (1993) calls an institutional change of the third degree. We will discuss Hall s typology of change and how it can be applied to administrative reform later in the paper. The remainder of the paper is organised as follows. We first define policy autonomy and discuss how such autonomy is related to agency age and administrative reform in an institutional perspective. This section will be followed by a brief description of the Swedish administrative context and the introduction of two hypotheses. We then turn to our empirical analysis by introducing our data set, specifying independent variables and presenting our empirical results. The paper closes with a discussion of the results and how they speak to the conceptual issues raised earlier in the paper. 2. Defining policy autonomy There are several different kinds of autonomy (see e.g. Christensen, 2001; Lægreid et al., 2006; Verhoest et al., 2010). The focus in this study is on the policy autonomy of public agencies. Policy is often associated with a decision or a sequence of decisions that define a course of action, for example what the goals of an agency should be and through what strategies these goals should be achieved (Hill, 2009:14 19). Policy autonomy is thus related to the authority to make decisions on goals and strategies. An agency enjoys a high degree of policy autonomy when these competencies are delegated from external actors, such as the government or the parliament, to the agency itself by reducing regulation and ex ante approval requirements of other actors (Verhoest, Peters, Bouckaert, & Verschurere, 2004). A complete delegation of these tasks is however unlikely. We therefore also consider agency involvement in policy formulation processes as an aspect of policy autonomy. Thus, an agency does not just enjoy policy autonomy when it sets its own policies protected from any influence of the parent department; policy autonomy is also at hand when an agency can influence early stages of the policy process, such as the initiation, the information, the consideration, and the decision-making stages (Jenkins, 1978:17) by having a high degree of access to the parent department or to the context in which these processes take place. We will return to the issue of how we operationalise and measure policy autonomy in the methodology section.

4 3. An institutional perspective on agencies B. Niklasson, J. Pierre / Policy and Society 31 (2012) Institutional analysis has been a major, if not the dominant subfield of political science for the past couple of decades, with its most recent version dating back to the mid-1980s and early 1990s (see Evans, Rueschemeyer, & Skocpol, 1985; Hall, 1986; March & Olsen, 1989; Steinmo, Thelen, & Longstreth, 1991). Over time, institutional theory has evolved into a group of distinct sub-theories such as normative, historical, and rational choice-based institutional theory (see Hall & Taylor, 1996; Koelble, 1995; Peters, Pierre, & King, 2005). For the present analysis, with the longevity of the agencies as the main explanatory variable, we draw on the historical institutional perspective. This approach raises the question of to what extent an administrative system needs to change before we may expect individual agencies that were founded before and after this reform to display different levels of policy autonomy. The historical institutionalist approach argues that overarching policy choices represent significant political and financial commitments to the state that contribute to making these policy choices de facto irreversible. Coupled with the institutionalisation of policy not least in the creation of agencies to implement policy overarching policy decisions define a path-dependency that famously shapes and constrains future decisions on policy and instruments (Thelen & Steinmo, 1991:13). The lock-in and path dependency that this version of institutional theory highlights has both a structural and a normative dimension. In terms of structure, major policy choices usually require significant investments in building an administrative apparatus for the implementation of policy. The normative dimension relates more to the ideational or ideological foundation of policy, articulating why this policy is important and fair. These two dimensions integrate in the creation of agencies so that these bureaucratic structures become important advocates and defenders of policy. This is not primarily a matter of a public organisation ensuring its future existence. Rather it is a matter of an agency as the administrative spearhead implementing a salient political project that becomes infused with the norms and values sustaining that project. This fusion of structure and norms is often seen as a defining feature of institutionalisation. Philip Selznick s (1957:40) famous definition of institutionalisation as the embodiment of values in an organizational structure is an eloquent capture of how norms and structure gel and mutually reinforce each other. In this paper, we argue that the norms and values regarding agencies policy autonomy that permeate the Swedish administrative system at a particular time similarly infuse the structures of the individual agencies that are created at that time. If norms and structure individually are known to be resistant to change, the combination of the two is even more resilient to exogenous pressures. It is indicative that a Swedish Government Official Commission some time ago now argued that it is in fact easier to change policy by creating a new agency and provide it with fresh policy directives than it is to alter the policy direction of an already existing agency (SOU, 1983:39). Thus, the norms infused in an agency structure at the time of the agency s foundation are likely to resist changes of these norms that may later occur in the agency s external environment. Institutional changes can and do occur, however, either incrementally, for example through displacement, layering, and exhaustion (Streeck & Thelen, 2005:31), or abruptly through radical shock and critical junctures caused by, for example, a performance crises (Collier & Collier, 1991, chap. 1; March & Olsen, 1989:64). The most dramatic kind of institutional change what Hall (1993) calls a change of the third degree is often associated with the latter reform process. Spurred by a dramatic incident or a window of opportunity (Kingdon, 1995), radical changes of basic assumptions, norms, ideals, and goals may gain currency and can be implemented. A third degree change can, however, also follow from several minor changes that take place over a longer period of time. Similarly, the gradual introduction of new policy instruments (the second degree of change) or adjustments of the way old policy instruments are implemented (the first degree of change) may result in extensive institutional changes over time (Baumgartner & Jones, 2009; Streeck & Thelen, 2005:8). Given the combined resilience of structure and norms, i.e. institutionalisation, of agencies, we expect that only third degree changes in the Swedish administrative system will be able to affect the ideas regarding policy autonomy held by individual public agencies. Focusing on the possible impact of a third degree change is a reasonable starting point, since changes of lower degrees are unlikely to affect the policy autonomy of agencies if third degree changes fail to do so. Furthermore, changes of the first and second degree are common and it would be a challenge to identify them all. The number of new agencies related to each first and second degree change would be very low, which would make it difficult to draw statistically reliable conclusions regarding their effect. In order to validate our results, however, we will compare the effect of a third degree change with that of a second degree change.

5 198 B. Niklasson, J. Pierre / Policy and Society 31 (2012) In the following section, we offer a brief description of the characteristics and the development of the Swedish administrative system. The purpose is to identify institutional changes of the third degree in this system that might have affected the policy autonomy of Swedish public agencies. 4. The Swedish context As Yesilkagit and Christensen (2010:72) point out, the Swedish system is distinct because a dualist principle distinguishing between government policy and policy implementation permeates not only government organization, but has been inherent in the constitution ever since the 17th century (Andersson, 2004). This dualism emphasises a strict division of labour among the departments, headed by political appointees (the ministers), and the public agencies, led by top civil servants. To be sure, there are even constitutional arrangements that uphold this ancient division of tasks and non-interference between these two kinds of governmental bodies. These arrangements confine the role of the departments to defining the policy objectives of the agencies; they are not allowed give precise directives on how agencies go about implementing these (Instrument of Government, chap. 12, 2). This arrangement was partially based on the expectation that a more independent civil service could secure the power balance between the executive and the legislative powers (RRV, 1996:50:43). Over time, a living constitution has evolved, however, according to which agencies may (and indeed in some cases are specifically required to) play a role in policy formulation (Instrument of Government, chap. 7, 2). There is also strong empirical evidence that departments and agencies have frequent informal but institutionalised contacts on policy matters. These networks have been persistent over time and appear to be in the interest of both parties. Through these informal exchanges, departments convey their views on policy to agencies while agency staff can offer government departments policy advice and expertise (Pierre, 1995). Furthermore, while the departments have been careful not to allow their staff to expand, agencies, by virtue of being more secluded from the political leadership and media scrutiny, have witnessed their staff and budgets grow more generously. The current number of staff in the government departments is about 4500 (Premfors & Sundström, 2007), to be compared with the circa 250,000 employees within the central agencies (SOU, 2008:118). Thus, while the dualist model is to a large extent premised on the assumption that departments should be more resourceful than the agencies, history has generated exactly the opposite pattern. The gravity of central government in terms of staff and budget lies with the agencies, not the central government office (Ruin, 1991). As a result, agencies today control more expertise on policy issues than the departments; they are much less exposed to the media, and despite the constitutional dualist model of government they are in constant dialogue with their parent department on matters of policy design and implementation. The question that we are interested in here is whether older agencies are more frequent participants in these types of dialogues compared to younger agencies Swedish administrative reform We claim that it is reasonable to divide Swedish agencies into two groups: those that were founded before 1963 and those that were founded this year or later. The significance of this particular point in time is that it represents the introduction of managerial ideas in the Swedish administrative system. Given the attention to NPM reform over the past couple of decades much of this reform in the 1960s has gone more or less unnoticed (but see Sundström, 2003). However, the 1963 public management reform represents the only case of a third degree change of the administrative system that is likely to have affected the policy autonomy of current Swedish public agencies. We use the concept managerial as a collective term referring to the rationalist ideal that civil servants should function as managers implementing policies efficiently and in a goal- and result-oriented way. The managerial perspective also suggests that civil servants are protected from political interference and not involved in politics. Such managerial ideas have gradually grown more dominant in the Swedish administrative system and today they are permeating Swedish public administration entirely (Pollitt & Bouckaert, 2004:286). Below, we will briefly depict the transformation of Swedish agencies from the engaged policy advocates they have been described as before 1963, to the smooth and efficient implementation machines they are expected to be today. The oldest still existing Swedish agency was founded in 1477 (Uppsala University). It is, of course, somewhat bold to claim that only one institutional change of the third degree has affected the policy autonomy of this more than 500 year old agency. Instead, we focus this study on the more recent institutional change of this dignity. We are thus

6 B. Niklasson, J. Pierre / Policy and Society 31 (2012) subjecting our institutional hypothesis to a relatively easy test; if this fairly recent, major institutional change does not have any effect on agencies policy autonomy, we should not expect older institutional changes to have such effect either. As discussed in the previous section, the government s authority to interfere in the implementation of their policies has traditionally been quite limited in Sweden. This tradition changed somewhat during the Second World War, however, when the public administration became increasingly centralised and regulated (RRV, 1996:52; Stjernquist, 1984; Tarschys, 1983). After the war, agencies could, in fact, make very few decisions on any matter regarding their organisation structures and finance without an explicit approval from the government or Parliament. Even quite minor changes in the agencies staffs required a decision on the political level, as did any changes in the budget. Furthermore, the budget was very detailed; it specified almost to the penny how the allocated funds of the agency should be spent. Shifting money between two fiscal years was not possible, an arrangement that led to panic spending towards the end of every budget year when agencies tried to deplete any remaining funds. If they failed, they risked receiving less money the following year, since they evidently could do with less (Government bill 1978/79:111:app8:3; Sjölund, 1989). In the post-war period, most of the war-time regulations were abolished. Those that remained contributed to the increasingly complex bureaucratic system that developed alongside the new welfare state and the growing public sector following the war (Stjernquist, 1984). Between 1946 and 1976, the number of people employed in the public sector increased from about 250,000 to almost 400,000 (RRV, 1996:48). The agencies played a pivotal role in the building of the welfare state. Many agencies did not just implement political decisions; they were also deeply involved in the formulation of those decisions (Rothstein, 2005; RRV, 1996:50). The most proactive reform bureaucracies were typically found in the social welfare, education, housing, and medical care sectors. Over time, these agencies evolved into big organisations with predominantly pro-welfare state staff and a strong intra-organisational ethos related to the role of the state in the establishment of social justice (Rothstein, 2005). These were the institutional roles and values that imprinted in the institutionalisation of the reform bureaucracies. To sum up, the building of the welfare state resulted in a large, complex administrative system, in which politicians focused on supervising how agencies spent their money rather than formulating policies on what they should spend it on. Politicians were overloaded in their struggle to steer the public service through detailed rules and regulations, almost up to the point where they did not have time to make important policy decisions. Instead, civil servants played an essential role in shaping public policy. Tarschys (1983:67) even suggested: In the old days, the civil servants ruled the country in the name of the king. /.../ Today, they rule in the name of the people. The administrative system at the time was often criticised for being expensive, inflexible, and slow (Hillman, Annerberg, & Carlsson, 1978), a critique that spurred interest in new forms of public budgeting, accounting, and evaluation. Attention turned towards the United States where the Hoover Commission s report had led to the implementation of the PPB (Planning, programming, budgeting) system in the 1950s (SOU, 1967:13:12; SOU, 2007:75:73 76). A defining concept in PPB was efficiency. The agencies spending was not to be controlled in detail. Instead, it should be evaluated in relation to how they had succeeded in reaching their policy goals. Exactly how these goals were to be achieved was left to the agencies to decide. These managerial ideas spread quickly. Shortly after the Hoover Commission presented its report in 1949, a Rationalisation Committee was initiated in Sweden. Yet another Committee was created in 1956, but neither of these Committees resulted in any actual administrative reforms. It was not until 1963, when the Program Budgeting (PB) Committee was formed that things started changing in practise. At that time, the criticism against the purported extremely inefficient public sector had grown very strong (Sundström, 2003:341). There was, therefore, a sense of urgency to reform the system. Consequently, pilot groups of PB-managed agencies were created even before the PB Committee presented its conclusions and recommendations in 1967 (Sundström, 2003:329). The most radical reform of the Swedish public administrative system in modern time had been set in motion. This reform changed the very ideas of how agencies work, how they should work, and how this should be achieved. The final report of the PB Committee argued, for example, quite firmly the importance of steering the administration through goals and budget allocations that were tied to overarching programmes stating what the agencies were expected to achieve rather than to specific items of expenditure. In this report, the Committee also made fairly detailed suggestions of how this kind of steering should be carried out. They provided, for example, illustrations of how letters of allocations and reports from the agencies should be formulated (SOU, 1967:13). The PB reform can thus be described as an institutional change of the third degree (Hall, 1993), triggered by a performance crisis (March & Olsen, 1989).

7 200 B. Niklasson, J. Pierre / Policy and Society 31 (2012) The performance crisis opened a window of opportunity that allowed for a change of the administrative system to be initiated, but the new ideas could not be implemented immediately. The change, which was related to ideas and norms of how the public administration should be run, was therefore followed by a series of second and first degree changes. New methods and techniques were introduced and old methods and techniques were improved in order to give agencies more freedom to implement policies and to supply the politicians with tools to evaluate if the agencies are successful in their tasks. Despite the continuous changes following the PB reform, the general direction of the development of the public administration was never seriously called into question (Sundström, 2003:328). On the contrary, it was constantly reconfirmed, not the least through the advent of the NPM (New Public Management) reform, which was introduced in the end of the 1980s. Although the Swedish administrative system never fully embraced the privatisation aspects of the NPM model, it has displayed a readiness to adopt the managerial ideas of the model, such as the emphasis on management by objectives and results, efficiency, and flexibility, which are very much in line with those ideas already established as a part of the PB reform. The NPM reform thus presented a second degree change that accorded agencies even more managerial autonomy, for example by expanding their rights to take loans and save parts of their allocated funds as well as manage their premises (RRV, 1996:142; Sundström, 2003:327). Today, Sweden can be listed among those countries in which the public administration has a strong NPM emphasis (Hood, 1995; see also ESV, 2007:23; Petersson, 2007; Rothstein, 2005). The ideological turn in reform in 1963 and the long sequence of minor changes that followed highlights two aspects of agencies autonomy, that have gradually changed since the initiation of the PB reform. Firstly, the policy autonomy of the agencies has decreased, at least formally; today, government departments keep a tighter grip on agencies and seek to secure compliance with government-formulated policy. Secondly, the operational and financial autonomy of the agencies have increased; the managers manage the implementation process. Thus, to a growing extent, agencies are steered through the setting of goals instead of the making of detailed rules and regulations. What also stands out very clearly in hindsight is the significance of the 1963 PB reform, emphasising efficiency and instilling a public management philosophy in the Swedish agency system. 5. Hypotheses Against this institutional and contextual backdrop, we can now better understand the analytical value of agency age. Agencies that were created before the managerial reforms began in 1963 were institutionalised in a political context that promoted norms and values of the centrality of the state in society, as well the active role of agencies in the expansion of the welfare state. These norms and values did not simply vaporise when new organisational goals like efficiency, flexibility, and competition emerged as keystones in the administrative vernacular. True, older agencies, too, were targets of the managerial reforms and thus were required to adapt their organisations to fit these new ideals. There is, however, an important difference between having been engrained with those values from the outset of the organisation and adopting those values at a stage where the organisation is institutionalised with its own subculture, routines, and preferred modus operandi. Thus, our task is to study differences in the policy autonomy of Swedish agencies using two dummy variables of agency age as the key independent variables. The first one, the PB variable, is a proxy variable for the extent to which agencies were created in a pre- (before 1963), or post- (1963 or later) managerial reform context. Our first hypothesis is that agencies founded before the managerial reforms will display higher policy autonomy than agencies formed after these reforms were implemented. The argument is that older agencies experienced their formative moment before the managerial reforms were implemented, which made them less susceptible to managerial ideals. According to Yesilkagit (2004: ): [t]he discrepancy between formal and real bureaucratic autonomy can be problematic in two situations. The first situation is where an agency has little formal autonomy but aspires for more real autonomy. /.../ A second situation occurs, as already briefly discussed, when an agency enjoys a high level of formal bureaucratic autonomy; however, politicians attempt to cut back the formal freedoms of its agent. We hypothesise that managerial reform has caused the second kind of discrepancy, i.e., that agencies that existed before these reforms had grown used to enjoying a higher level of policy autonomy that they were unwilling to surrender.

8 B. Niklasson, J. Pierre / Policy and Society 31 (2012) Table 1 Public agencies in the study. In sample Response rate In analysis County administrative boards 9 (21) 67 (14) 8 (8) Universities and colleges 16 (36) 72 (26) 22 (23) Museums 4 (8) 88 (7) 7 (7) General public services 71 (159) 69 (110) 63 (67) All 100 (224) 70 (157) 100 (105) Comment: The figures represent percent. The figures in parenthesis represent numbers. Our second hypothesis is that the PB variable will perform better in explaining policy autonomy in a Swedish context than the second age variable that compares agencies that were created before and after 1990, which has proved fruitful in other countries where managerial ideas were still more of a novelty at the arrival of NPM (Verhoest et al., 2010). This hypothesis is based on our claim that the NPM reform did not represent a third degree institutional change in Sweden, but only a second degree change. Hence, we do not expect it to have as strong an impact on the policy autonomy of Swedish agencies as the PB reform. 6. Method The empirical analysis is mainly based on a COBRA (Comparative Public Organisation Data Base for Research and Analysis) 1 survey directed to the chief executives (CEO) of 224 Swedish government agencies between November 2008 and April % of the responses were filled out by the CEOs while the remainder were filled out by vice CEOs, or managers of human resource management, finances, and of specific departments. The agencies selected are those that: (1) are directly responsible to the government, opposed to being responsible to the Parliament, or to another central agency (Instrument of Government, chap. 12, 1), and (2) have their own staff. These agencies fulfil at least some of basic prerequisites that should enable them to act autonomously. They possess, for example, a distinctive area of competence, a clearly demarcated clientele or membership, and undisputed jurisdiction over a function, service, goal, issue or cause (Clark & Wilson, quoted in Downs, 1967:157). Consequently, the sample includes county administrative boards, universities and colleges, museums, and general public agencies (see Table 1). A total of 157 agencies answered the survey, which equals a response rate of 70%. In comparable studies, the figure has varied between 30 and 70%, 2 so this must be considered an acceptable turn out. The response rate varies somewhat between the different agencies however. It is the highest among museums (88%) and the lowest among county administrative boards (67%). This means that museums are overrepresented in the empirical analysis. As shown in Table 1, the analysis in the empirical section will only include 105 cases (46% of the sample), since it is only 105 agencies that have answered all the questions required for the analysis (see the discussion on measures below). Museums only constitute a very small share of the agencies included in the analysis, so it is not very likely that this overrepresentation has affected the analysis significantly. The over-representation of universities and colleges and the under-representation of agencies supplying general public services may be greater reasons for concern, something that we will discuss further in the empirical section Measuring policy autonomy Policy autonomy is measured by an additive index (0 1) that measures to what extent agencies claim to participate in activities involving initiation, formulation, or decision-making on policies relevant to their own organisation. The index is constituted by three survey questions (see Appendix for details) that are significantly correlated with each other (Pearson s R > 0.699, p =.000). The survey questions all load on the same dimension in an unrotated factor 1 COBRA is a research collaboration initiated by Guy B. Peters and Geert Bouckaert in It offers a unique opportunity to compare public administration systems in different countries (see the COBRA web page). 2 So far, the COBRA collaboration has resulted in comparable web surveys in 22 European countries apart from Sweden.

9 202 B. Niklasson, J. Pierre / Policy and Society 31 (2012) analysis (factor loadings > 0.895), using Kaiser s criterion. Cronbach s alpha for the index is The index is normally distributed Measuring the effects of administrational reforms The main independent variable in this study is related to the third degree institutional change of the Swedish public administration that began when the PB reform was initiated in This variable is thus a dummy of agencies that were founded before 1963 (1) and agencies that were founded in this year or later (0) and we will refer to it as the PB variable in the analyses. Since we take an institutional perspective that stresses the long-term effects of the norms permeating the organisation when it was first created, we allowed the agencies to define their starting point themselves. We believe that this is an appropriate method of capturing what year those who work in an organisation perceive as the most formative moment. The information about when agencies perceive that they were founded has been collected from the web pages of the agencies, or from other kinds of publications produced by the agencies. According to this information, 56 agencies were founded before The explanatory power of the PB variable will be compared to that of another age variable that we will refer to as the NPM variable. The NPM variable is based on when agencies were founded in relation to the NPM reform that started in Agencies created before 1990 score 0 on this dummy variable, whereas agencies created 1990 or later score 1 (51 agencies). In the regression analysis (see Table 2), we include the PB variable as well as the NPM variable. This means that the b-values for these variables represent the change that occurs in policy autonomy when an agency is either created before 1963 (PB = 1), or 1990 or later (NPM = 1). Those agencies that do not score 1 on neither of these two variables were created between 1963 and 1989 (50 agencies). The level of policy autonomy of these agencies is represented by the b-value of the constant Control variables The effects of the two age dummies will be controlled for by three different kinds of variables: Variables related to (1) the external relationships of the agencies, (2) the tasks of the agencies, and (3) the organisational structures of the agencies The external relationships of the agencies One of the most important external relationships of an agency is that to the parent department. It is from the department the agencies receive their instructions and, usually, most of their budget. It is also to the departments that the agencies report their performance. If this relationship is working poorly, the policy autonomy of the agency is likely to suffer. Why would the government allow high levels of policy autonomy to an agency it does not consider trustworthy or reliable? It would be wiser to keep a close eye on such agencies (Lægreid et al., 2006). Thus, we include the level of mutual trust between the agencies and the government in the analysis. The government, in this case, is the political leadership and the civil servants working at the parent department. The level of trust is measured through an additive index (0 1) built on three survey questions (see Appendix for details). These survey questions are significantly correlated with each other (Spearman s Rho 3 > 0.531, p =.000) and they load on the same dimension in an unrotated factor analysis (factor loadings > 0.822), using Kaiser s criterion. Cronbach s alpha is The confidence index is not quite normally distributed. It is skewed towards the higher values. Another external factor that might affect the level of policy autonomy of an agency is the public attention attracted by that agency. An agency that faces a lot of external criticism in Parliament, by other public agencies or NGOs, or in the media might also draw the critical eye of the government. As the institution that is held democratically accountable for the achievements of the public agencies, the government is likely to allow less autonomy to an agency that appears to have problems handling its assignment in a satisfactorily manner. To what extent an agency has received public attention is measured by a normally distributed additive index (0 1) that is based on four survey questions 3 Spearman s Rho is used in this case since the answers to some of the survey questions are not normally distributed.

10 Table 2 Regression analyses (OLS) of policy autonomy. B. Niklasson, J. Pierre / Policy and Society 31 (2012) Model 1: reform variable Model 2: external variables Model 3: task variables Model 4: structural variables Model 5: all variables b Beta b Beta b Beta b Beta b Beta Constant.416 ***.290 **.444 ***.534 **.712 *** (.05) (.14) (.08) (.19) (.25) Reform variables PB (founded before 1963).203 *** *** *** ** ***.346 (.07) (.07) (.07) (.07) (.07) NPM (founded 1990 or later) (.07) (.07) (.07) (.08) (.08) External variables Trust (.16) (.17) Public attention.385 *** ***.331 (.11) (.12) Task variables Regulatory task (.06) (.07) Advisory task (.06) (.06) Social welfare policies (.06) (.07) Structural variables Employees (logged) (.26) (.25) Budget (logged) *.222 (.38) (.35) Share of budget provided by the government (.11) (.12) Agency board (.07) (.08) Adjusted R SEE N * p < 0.1. ** p < 0.5. *** p <.01.Robust standard errors in parenthesis. The dependent variable is an additive index that runs from 0 to 1. A detailed description of the variables can be found in Appendix. (1) All models are based on the same respondents: those for whom we have valid measurements on all the variables included in Model 5. (see Appendix for details). The survey questions are significantly correlated (Spearman s Rho 3 > 0.425, p =.000) and they load on the same dimension in an unrotated factor analysis (factor loadings > 0.732), using Kaiser s criterion. Cronbach s alpha is Agency task Pollack (2002) has stressed the importance of controlling for explanation factors internal to the agency itself when studying the relationship between principals and agents. One example of an internal factor is agency task, which Painter and Yee (2011) point out as a central feature in the literature studying why public bodies are granted autonomy. Agencies with regulating tasks are often said to enjoy greater policy autonomy than those mainly dealing with general public services, or defence for example (Bendor, Glazer, & Hammond, 2001; Epstein & O Halloran, 1999; Huber & Shipan, 2002; Painter & Yee, 2011; Pollack, 2002; Verhoest et al., 2010). The government has a greater interest in ensuring that regulatory agencies appear to act independently of the political leadership, so that they may harmonise with a policy field that is largely dominated by private and international corporations. It is also a matter of

11 204 B. Niklasson, J. Pierre / Policy and Society 31 (2012) convincing free market actors of the long-term stability and objectivity of the economic policies implemented. We will, therefore, identify those agencies that have a regulatory task. Whether an agency has such a task is measured by a survey question (see Appendix for details). Having an advisory task is also likely to increase the policy autonomy of an agency. Some agencies are explicitly instructed to provide the government with expert knowledge, inquiry recommendations, and background information. These are typically highly professional organisations that possess a high degree of policy-relevant knowledge. It is, therefore, likely that they will play a great part in the policy formulation processes regarding their own activities as well. Moreover, agencies like these are harder to control and it might also be less suitable to try to do so (Pollack, 2002). To find out what agencies have advisory tasks, we have scrutinised the legal instructions and the annual letters of allocations that the agencies receive from the government every year (see Appendix for further details). Other kinds of tasks may decrease the autonomy of an agency. According to the study by Verhoest et al. (2010), agencies dealing with social welfare policies display lower autonomy. This might be because this policy field has a large impact on people s everyday lives and, thus, tends to be strongly correlated to the popular support for the government. Consequently, the government is less willing to give up too much control over these agencies. The dummy variable of whether agencies deal with social welfare issues is based on a survey question (see Appendix for details). In line with the previous study by Verhoest et al. (2010:97), we have defined social welfare policy as issues regarding housing, health, recreation, culture, religion, education, and social security Agency structure The last category of control variables is that related to the agency structure. The first one of these variables is the number of employees at the agency. As Verhoest et al. (2010) point out, agencies controlling a lot of manpower have greater resources to act autonomously. At the same time, smaller work places are more likely to develop a homogeneous culture with strong, shared norms, something which might increase the autonomy of the agency. Previous research indicates that it is this latter understanding of the relationship between agency size and agency autonomy that is the more accurate one (Lægreid et al., 2006; Lægreid, Roness, & Rubecksen, 2008), which is also in line with our institutional perspective. The information regarding the number of people working at an agency is retrieved from a report published by the Government s Survey Support (Statskontoret, 2007). This report states the number of employees in December 2007, 2 years before the web survey was collected. The second structural variable is the size of the budget controlled by the agency. As in the case of staff size, this variable might affect the degree of autonomy in two different ways. On the one hand, sizeable economic resources increase the possibilities for an agency to act independently. On the other hand, the government is likely to feel a greater need to control agencies that are more costly (Verhoest et al., 2010). We will, therefore, conduct two types of controls related to the agencies budget; one in which we measure the size of the agencies budgets and one in which we measure how large a share of their budget is provided by the government (see Appendix for details). This information has been provided by the Swedish National Financial Management Authority (ESV). The last control we will do regards the leadership structure of the agency. The existence of an agency board, for example has frequently been identified as a factor that has a positive impact on agency autonomy (Christensen & Laegreid, 1999; Painter & Yee, 2011; Verhoest et al., 2010; Yesilkagit & Christensen, 2010). The agency board works as an extra layer that the information and control exercised by the political leadership has to filter through. During this process, the information might become muddled or muffled. The CEO is also likely to be able to take a stronger stand against the government and the parent department in cases of disagreement if she has an agency board supporting her than if she stands alone. The information regarding the existence of an agency board is taken from a report by the Government s Survey Support (Statskontoret, 2008). 7. Results There are two hypotheses that will be tested in this empirical section. The first is that agencies founded before 1963 will state that they enjoy higher levels of policy autonomy than agencies founded later. The second hypothesis is that this categorisation of agencies contributes to a better understanding of the policy autonomy of Swedish agencies than a categorisation that is based on when in time agencies were created in relation to the NPM reform, which was not a third degree change in the Swedish context, but only that of a second degree. The effect and explanatory power of these two different ways of categorising public agencies based on their age will be controlled for by other relevant variables in

12 B. Niklasson, J. Pierre / Policy and Society 31 (2012) four multivariate regression (OLS) models focusing on (1) the external relationships of the agency, (2) agency task, (3) agency structure, and (4) all of the above. Let us start by testing the first hypothesis. Do older agencies perceive that they have higher policy autonomy than younger ones? The results from the OLS regressions are presented in Table 2. The main result in Table 2 is that the PB reform that was initiated in 1963 display a strong, significant, and negative effect on agencies perceived policy autonomy. Agencies that were founded before this reform claim that they enjoy higher policy autonomy than those founded between 1963 and The difference in perceived policy autonomy between these agencies is on a scale that varies between 0 and 1 in the bivariate model (Model 1). This result does not change much regardless of what control variables we include in the analysis. Agencies that were founded when the managerial reforms had been set in motion in 1963 state that they are much less involved in the initiation, formulation, and decision-making on policies compared to agencies that already existed when these reforms reached the Swedish shores. The first hypothesis is thus confirmed. We will now turn to the second hypothesis. Is 1963 a more suitable cut-off point than 1990 if we strive towards a better understanding of the policy autonomy of public agencies in Sweden? In order to answer this question, we have included a second age variable, the NPM variable, which is based on when an agency was founded in relation to the NPM reform. The results in Table 2 show quite clearly that the NPM variable is not nearly as good at explaining the policy autonomy perceived by Swedish agencies as the PB variable. The NPM variable does not display a significant effect on policy autonomy in any of the five regression models tested. Agencies created after the NPM reform in 1990 do not perceive that they have any less policy autonomy than the agencies that were created between 1963 and Apart from the PB variable, only two variables display a significant effect on agencies perceived policy autonomy. The most obvious one is the degree of public attention that an agency has received during the passed 5 years. 4 Surprisingly, it is the agencies that have been subjected to more public attention that also enjoy a greater degree of policy autonomy. This is the opposite pattern from what we expected. Furthermore, the effect seems very strong; the difference in perceived policy autonomy between the agencies that have received the least public attention and those that have received the most equals (Model 5) on a scale that runs between 0 and 1. The other variable that has a significant effect is agency budget. This effect is only significant in Model 5 and only on a 0.1 level, but it does point in the expected direction. Agencies with larger budgets enjoy a lower level of policy autonomy than agencies with smaller budgets. As mentioned in the methodology section, we have included the same cases in all five models. This means that the analyses are only based the agencies for which we have information on all the variables. The overall conclusion does not change even if we do maximise the data (N) in each and every model, but the effect of the PB variable becomes less obvious. It is no longer significant in Model 3 (b =.120, p =.060) and in Model 4 (b =.100, p =.143) that are the two models with the largest N (138). Possibly, this change is related to a slight overrepresentation in our analysis of universities and colleges (22% compared to 16% in the sample). Universities are often old institutions that are also claimed to be relatively autonomous. The majority of the Swedish universities and colleges were founded after 1963 however (72%). A university dummy included in a bivariate regression model, or in multivariate regression models together with (1) the PB variable, or (2) all variables in Model 5 does not show any significant effect on policy autonomy, nor does it change the main results in Table 2. The overrepresentation of universities and colleges in our analysis is therefore not likely to have affected our results. A group that is, on the other hand, underrepresented in our analysis is agencies dealing with general public services. They make up 71% of the sample, but only 63% of the agencies included in the analysis. 23% of the general public service agencies in the sample were founded before There does not seem to be a relationship between this kind of agency task and policy autonomy, however. A dummy separating agencies with a general public service task from the rest does not display a significant effect on policy autonomy in neither a bivariate nor in multivariate (including the PB variable, or all the variables in Model 5) regression analyses. There are no reasons to suspect that the underrepresentation of agencies providing general public services in our analyses have affected our results significantly. 4 A model in which maximum data is used and the only independent variables included are the PB variable and the public attention of the agencies, the N = 134, adjusted R 2 =.101, SEE =.291, b-value for the PB variable =.142 ( p =.008), b-value for public attention =.320 ( p =.003).