Effects of Autonomy, Performance Contracting, and Competition on the Performance of a Public...

Introduction

In many industrialized and developing countries a shift can be observed in the way governments control their public organizations that are involved in policy implementation and service delivery (henceforth called "public agencies"). Control on inputs by the government is reduced, implying more managerial autonomy for the public agency. Three new mechanisms of control replace input control: control by results in the form of performance contracting, control by means of financial incentives, and control by the introduction of competition. Public services are no longer delivered by strict input controlled and incrementally financed units within monolithic and monopolistic government bureaucracies. Instead, they are increasingly provided by public agencies that have considerable managerial autonomy with respect to the use of their inputs. These agencies are regulated by result-oriented performance contracts and the introduction of performance enhancing incentives or market pressures. Elements of this trend have been described in detail by several scholars (OECD, 2002; Peters & Bouckaert, 2004; Pollit & Talbot, 2004). Also in Belgium and Flanders, such a shift could be observed.1 Moreover, influential international organizations and national governments are major proponents of such a shift in control mechanisms (see, e.g., OECD, 2002). However, a crucial question is to what extent such a shift in control mechanisms results in a better performance of these agencies, as is assumed in the New Public Management (NPM) doctrines.

The assumption is as follows: The performance of public agencies can be enhanced only if more managerial autonomy (i.e., less input control on financial and human resources matters) is devolved to them by government, and if they are forced by result control, financial incentives, and competition to use that autonomy in order to increase their performance (Bouckaert, 1997). According to NPM, public managers cannot be trusted to perform in an optimal way unless they are forced to because they serve their private interests, which are not always congruent with those of central government (Schick, 1996,18-19). Therefore, information about their (relative) performance should be available to the government, as well as incentives to align their interests with those of the government.

As most contemporary studies on the effect of autonomization and control of public agencies on performance seem to show ambivalent results (for a review of the literature, see Pollitt, 2004, 331-34; Verhoest et al., 2004), it is not possible to confirm or reject any of these assumptions. Most studies focus on the effect of only one or two of the four control mechanisms and thereby neglect the possible effects of changes in the other control mechanisms. However, mostly NPM-minded governments tend to combine the control mechanisms under review. Therefore, there is a need for additional systematic empirical research, based on a theoretical positioning of the four control mechanisms and their interactions. This article presents a theoretical model based on principal-agent theory, which enables to hypothesize and to study the combined effects of (changes in) different control mechanisms on the performance of the public agencies. Furthermore, it presents some first empirical material from an embedded in-depth case study. The empirical data presented in this article focuses on the control by the Flemish government of the Flemish Employment Service in the period from 1992 to 1999 (Verhoest, 2002). The extent to which the theoretical model is corroborated by the case study enables us to make an assessment on the value of the model for further empirical research and on the features of such research. As such, the case study has, besides an explanatory focus, an important albeit strongly theory-driven explorative component, aiming at developing pertinent and more detailed hypotheses for further inquiry (Yin, 2003,6). This study fits into the tradition of empirical research on governance (Lynn, Heinrich, & Hill, 2001).

In this article I first develop the theoretical model based on principal-agent theory. The research methodology is set out in the second section of the article. In the third section, the findings of the case study are presented. I conclude by assessing the value of the theoretical model for further empirical research by linking the outcomes of the case study to the broader public management literature.

The Theoretical Model

Principal-Agent Theory

Principal-agent theory (Jensen & Meckling, 1976; Pratt & Zeckhauser. 1991) is a particularly promising theoretical framework to underpin the NPM-assumptions about the effect of autonomy, result control, and competition on the performance of public agencies. First, principal-agent theory focuses on the central question of how the principal can control the agent in a context of information-asymmetry and goal conflict. Situations of information-asymmetry and goal conflict are particularly prevalent in the public sector (Boorsma & Halachmi, 1998, 2). Moreover, this theoretical framework is appropriate to study the control relationship between a government and its agencies, because "where one actor has the power to the purse and can formulate goals for another actor, who receives the money and is better situated (e.g., by virtue of having more knowledge or time) to pursue them we may apply the theoretical framework for analysis" (Boorsma & Halachmi, 1998, xxv). Second, in several OECD countries, principal-agent theory is used as a theoretical justification to underpin NPM reforms, such as agencification and marketization (e.g., Schick, 1996). Third, just like in NPM, agents are perceived as maximizing their own interest at the detriment of the interest of the principal. Fourth, as I will show later, the control mechanisms under study can be linked to the basic control devices in principal-agent theory, which enables one to hypothesize why these control mechanisms foster performance of public agencies. Other theoretical frameworks that have been used to explain performance of public agencies such as open system theory, network theory, theories on contract design theory, or sociological institutionalism are unlikely to provide precise and detailed hypotheses of how these control mechanisms work ( see Bouckaert & Verhoest, 1999; Pollitt, 2004, 331-34). Even when using rational choice theories like principal-agent theories to examine the effect of "autonomization" or "contractualization,. . . we need some very precise studies" (Pollitt, 2004, 332). Providing such a study is the aim of this article.

Central in principal-agent theory is the relationship between, on the one hand, the principal, and, on the other hand, the agent that is supposed to execute some tasks on behalf of the principal. Because of specialization the agent has some discretion in executing these tasks. Moreover, there is ex ante and ex post information asymmetry between the principal and the agent concerning the production process, as the agent knows more about the used processes, the achieved results, and important circumstances. The interests of the agent do not necessarily coincide with those of the principal (i.e., goal conflict).2 The agent, as a utility maximizing actor, will try to use its discretion to pursue his or her own goals at the detriment of the principal. As such, the agent could behave in an opportunistic manner, resulting in "adverse selection" and "moral hazard" problems (Alchain & Demsetz, 1974; Arrow, 1991, 37-39). The principal may use three control devices to motivate the agent to perform well by reducing problems of information asymmetry and goal conflict between itself and the agent: (1) monitoring systems to measure and evaluate the performance of the agent, skills of the agent, and interfering environmental conditions; (2) bonding arrangements by which the agent can provide guarantees that he will act in accordance with the principal's interests or by which "contractual limitations on the manager's decision making power" are set (Jensen & Meckling, 1976, 325); and (3) systems of financial incentives that link reward to the performance of the agent, resulting in risk-turnover from the principal to the agent.

This basic principal-agent model has been applied with great sophistication in economics resulting in a normative "theory of principal-agent," which focuses on the design of optimal incentives schemes (e.g., Sappington, 1991; for an application to the public sector, see Dixit, 2002), and in a positive "agency theory" with a focus on issues like corporate governance, management control, and accounting. The model has increasingly been applied by political scientists to the public sector, leading to a vast body of literature on the positive "theory of political control of bureaucracy" (e.g., Moe, 1984; Wood & Waterman, 1994). Moreover, public management scholars have applied the model to issues such as autonomization, control, financial systems, and contracts in government (e.g., Boorsma & Halachmi, 1998; Horn, 1995). This article fits mainly in these two last strands of literature.

The Principals and the Agent in the Theoretical Model

Taking into account the difficulties of applying principal-agent theory to the public sector (such as the existence of multiple principals, multiple and poorly observable goals; see Boorsma & Halachmi, 1998; Dixit, 2002; Moe, 1984, 765), I develop a model that allows the study of the combined effects of different control mechanisms on the performance of the public agency. The model considers a public agency as the agent. The government that controls the public agency is defined as the principal. However, in a traditional, functionally structured government bureaucracy, a single unit is controlled by multiple actors. The spending minister, the minister of the Budget, and the minister of the Civil Service, their departments and control agents: each control one aspect of the production process of the public agency. These multiple principals set their own objectives, which may be conflicting (Wood & Waterman, 1994, 147). This is known as the "multiple principals' problem" in the public sector. My analysis takes this multi-actor setting into account, and therefore, it aims to be more sophisticated than most principal-agent theoretical models in political sciences and public administration literature that limit themselves to dyadic constellations (i.e., an unified agent and an unified principal) (see also Waterman & Meier, 1998).

The Control Devices in the Theoretical Model

In the theoretical model, I insert the changes as to the four control mechanisms as independent variables: i.e. "reduction of input control," "increase of result control," "increase of performance related funding," and "increase of competition." These four control mechanisms are each considered to be referring to one of the original principal-agent control devices, that is, bonding, monitoring, and incentives. The use of these control mechanisms is, according to the theoretical model, supposed to result in a higher motivation of the public agency with respect to the fulfillment of the objectives of the government. The public agency's motivation is assumed to rise because of a change in some intermediate variables: by applying the four control mechanisms (1) the government gets more information about the performance and efforts of its public agency (i.e., reduction of information asymmetry); (2) the goals of. the public agency and the government are better aligned (i.e., reduction of goal conflict) (Pratt & Zeckhauser, 1991,15); (3) the public agency has more room to decide itself on the optimal mix of inputs (e.g., finances and personnel) (Jensen & Meckling, 1976, 325); and (4) the public agency is less confronted with mutually conflicting objectives of the different principals (i.e., reduction of the "multiple principals" problem"). Figure 1 presents the effects of the different control mechanisms on these intermediate variables and ultimately on the performance of the public agency. The distinctive causal relations between the . different control mechanisms and the intermediate variables are numbered in Figure 1. Hereafter I will refer to these numbers when discussing the different relationships.

"Input control" refers to the extent to which the government, in particular, the ministers of the Budget and the Civil Service, constrains the choice and the use of inputs (e.g., human and financial resources) by the public agency in its production process.3 Input control may be considered in the perspective of principal-agent theory as the set of major bonding arrangements, which limit the decision-making power of the agent (Heyman, 1988). Therefore, input control restricts the managerial autonomy of the public agency (for different dimensions of autonomy; see Verhoest et al., 2004). A decrease in the level of input control (i.e., an increase in managerial autonomy) could have a positive effect on the "performance" of the public agency, for it increases the extent to which the public agency itself can decide on the optimal mix of inputs to produce the asked results (see relation 2a in Figure 1). Moreover, a decrease of input control reduces the "multiple principal problem" (see relation 2b in Figure 1), because the involvement of the ministers of the Budget and the Civil Service in controlling the agency is diminished.

"Result control" means that the government controls the public agency by setting clear objectives and targets, by monitoring, and by evaluating the achieved results against the set targets. Result control is increasingly used in the public sector with performance contracts, performance measures, monitoring systems, and performance audits as its main instruments. Result control emphasizes ex post monitoring of the achieved results. Result control is believed to temper opportunistic behavior by the public agency and to enhance its performance in several ways (see Bouckaert, 1998,142; Lane, 1995). First, the information asymmetry between the government and the public agency as to the performance of the latter is lessened by the use of information revealing instruments (e.g., business plans, cost accounting, performance evaluations, audits; see Besanko, Dranove, & Shanley, 1996, 628; Dixit, 2002, 701) (see relation 2c in Figure 1). Second, the goals of both parties are aligned more closely because clear objectives and targets are set and negotiated (see relation 2d in Figure 1). Third result control instruments like performance contracts may set priorities among the objectives of the different involved ministers, reducing the "multiple principals' problem" (see relation 2e in Figure 1) (Garner, 1996, 94).

The extent to which the funding of the public agency is linked to its performance is another control mechanism that the central government may use. The link of funding with performance can be strong or weak depending on the extent to which the relative increase or decrease of funding is equal or less than equal to the relative performance of the agency. In the view of principal-agent theory, performance-related funding refers to incentives and risk-turnover as control devices (Bamberg & Spremann, 1987; Dixit, 2002; Frant, 1996; Vosselman, 1996). Performance of the public agency is enhanced if funding is linked more strongly to achieved performance, resulting in a greater alignment of the goals of the public agency to those of the government (see relation 2f in Figure 1).

The central government can also control the public agency by introducing competition from other providers in the involved market (Common, Flynn, & Mellon, 1992; Douma & Schreuder, 1992, 80-81). Competition strengthens the motivational effects of result control and performance-related funding. The existence of other providers on the markets enables government to benchmark the performance of the public agency with the performance of other providers (Arrow, 1991, 46^18). Such a benchmarking reduces the ex ante and ex post information asymmetry between the government and the public agency (Walsh, 1995; see relation 2g in Figure 1). Moreover, if competition is combined with a performance-related funding, the risk-turnover, caused by the latter mechanism, is even enhanced (see relation 2h in Figure 1).

Hypotheses in the Theoretical Model

Reflecting the assumptions in NPM and in international practice, the main hypothesis states: "A higher performance of the public agency will be achieved by a change of the control arrangement in which a reduction of input control by the government (implying more managerial autonomy) is combined with an increase of result control, an increase in financial incentives (i.e., performance related funding of the public organization) and an increase in competition." In order to assess fully the validity of the developed model in an empirical study, all the relationships in Figure 1 should be confronted with empirical data. Two levels of analysis can be distinguished in the empirical study. The level 1 analysis looks for evidence of causal relationships between the changes in control mechanisms and changes in performance:

1. There must be an association between the level of the control mechanisms and the level of the performance of the public agencies under study in the way as formulated in the main hypothesis.

2. The time-order between changes in control and changes in performance must be such that the former can be denoted as cause of the latter.

3. Rival explanations to explain the association between the level of control mechanisms and level of performance do not hold. The level 2 analysis examines whether the observed causal relationship between control and performance is due to the intermediate variables as shown in Figure 1.

4. There must be an effect of the level of control mechanisms on the intermediate variables as is assumed in the theoretical model (see relations 2a to 2h on Figure 1).

5. There must be an effect of the level of intermediate variables on the motivation of the public organization to achieve the goals, set by government, as is assumed in the theoretical model (see relations 3a to 3d in Figure 1).

Research Methodology

The Research Field

I focused on the control of the Vlaamse Openbare Instellingen (VOI's) by the Flemish government in the period from 1992 to 1999. The VOI's are statutory public law agencies, which have some managerial autonomy. Most VOI's are externally decentralized agencies with a governing board. The spending minister has no direct hierarchical authority over these VOI's but the minister can control them only by using the instruments that are stipulated in the legislation concerning these VOI's. Traditionally, the control system relied heavily on ex ante and input control, limiting the autonomy and flexibility of the VOI's. In the period from 1992 to 1999, the Flemish government changed its policy concerning the control of the VOI's from an ex ante input orientation to an ex post result and market orientation.

The Research Strategy

The research aims to assess the value of the theoretical model and its hypotheses (see the concept of analytical generalization in Yin, 2003, 10) for further empirical research using case studies as research strategy. Case studies allow us to better deal with the complexity of the model and the context than surveys would. Given the theoretical model, the public agencies under examination should preferably have experienced a shift of all of the four control mechanisms, so that the effect of each of them-separately and in interaction with the other control mechanisms-could be analyzed. In order to select appropriate cases, I screened by means of document analysis and interviews the evolutions of the control of 32 VOI's by the Flemish government in the period from 1992 to 1999.4 This review of the research field showed that only one VOI had experienced such combined changes in the four control mechanisms during that period: the Flemish Employment Service (FES, Vlaamse Dienst voor Arbeidsbemiddeling en Beroepsopleiding). This public agency could be considered a "critical case" or even a "unique case," which allowed for a single case study as an appropriate research strategy (Yin, 2003, 40-42). The FES is the largest VOI, with the number of personnel at 3,680 full-time equivalents at the end of 2002 (VDAB, 2003, 63), and the second largest as to its budget (EUR 300 million in 2002; see VDAB, 2003, 69). In the 1990s, it performed two main tasks, job brokerage and vocational training, and it was internally structured according to these lines. During this period, the management of the FES and of its divisions did not change: the same leaders remained in charge. Starting in 1992 with quite similar control arrangements, the two organizational divisions of the FES were until 1999 each subjected to somewhat diverging control arrangements. The analysis of the control and performance of these two divisions provided for some variation in the independent variables within the case of the FES, which allowed me to compare the effects of different control arrangements on performance within one embedded case. As such, the case of the FES consisted of two embedded cases, i.e. the job brokerage division and the vocational training division.

The Steps in the Research

The case study of the control and performance of the FES-divisions during the period from 1992 to 1999 (Verhoest, 2002, 214-50, 273-309) was constructed along the two levels of analysis, each involving specific data collection and data analysis methods. The level 1 analysis encompassed (1) the analysis of the changes in control arrangements, on the one hand, and the evolution in performance of the FES-divisions, on the other, in order to assess an association by the means of patternmatching and to assess a time order of cause and consequence; (2) the analysis of the causal link between control arrangement changes and evolution in performance by excluding rival explanations.

For the level 1 analysis two complementary methods were used. First, I constructed an "objective" image of the evolution of the different variables in time.5 Therefore, I conducted an extensive document analysis and time series analysis with quantitative data, combining FES- and external sources.6 In addition, 23 semistructured interviews7 were done to reconstruct facts for the period from 1992 to 1999. Second, this analysis was supplemented by a "subjective" analysis of the perception of the involved key-actors, of the evolution of the different variables (i.e., control, performance, and causal link) during the period from 1992 to 1999. The keyactors were the top management of the FES and its two divisions, the members of the governing board of the FES, the spending minister and its political secretariat, and the minister of Finances and its political secretariat. This was done by 12 structured interviews based on a questionnaire with fixed response categories that the respondents had to fill in before the interview. The high degree of standardization enhanced the comparison between the key-actors' individual perception. Although these interviews focused on ex post perceptions of involved decision makers, checks existed to avoid social desirable or biased answers. The results of the "objective" analysis provided for the first check. Moreover, I checked the reported perceptions of the key-actors with their real-life behavior where possible. Because of differences in the interests and position of these key-actors, their perceptions ought to be considered as complementary, rather than equal. And the interviews encompassed open questions and more in depth questions to check for correct interpretations of concepts and solid motivations for stated answers.

The level 2 analysis focuses on the analysis of the link between the control mechanisms and the intermediate variables by analyzing perceptions of key-actors with respect to the intermediate variables. This was mainly based on the "subjective" analysis of the perception of the key-actors, which allowed the opportunity for explanation building and the elimination of rival explanations.

The construct validity of the case was fostered by the careful creation of sensitive measurement instruments for most of the variables, which were validated by the use of other measurement techniques for the same variables, data and method triangulation, and respondent validation. Internal validity was stressed by the abovementioned techniques of pattern-matching, rival explanations, and explanation building (see Miles & Huberman, 1994, chapter 10; Yin, 2003,34-39, and chapter 5). Both level 1 analysis and level 2 analysis placed a heavy emphasis on addressing rival explanations. Although they were not included in the theoretical model, plausible rival models of variables were developed during the case study to explain the change in the dependent variable. The different rival models will be denoted with "rival model A," "rival model B," etc. When discussing the outcome of each step in the research, more details about the methodology used shall be given. The research was conducted in the period from 1999 to 2002.

Results of the Case Study

Level 1 Analysis: Evidence for Association, Sequence, and Causality between Control and Performance

Changes in Control, From 1994 on, in both the job brokerage division and the vocational training division the level of input control was reduced, and the level of managerial autonomy increased because the government relaxed the regulations concerning the management of financial and human resources. On the other hand, result control and financial incentives became important as control mechanisms by the introduction of a performance contract between the government and the FES. By increasing managerial autonomy and using performance contracts, the involved ministers sought to get more of a grip on the strategy of the FES and to increase its efficiency and effectiveness.

After 1994 different control mechanisms were dominant in respectively the two divisions, as is shown in Table 1.

For the job brokerage division, the change in managerial autonomy, result control, and negative financial incentives was smaller than for the vocational training division, while the change in and the level of competition was much higher in the first division. This increase in competition on the job brokerage market was due to the gradual legal liberalization of the job brokerage market in the 1990s. A final observation is that the government relaxed some control mechanisms, like negative financial incentives, after 1994. These observations were made in both the "objective" and the "subjective" analysis.

Nevertheless, according to the theoretical model, this change in control arrangement should result in an improvement of the performance in both divisions with respect to the objectives set by the Flemish government, compared with the period before 1994.

Changes in Performance. As to the measurement of the performance of the FES-sections, this was tackled by a three-way strategy. First, by doing time-series analysis of performance data based on all available FES- and external (e.g., scientific) data sources I checked to what extent the two divisions of the FES had achieved the result norms in the performance contracts. Second, I validated the results norms and the performance of the FES with respect to these norms by a comparison with the achieved results before the change in control arrangement (in 1992-93 and before). Additionally, the key-actors were asked for their perceptions concerning the evolution in performance of the FES with respect to the objectives in the performance contracts (cf. the "subjective" analysis).

The analysis of the performance shows that in the two divisions there was an improvement for most of the results concerning the quantified objectives in the period from 1994 to 1999 in comparison to the period before 1994 (see Table 1). In the case of the job brokerage division, a major performance improvement could be observed, even taking the low level of performance norms into account, while in the vocational training, performance improvement was rather small. One major exception on the good performance report for both divisions were the low results concerning the objectives concerning equity and justice of service delivery, with an insufficient representation of hard-to-place groups of jobseekers in the services and outflow. The key-actors also perceived the performance record of both FES-divisions as mixed. The involved ministers and the governing board as key-actors reported that in their view the FES-divisions put too much emphasis on strengthening the online services and the charged services for the broader groups of jobseekers and employers to the detriment of the face-to-face services and special services for the groups of hard-to-place jobseekers.

Evidence of Causality between Control and Performance: Addressing Rival Explanations. For both the job brokerage division and the vocational training division a change in control mechanisms was followed by a positive evolution in the performance with an exception for hard-to-place groups of jobseekers. Association and sequence of variables are however not sufficient to assess causal relationships. The possibility of spuriousness needs to be ruled out as much as possible: to what extent does the evidence exclude that other factors caused the partial performance improvement in both divisions? Three groups of relevant rival explanations were addressed by both extensive document analysis and time series analysis of data (i.e., the "objective" analysis), and the perception of the key-actors involved (i.e., the "subjective" analysis).

First, I examined to what extent the observed increase of performance could be explained by (1) the level of result norms over time, (2) an increase of financial resources (by an analysis of budgets and accounts), (3) quality deficits of the used performance measurement system as to technical validity, legitimacy and functionality, or (4) a simultaneous neglect of the nonquantified objectives in the performance contracts (rival model A).8

Analysis for both divisions showed that none of these factors could account solely for the observed increase of performance. This means that rival model A cannot account for the observed change in the dependent variables. As to the performance of nonquantified objectives, a rise in performance could even be observed for several nonquantified objectives in both divisions, while for the other nonquantified objectives performance remained at least constant.

The two others groups of rival explanations that were addressed were the influence of relevant external factors on the one hand (rival model B), and the influence of internal factors on the other hand on performance (rival model C). Out of the analysis of documents and interviews, I distilled a number of relevant internal and external factors, which changed in the period from 1992 to 1999 and which could have influenced the performance of the FES-divisions.9 The examination of some external factors (e.g., the economic climate and labor market situation, Federal and Flemish employment programmes, the influence of federal employment policy) and some internal factors (e.g., the introduction of online services in job brokerage, the introduction of new charged services) were included in the "objective" analysis, based on statistical data sources. All the factors were examined in the "subjective" analysis, in which the key-actors were asked (1) to what extent these factors were present or emerged in the period 1992-99, (2) to what extent and in what way these factors influenced the performance of the involved FES-division, and (3) to what extent and in what way the internal factors in their turn were influenced by the four control mechanisms under review (positive, neutral, negative) or were independent of them. In addition, the respondents were asked to what extent and in what way the control mechanisms had a direct influence on the performance of the involved division. Using the number of positive responses and the nature of the key actors as criteria, I categorized the observed positive causal links between control mechanisms, factors and performance as being strongly supported, modestly supported, and limitedly supported. A set of open questions at the beginning of the interviews and in-depth questions, looking for motivation of answers, aimed to validate the responses on the highly structured questions on influence.

Three conclusions emerge from both the "objective" and the "subjective" analysis (see Table 1). Neither the external factors (rival model B) nor the internal factors (rival model C) could account for the increase of performance in both FES-divisions, although some of them did matter quite a lot. This means that the control mechanisms had a "direct" influence on the performance of the FES-divisions, as was supported by the perceptions of the key-actors and their justification in the interviews. Moreover, the control mechanisms influenced performance in an "indirect" way by enhancing some internal factors that in turn affected the performance of the public agencies. The level of influence of the respective control mechanisms, as reported and substantiated in the interviews with the key-actors, differed in the two divisions, reflecting the level of these control mechanisms in the two divisions. As to the job brokerage division the level of competition was considered to have the largest influence, while in the vocational training the change in result control was reported to have the largest positive effect on performance. As the Level 1 analysis ruled out the most relevant rival explanations, I found substantial evidence of a causal relationship between the use of the control mechanisms and the observed changes in performance in both of the two FES-divisions.

Level 2 Analysis: The Role of Intermediate Variables such as Goal Conflict, Information Asymmetry, and Legitimacy

The level 1 analysis shows evidence of causal relationships between the control mechanisms and the performance of the FES-divisions. But how did these control mechanisms influence the performance? To what extent did the control mechanisms motivate the FES(-divisions) by affecting the intermediate variables in our model, that is, by increasing the possibility to choose the optimal mix of inputs, reducing information asymmetry between the FES and its political principals, reducing goal conflict between them, and reducing the multiple principals' problem? This question is at the center of the level 2 analysis.

At least two empirical observations, which emerged from the level 1 analysis, seem contra-intuitive when viewed in the perspective of the theoretical model. First, after 1994 there was an increase in performance over time despite that the level of some of the control mechanisms during the period from 1994 to 1999 decreased or evolved inconsistently. second, the performance with respect to the nonquantified objectives increased or at least remained constant. The FES-divisions refrained from moral hazard behavior as to these nonquantified objectives, despite that the result control by the government did not focus on these objectives.

What could both these observations point at? Again, several rival explanations can be elaborated: Maybe, other control mechanisms, which were not included in the study, did also motivate the FES (rival model D). Or maybe, other motivating intermediate variables were at stake. It could be that FES-divisions were not (or not exclusively) motivated as such by a reduction of, for example, information asymmetry or goal conflict between itself and the government, as is supposed in the theoretical model. Which other motivating intermediate variables could possibly have played a role? First, it could be that the objectives of the FES-divisions did align to a great extent or even completely with those of its political principals. In the latter case, there would be a goal congruity between the FES-divisions and their principals, rather than a goal conflict. In that case, intensive and harsh control was unnecessary to motivate the divisions, because the FES-divisions had an inherent motivation to perform well (rival model E).

Second, even in the case of goal conflict, it is possible that the FES-divisions were motivated to performance enhancing behavior by an urge to strengthen or restore legitimacy toward customers or their political principals (i.e., the spending minister, the minister of Finance, and the Minister of Civil Service), because these principals held the levers for the future (rival model F).

The level 2 analysis had to take these possible rival explanations into account. I asked the key-actors to what extent, in their perception, the intermediate variables changed, and to what extent such a change was due to the control mechanisms. Therefore, I asked the key-actors to what extent they agreed with several statements in the questionnaire, after which their most relevant responses were discussed in highly structured interviews. Although they were phrased in a more indirect way, the statements referred to the causal relationships between the independent and the intermediate variables in the theoretical model (see relations 2a till 2h and relations 3a till 3d in figure 1).10

The subsequent interviews mainly focused on the right interpretation of propositions and on the justification of randomly chosen and diverging answers. The questionnaire and the interview were also used to examine other possibly relevant intermediate factors, like goal congruity and legitimacy. The question of (the absence of) goal conflict was tackled by asking the key-actors to make a ranking of their priorities and those of the other key-actors on the basis of a predefined list of goals in the questionnaire. In the subsequent interviews their responses were confronted with their observed real-life behavior. As to legitimacy(-building) as an explanatory factor, some indirectly phrased statements referring to this notion were included in the questionnaire. By the analysis, I was able to reconstruct why and by which factors (the top management of) the FES-divisions were motivated to perform as was observed.

Now, what are the main findings of the analysis of the intermediate variables (i.e., the level 2 analysis)? First, there was at least a partial goal conflict between the FES(-divisions) and its political principals during the period from 1992 to 1999, indicating that rival model E does not hold. Second, the control mechanisms did influence motivation by affecting most of the intermediate variables in the theoretical model, but only under specific conditions. Third, the motivation to perform well was enhanced even more by the urge of the FES-divisions to restore their legitimacy towards their customers and political principals. This implies that besides the theoretical model, the rival model F also has explanatory power. Fourth, some organizational features influenced the extent to which the divisions reacted on performance-enhancing stimuli by the control mechanisms or by the threats for legitimacy. This then indicates that the public agency only reacts as assumed in both the theoretical model and the rival model F when certain organizational features are present. Figure 2 shows the model as it shows from the FES case study on which these findings are presented with numbers referring to them. These four findings will now be discussed in further detail.

Goal Conflict between the FES and the Government

The top management of the FES and of its divisions perceived themselves as emphasizing other goals concerning management of resources and substantive issues than its political principals during the period from 1992 to 1999, at least partially. This perception was shared by the other key-actors, that is, the spending minister, the ministers of the Budget and the Civil Service, and the governing board of the FES. Moreover, the acts and decisions of the FES-management reflected its priorities.

First, some goal conflict seemed to exist as to substantive objectives: Unlike its spending minister and its governing board, the top management stressed the economic function of the FES (i.e., "the quick match of offer and demand on the labor market"), rather than its social function (i.e., "the FES needs to give its services towards target groups priority"). Also, the FES-management emphasized the central market position and market share of the FES to the detriment of other private profit and nonprofit actors on the job brokerage and vocational training markets, while the spending minister, the minister of the Budget and the FES-governing board were in favor of a more active role of private actors on these markets.

Second, with respect to goals concerning the management of resources, the FES management stressed the acquisition of more financial resources and more managerial autonomy, rather than compliance to financial and other regulations and uniformity between the FES and other public agencies that were the objectives of the ministers of the Budget and the Civil Service.

There was thus no complete congruity between the FES-management and its political principals as to management and substantive goals, which could function as an inherent motivation to performance-enhancing behavior. Rather, and in line with the principal-agent theoretical model, there was at least a partial goal conflict between the FES-management and each of its political principals, calling for external motivating incentives to align its objectives with those of the principals. Rival model E does not hold in the case of the FES-divisions.

Effects of the Control Mechanisms on the Intermediate Variables In the period from 1992 to 1999 the motivation of the two FES-divisions to achieve the objectives as set by the Flemish government was reported by the key-actors to have increased (see element 4 on Figure 2). But was this increase in motivation due to changes in the intermediate variables? According to the FES-management and the other key-actors, it was partially induced by the following intermediate variables (ranked from strongly to weakly supported as to their motivating power): the increase of information on part of the government about the performance of the FES (i.e., reduction of the information asymmetry-see relation 3b in Figure 2), the closer alignment of goals of the FES and the government (i.e., reduction of goal conflict-see relation 3c in Figure 2), and the enlarged possibility for the FES to choose an optimal mix of inputs (see relation 3a in Figure 2).11 As to the reduction of goal conflict, the top management, as well as the other key-actors, perceived the alignment of its objectives to those of the government to be only limited and incomplete. In particular, concerning the substantive issues the FES-management kept on emphasizing the economic function and the central market position of the FES over the social function and a greater involvement of private actors, while the political principals stressed more the latter issues.

How did the four control mechanisms used by the Flemish government affect these intermediate variables, according to the perception of the key-actors? The comparison of the two FES-divisions points at several conditions under which the control mechanisms increased the motivation for performance-enhancing behavior by affecting the intermediate variables. Result control by performance contracting was perceived as a strong motivator by the means of reducing information asymmetry (see relation 2c in Figure 2) and goal conflict (see relation 2d in Figure 2) to the extent that the result norms were considered as challenging for the organization and to the extent that the government had influence on the objective- and target-setting. As such, the result control of the vocational training division that encompassed relatively high result norms was perceived to be more performance-enhancing than the result control with rather low norms of the job brokerage division. Information asymmetry was said to be reduced by other information-revealing instruments as well, such as external scientific evaluation studies and external organizational audits (rival model D).

Financial incentives were perceived as a modest motivator by reducing goal conflict (see relation 2f in Figure 2), only when the financial incentives were not perceived as too harsh or unfair and when resulting negative behavior could be avoided. In the vocational training division, the negative financial incentives were perceived as being too harsh and unfair because of financial problems, the lack of counter-balancing positive incentives and the automatic nature of the negative incentives. The strong negative financial incentives encouraged a less transparent attitude toward performance reporting and unilateral adjustments of performance targets in the vocational training division. They even resulted in a greater emphasis on investments in charged services against the will of most principals. As such, the harsh negative financial incentives even strengthened goal conflict as to the vocational training division.

Competition did not reduce information asymmetry (see relation 2g in Figure 2) because market actors were perceived as being incomparable to the FES-divisions and the government was ill-equipped to make comparisons. Competition made the objectives of the performance contract harder to achieve, and had as such a motivating function (see relation 2h in figure 2) as was predicted in the original model. However, in the job brokerage division, competition was perceived as a much greater motivator than result control. This indicates that competition increases motivation in itself and not just in combination with another control mechanism (see below).

Managerial autonomy was more a factor that enabled both desirable and undesirable behavior from the point of view of the political principals (see relation 2a in Figure 2). For example, the enlarged financial managerial autonomy with respect to budget shifts enabled the job brokerage division to increase the financial resources for the further development of the online services and charged services on its own initiative, rather than to invest in face-to-face and special services for hard-to-place groups of jobseekers.

These findings indicate that most of the causal links between the four control mechanisms and the performance via the intermediate variables, as assumed in my theoretical model, seem to be valid only under some conditions.

Legitimacy as Motivational Factor But was the motivation of the FES-management only enhanced by the control mechanisms used by the Flemish government? According to the FES-management, another factor did play a major role. The FES-management was intensively worried by the uncertainty of its future existence (see number 5 in Figure 2). Both FES-divisions were motivated to performanceenhancing behavior by an urge to strengthen or restore its legitimacy towards its customers, stakeholders, and, most importantly, its political principals (i.e., ministers, cabinet, and parliament) (cf. rival model F). In the job brokerage division, this urge clearly stemmed from the perceived risks, caused by the gradual liberalization of the job brokerage market and the presence of other providers. In the perception of the FES-management, this competition increased the risk of a reduction or a restriction of transactions with the Flemish government in the future. And, more important, it increased the risk of abolition or reform of the FES-division. The other providers on the job brokerage market were more and more perceived as alternatives for the FES, and they mobilized political pressure on the political principals of the FES. According to the FES-management, the job brokerage division had to prove itself and its usefulness continually to its customers, its stakeholders, its ministers, the cabinet, and the parliament. Its legitimacy toward its different principals was continually questioned and threatened, which caused, among others, performanceenhancing behavior. As such, competition in itself increased the motivation of the job brokerage division.

But did competition by others providers also induce the vocational training division to performance-enhancing behavior? As to this division, the rather low level of actual competition for market shares was only one cause of the perceived threat of declining legitimacy. The legitimacy of the vocational training division was mainly threatened by the competition by other actors for governmental subsidies, the lack of political consensus about the role of the FES-vocational training in a free market, and its increased political salience because of societal evolutions.

Performance-enhancing behavior was only one of the behavioral strategies that the FES-divisions followed in order to retain or restore their legitimacy. Competition-inhibiting behavior, well-balanced communication and image-building, and symbolic responsive behavior toward the political principals were elements of other strategies.

Legitimacy and Goal Conflict It is clear that these threats for legitimacy explain why the good performance record was both on the quantified and the non-quantified objectives in the performance contracts. But how is the low performance on the objectives regarding the hard-to-place jobseekers to be explained?

Since motivation was enhanced by legitimacy-threatening factors, one can indeed explain why the FES-management did not or could not align its objectives completely with those of the Flemish government. The quest of the FES for a stronger and broader legitimacy induced the FES management to stress the economic function of the FES rather than its social function. A better performance in its services toward the largest group of jobseekers and employers would create, in the view of the FES-management, a larger and more stable support for the continuing existence of the FES and a more stronger market position, than would a dominant orientation towards more marginal hard-to-place groups of jobseekers. An orientation towards its social function would not support what seemed to be the ultimate objective of the FES- management, that is, guaranteeing the future existence of the organization with at minimum the same level of resources and tasks, nor would a more humble stance towards its competitors. This explains the reported low level of performance of the FES on the objectives concerning services to hardto-place groups.

This adds two more conditions with respect to the motivating power of result control and competition as control mechanisms in the model. Result control by performance contracting only reduced goal conflict to the extent that the objectives and result norms in the performance contract did not endanger the legitimacy of the FES. Competition reduces goal conflict only to the extent that the objectives propagated by the market are in line with those of the government. In the job brokerage division, competition was-looking from the perspective of the political principals-too strong as a motivator, inhibiting a shift of focus of the division to service delivery for marginal target groups. Competition pushed the organization to seek its legitimacy base in market shares and market impact.

Features Affecting the Receptiveness for Stimuli Still, one question remains: Why did the vocational training division only improve its performance slightly, compared to the substantial increase of performance of the job brokerage division? On the one hand, this is to be explained by negative consequences of some control mechanisms, such as negative financial incentives, and by the lack or the impossibility of a technological breakthrough like the development of online services. On the other hand, the vocational training was perceived by the key-actors to have some inherent features, which made it less receptive and responsive to stimulation by control mechanisms or legitimacy-threatening factors (see number 6 in Figure 2). Table 2 lists the organizational features, which made the FES-divisions more or less receptive for such stimuli for performance-enhancing behavior.

Unlike the job brokerage division, the vocational training division was characterized by the relative autonomous position of its subunits and the subsequent need for indirect internal control systems. The short tradition in internal result control and the higher need for control on quality and content were other intervening factors. The organizational culture was more strongly dominated by "professional" norms. Moreover, the vision and strategy of the FES management of the vocational training division was not oriented at network management as although this was needed in its context of diverse and autonomous (internal and external) actors.

Discussion and Conclusions

In this article I developed a theoretical model which reflected the NPM-assumptions with respect to the effect of new control mechanisms on the performance of public agencies. The theoretical model was then confronted with empirical data from a single embedded case study. The case study focused on the effects of the control by the Flemish government on the performance of the job brokerage division and the vocational training division of the Flemish Employment Service. The internal validity of the case study was emphasized strongly by doing pattern-matching, excluding rival explanations and by distinguishing two different levels of analysis.

The case study findings suggest a refinement of the original theoretical model with at least four elements, which seem relevant for further empirical research.

First, the notion of goal conflict should be refined in the model in case of further research in order to take multiple sets of goals into account, like goals concerning management and substantive issues. Following from this and taking the existence of multiple principals into account, the public agency can have a partial goal conflict with each of its principals, which can be different according to the principal concerned (see Waterman and Meier, 1998).

second, the case study suggest that a public agency can be induced to performance enhancing behavior by means of managerial autonomy, in combination with more result control, financial incentives and competition, but only if these control instruments are used in a well-balanced way and if certain conditions are taken into account. However, the external validity of this finding for other public agencies than the case under review needs to be tested in further multiple case research, which should allow for theoretical and literal replication (Yin, 1993).

Third, a second, and a more fundamental motivational force for performance-enhancing behavior is at stake in the case study: the need of the public agency for strengthening its legitimacy towards its customers and especially towards its political principals (in particular, ministers, government, and parliament) in response to certain legitimacy-threatening factors. Legitimacy as a major motivational force for performance has been mentioned in other research. In his research on innovation in public services by voluntary and nonprofit organizations, Osborne found that the major impetus for innovation is the search for the search for legitimacy in the eyes of their beneficiaries, their staff, their peers, or perhaps most significantly, their funders (Osborne, 1998, 160, his emphasis). However, performance-enhancing behavior is only one of the possible behavioral strategies that the public agency may follow in order to retain/restore legitimacy (cf. Osborne, 1998, 160). Further empirical research could contribute by dealing with the interplay of motivation by control mechanisms and motivation by legitimacy-threats. Both kinds of motivations may strengthen or compensate for one another, as was shown in the case study.

Nevertheless, and this is the fourth element, the public agency may be less receptive to receive stimuli for performance-enhancing behavior because of certain organizational features (see Table 2). Some of these factors are not at all new in the literature of "organizational effectiveness" (Wolf, 1993) or pop up in research on autonomy, control and performance (e.g. service characteristics in Pollitt, Birchall and Putnam, 1998; organizational culture and mission in Ter Bogt, 1998). This could lead us to another explanation of the inconclusiveness in contemporary research with respect to performance of autonomous public organizations: that is, the limited extent to which these studies include the intervening effects of such organizational features on the relation between organizational autonomy and performance in their initial research design. The outcomes of the case study research seem to stress the importance of integrating such factors ex ante in the design of future research on autonomy, control and performance.