Successful Problem Solvers? Managerial Performance Information Use to Improve Low Organizational Performance

Successful Problem Solvers? Managerial Performance Information Use to Improve Low Organizational... Abstract Performance management is increasingly the norm for public organizations. Although there is an emergent literature on performance information use, we still know little on how managers engage in functional performance management practices. At the same time, growing evidence suggests that managers face pressure to improve low performance because of a negativity bias in the political environment. However, in managerial performance information use, the negativity bias might be reconsidered as a prioritization heuristic with positive performance attributes, directing attention to organizational goals with a favorable return of investment. I test this argument with data from public schools. A fixed-effect estimation is used to analyze how principals prioritize when they are provided with performance information on a number of different educational goals. Furthermore, a difference-in-differences model tests whether the prioritizations of certain goals have performance-enhancing effects over time. The analysis shows that principals prioritize goals with low performance and that prioritizations result in performance increase. The improvements primarily occur for goals that have a low performance level and that are repeatedly prioritized. Introduction During the last decades, one of the most widespread trends in the governing of public organizations has been the introduction of performance management. The core of performance management systems is performance information, which should inform managers’ decisions and thereby improve organizational performance (Andersen and Moynihan 2016; Heinrich 1999; Kroll 2015a; Nielsen 2014a; Snyder, Saultz, and Jacobsen et al. 2018). However, evidence of success is scarce. This is partly because managers show reluctance to integrate performance information in their routines and decisions (Kroll 2013; Melkers and Willoughby 2005; Moynihan and Lavertu 2012) and partly due to the inherent difficulty in identifying examples of purposeful use. Taken together, these circumstances leave us without much empirical and theoretical understanding of functional performance management practices (for an exception, see Kelman and Friedman 2009). So, how could managers use performance information to create organizational improvements? While the literature on performance management points to informed decision-making as the key to this process (e.g., Moynihan 2008; Taylor 2009; Vakkuri 2010), it offers little guidance on how data could be transformed into meaningful messages that guide organizational change. This study addresses this shortcoming by presenting two ways managers could utilize information: Either by reacting to signals of failure and initiating problem solving (Cyert and March 1963; Meier, Favero, and Zhu 2015), or by exploiting successes to develop expertise and specialize (March 1991). To provide insights into which strategy managers pursue, this study examines how performance on different organizational goals influences managers’ willingness to prioritize certain goals. While previous research has provided important insights into the ways to increase performance information use (Moynihan, Pandey, and Wright 2011a; Yang and Hsieh 2007), most studies rely on self-reports from managers to evaluate their use. By testing the influence of information on a tangible decision outcome, this study offers one of the first accounts of performance information use in actual decision-making. A key argument in the article is that prioritizations do not only concern rationales for improvements, signaling effects to the political environment also play a part (Bourdeaux and Chikoto 2008; Lavertu, Lewis, and Moynihan 2013). The study extends previous work in this area by testing the idea that the strong emphasis on failures (i.e., a negativity bias) in the political environment (Hood 2011; Nielsen and Moynihan 2017; Olsen 2017) encourages managers to prioritize goals with low performance. The distinction between internal and external rationales is not only important for understanding a manager’s choice of priorities; it also has implications for the outcome of this decision. Priorities that are chosen solely for their signaling value to political principals might end up being nothing more than symbolic (Feldman and March 1981). However, even when the decision initiates functional organizational change, success is no guarantee. I argue that prioritizations influence outcomes with a diminishing return, meaning that they primarily lead to performance increases for the most problematic goals. In this way, managers’ tendency to focus on performance failures may actually have positive attributes, directing scarce attention to goals with the highest potential for improvement. Other than being one of the first studies to address these questions, the article contributes to the literature on performance management in three general ways. First, I provide an empirical example of how managers engage in functional performance management practices, namely by using performance information to solve problems. The context for this finding is Danish public schools where principals prioritize the educational goals (for instance grades, truancy, wellbeing, parental satisfaction, and student overweight) where their school is performing the worst. After a goal have been made a priority for several years, the performance level increases. While it is not possible to attribute these improvements directly to the use of information, the analysis shows that performance primarily increases for goals with a low initial performance level, thereby indicating that the underlying mechanism is managerial problem solving. On this basis, the study provides an important perspective to the many studies that tell a pessimistic story on performance management in the public sector (e.g., Andersen 2008; Gerrish 2016; Heinrich 2010). In addition to the empirical account, the article contributes theoretically to the debate on how managers could use performance information to make decisions that improve organizational performance (Askim, Johnsen, and Christophersen 2008; Kroll 2015a; Moynihan 2009). More specifically, I develop a comprehensive theoretical framework that theorizes on the rationales for utilizing performance information in strategic decision-making, but that also considers the pitfalls when these decisions unfold as attempts to improve performance. Finally, a valid test of the research questions requires addressing the methodological challenges in the literature. To do this, I operationalize managers’ use of performance information in a novel way. Since the principals make an explicit prioritization, it is possible to couple their decisions with the performance level of each goal. This way of structuring data avoids the problem of common source bias (Meier and O’Toole 2013) and allows the use of a fixed-effect model and a difference-in-differences setup that can address the problem of omitted variable bias more exhaustingly than previous research. Purposeful Performance Information Use The theory consists of the following sections. The first section outlines the conceptual framework for understanding performance information use in strategic decision-making. In the next section, I present the internal and external managerial considerations for prioritizing goals with low- or high-performance levels. The following sections concern the effects of goal prioritizations on performance. Here, I consider situations with symbolic prioritizations, as well as instances where prioritizations initiate functional organization change. Expectations for the latter situation are developed by theorizing on the production process, including the possibility of diminishing returns, the importance of sustained attention, and performance trade-offs between goals. Strategic Performance Management This article focuses on managerial use of performance information in strategic decision-making, which broadly concerns a manager’s choice of strategies and actions to ensure goal achievement (Poister 2010). To theorize on this process, two shortcomings in the literature on performance management need consideration. The first relates to goal multiplicity. Public organizations often pursue multiple goals, and performance indicators at least partly reflect this multiplicity (Christensen et al. 2018; Moynihan et al. 2011b). However, studies of performance information use rarely consider this point, neither theoretically nor empirically. Performance is either treated as an abstract concept (in studies based on self-reports), or only one dimension of performance is examined (e.g., students’ final grades). Second, research on performance management needs a more precise conceptual understanding of what “use” entails. Previous research relies on self-reports from managers to solve this issue, thereby leaving what constitutes use to be a subjective matter. This study takes a more independent approach and conceptualizes use as the process in which performance information—consciously or unconsciously—influences managers and thereby contributes to a specific thought, a decision or action (Rich 1997, 15). In relation to these points, performance information use in strategic decision-making concerns the relationship between performance levels across organizational goals and a manager’s willingness to make certain goals a priority. Simply put, this means that managers can use the information to direct attention to goals with a (relatively) low-performance level or goals with a (relatively) high-performance level. However, there is also a possibility that managers do not let their prioritizations influence by the results at all. Such a tendency would reflect the many studies that find managerial reluctance to inform decisions and practices with performance information (Kroll 2015b; Moynihan and Lavertu 2012; Moynihan, Pandey, and Wright 2011a; Taylor 2009). Even though the studies do not examine use in relation to strategic prioritizations, the findings suggest that managers’ decisions in performance management systems often reflect other factors than the content of the information. Rationales for Using Performance Information to Prioritize Goals In the situation where managers actually consult performance information when deciding on organizational goals, where do they turn their attention? An important distinction in answering this question is between internal and external considerations (O’Toole and Meier 2011). The internal dimension concerns management in relation to processes that take place within the organization with the objective of creating performance improvements (Meier, Favero, and Zhu 2015). While external management could also concern improvements, this aspect of public management has a more intermediate objective, namely to maintain autonomy and support for the organization (Carpenter 2001). Internal Considerations: Problem-Solving and Specialization An obvious starting point for the internal consideration is the research tradition that originates from the behavioral model developed in the work by Cyert and March (1963). A key idea in this model is that managers improve performance through three sequential phases, namely (1) identification of problems, (2) search for solutions, and (3) implementation of the most appropriate solution (Cyert and March 1963, 34). The identification phase starts with a stimulus that indicates a problem and a need for managerial action (Mintzberg, Raisinghani, and Théorêt 1976). Performance information is a key component in this process, quantifying the results of activities and thereby easing the identification of problematic areas in the organization’s work. The main theoretical developments and empirical tests of the problem-solving model have taken place in the private sector (for a review, see Shinkle 2012). However, recently, a number of studies have used the framework to explain public managers’ response to performance data. For example, Salge (2011) finds that performance shortfalls lead public organizations to increase search for innovative solutions, and Nielsen (2014b) and Rutherford and Meier (2015) show that managers redirect attention to various organizational outcomes in response to performance declines and negative social feedback. While these studies have provided vital first steps in adapting the behavioral model to the public sector, their conceptualization of performance shortfalls does not capture the complexity in modern public management completely. The underlying assumption in the studies is that managers evaluate performance shortfalls and successes either along a single performance dimension or in serial processing between multiple dimensions, comparing absolute results to an aspiration level. Thus, the performance hierarchy between goals is not taken into consideration, which leaves out an important aspect of how managers form opinions about what constitutes low and high performance. On a theoretical level, the idea of a problem-solving manager has also been subject to discussion. On the one side is the line of research that draws on the psychological literature on self-affirmation and points to self-enhancement as a motive when managers process performance information (Jordan and Audia 2012). Empirical support for this idea is found in the studies that show that managers adapt various strategies to cope with performance shortfalls, for instance, by choosing social and historical reference points that portrait performance in a positive light (Audia, Brion, and Greve 2015) or shifting focus to favorable performance indicators (Audia and Brion 2007). The last finding is particularly interesting as it connects to another research tradition that focuses on how managers create improvements from successes (e.g., Baum and Dahlin 2007; Diwas, Staats, and Gino 2013). A key point in this literature is that an organization could build competence and gain improvement through exploitation and utilization of tacit knowledge about functional organizational activities (March 1991). Knowledge of such activities could be employed in two different ways to create further improvements. One way is to dismantle information on the functional activities throughout the organization in an attempt to create best-practice learning (Askim, Johnsen, and Christophersen 2008). Another way is to strengthen the work on the successful goal dimension. Here, the knowledge would enter as competence building in an effort to specialize activities, for example, by reorganizing resources to enhance the practices that have proven functional. External Considerations: Buffering and Reputation Building Managerial reactions to performance information do not take place in an environmental vacuum; rather, they are a vital part of hierarchical accountability relations. This point is evident in many studies that show that an engaging political environment significantly increases the likelihood of managers responding to the information (Bourdeaux and Chikoto 2008; Lavertu, Lewis, and Moynihan 2013). While the findings suggest that managers pay attention to their environment in performance information use, there is little theoretical basis for understanding how political considerations influence the decisions made in the process. One important thing to note about the decision to prioritize certain organizational goals is the signaling effect to the environment (Spence 2002). This signaling effect occurs is particularly relevant when priorities have a clear connection to a performance result (Poister 2010). For goals with low performance, a prioritization signalizes responsiveness to organizational problems; for goals with high performance, a prioritization signalizes competence building. Both signals could be used strategically by managers in their quest for political support and autonomy. Prioritization of goals with low performance is a managerial strategy to buffer the organization from political interventions (Meier and O’Toole 2008). Politicians have a strong electorate incentive to avoid blame for poor performance results (Krause 2003), which could lead them to disclaim responsibility and instead attribute control over outcomes to managers (Nielsen and Moynihan 2017). In addition, low performance could offset a political reaction, for instance, in terms reduced autonomy (MacDonald and Franko 2007) and funding (Gilmour and Lewis 2006). Managers can use their prioritizations to avoid such a reaction by signalizing to the political environment that the problem is recognized and action is being taken. Prioritization of goals with high performance is a strategy to improve the organization’s reputation. Since a reputation describes the stakeholders’ perceptions of the unique capacities, roles, and obligations of the organization, it is a valuable political asset (Carpenter 2010). For this reason, managers spend time creating, maintaining, and protecting their organization’s reputation (Maor, Gilad, and Bloom 2013). One way to achieve this objective is to highlight the organization’s contribution to the public good by appearing successful and distinct from other organizations (Busuick and Lodge 2017). A prioritization of a goal with high performance can serve to fulfill this purpose because the signal should draw political attention to well-functioning areas of the organization’s work, thereby highlighting the organization’s impact on society. How Do Managers Weight the Different Considerations? The previous sections have identified both internal and external rationales for how managers could integrate performance information in their prioritizations of organizational goals. So, how should we expect managers to weight these different considerations? One aspect that could factor into this decision is accountability relations. In this regard, an important feature of political environment is a negativity bias, which creates an environment where politicians and citizens evaluate performance information with a strong emphasis on failures (Hood 2011; Nielsen and Moynihan 2017; Olsen 2017). Because of these tendencies, managers might perceive the problem solving and buffering strategies as particularly beneficial to them. This expectation is tested in the first hypothesis. H1 Public managers will react to performance information on multiple goals by prioritizing goals with the lowest performance level. Effects of Goal Prioritization Symbolic Decisions The distinction between external and internal rationales for prioritizing certain organizational goals is particularly important when considering the potential performance effects of the decision. The reason is that prioritizations based solely on external considerations risk being symbolic use of performance information (Feldman and March 1981). Symbolic use occurs in a situation where managers decide in a way that reflects the content of the information, but they do nothing more than making the decision publicly available. Such perverse engagement in performance management systems has been found in a number of studies, which show that public servants perceive the systems as irrelevant for their work (Lavertu, Lewis, and Moynihan 2013; Soss, Fording, and Schram 2011) and as a structural obstacle to game (Bevan and Hood 2006). If this type of behavior also characterizes managers’ goal prioritizations, we should not expect any organizational change and therefore no performance improvements. This scenario is tested in hypothesis 2. H2 Public managers’ prioritization of goals with low performance is symbolic and does not lead to performance improvements. Performance Potentials and Diminishing Returns While the hypothesis above expects managers to engage in performance management practices with a minimum of effort, a prioritization could also entail a real effort to create performance improvements. In this situation, a prioritization initiates a process that adds resources to the particular goal dimension. The resource allocation should be understood broadly, encompassing both material resources (for instance, new equipment, better facilities, or supporting staff) and activation of mental resources (for instance, motivation and abilities). No matter the specific content of the prioritization, the manifestation should follow the internal management rationales for making the prioritization. This means that prioritizations of goals with low performance start a process where the organization uses resources to analyze the reasons for the shortcomings and correct the problems (Cyert and March 1963). For goals with high performance, resources add to a process where successes are scrutinized to create knowledge that builds competences and commence specialization efforts. One thing to note about both processes is the risk that the investment does not yield performance payoff. Research on both financial (Andersen and Mortensen 2010; Hanushek 2003) and human resources (e.g., Banerjee et al. 2012) have consistently shown that the mere addition of resources to achieve a goal does not equal success. Instead, there are good reasons to believe the marginal return of resources should be taken into consideration. To understand why we need to consult the production function that describes how resources transform into outcomes on a goal dimension. This relationship is often characterized using the Cobb–Douglas production function. Here, resources have a diminishing return, meaning that addition of resources to the production increases outputs at a diminishing rate. The reason is that the benefits of the capital stock are being exploited as production increases, thereby lowering the product of each additional worker. Because of this exploration, the marginal return of resources is highest when few resources are used in the production and the corresponding output level is low. Although resources’ diminishing return typically is used to explain how companies can maximize their production of material outputs, the idea has also been applied to describe production processes in the public sector, concerning service delivery and the creation of humane outcomes (Heckman 2000). The theoretical argument for making this application is that production processes in public organizations also risk exploration of the capital stock, thereby resembling those in the private sector (Pritchett and Filmer 1999, 223–4). Substantially, the argument provides an explanation for why resources do not always yield a performance payoff. The reason is that the marginal product trends toward zero, or even becomes negative, at high production levels because the many resources cause inefficiency. While the argument for resources’ diminishing return focuses on the input side of a production, the equation could also be turned around, implying that output or outcome levels could be used to assess a production’s potential for improvement. Following this idea, goals with low performance have a great potential because the production is at a low level and accordingly, the marginal product of adding resources should be high. This expectation is tested in the following hypothesis. H3 Public managers’ prioritization of goals will improve performance, primarily for goals with an initial low level of performance. The Importance of Sustained Attention A standard way to define performance management is as an ongoing system that is bound by performance information creation, decision-making, and outcomes (Moynihan 2008). A system approach, by definition, includes a temporal component, as repeated measurements produce new information over time (van Helden, Johnsen, and Vakkuri 2012). However, this temporal aspect is a rather unexplored aspect of performance management, especially in relation to managerial decision-making. The cyclic aspect of performance systems means that an initial decision (in the first cycle) can be revisited in later cycles (Pollitt 2013). In the situation where managers prioritize resources, they face a choice in the second cycle of either reprioritizing the same goals as in the first cycle or choosing to prioritize a new set of goals. To understand the implications of this decision, it is necessary to open the black box, which has been labeled the “production process” so far. Here, a relevant point of departure is the classical input-output model (Boyne 2002). In this model, organizational activities are the key mechanism that transforms resources into performance (Boyne 2003). A prioritization could aim at improving existing activities or developing new activities. An example is the principle that prioritizes the goal “student health.” At the school, a certain amount of resources is already used on student health, such as the provision of gym classes and green areas with playgrounds. When the goal is made a priority, the principal can increase the number of gym classes or develop a new initiative. Neither solutions quick fixes because they cannot be implemented and provide a return within a day, week, or month. In this way, the example highlights two general traits of production processes in public organizations, namely that (1) they consist of a long chain of complex subprocesses and (2) there could be a substantial time lag before addition of input to a goal actually has an influence on the outcome (Ostrom et al. 1978). When these considerations are related to the cyclic aspect of performance management, it becomes evident why managers are facing an important decision in later cycles of the system. In a situation where a goal is only prioritized in one cycle, it is unlikely that the prioritization will lead to improvements, simply because the input to production processes lack the sufficient time to transform into improvements. If the manager chooses to prioritize a completely new set of goals, there is a risk that the initial prioritization will only have superficial attention without any real improvement as a result. This argument is tested in hypothesis 4. H4 A long-term strategy (repeated prioritizations of a goal) increases the likelihood of performance improvements. Performance Trade-Offs An important aspect of hypothesis 2.1 and 2.2 is what happens to the performance level for nonprioritized goals. Part of the answer to this question depends on the level of available resources in the organization. In one scenario, the organization finds itself in a situation where resources are used for activities that do not benefit the users (i.e., slack) (Migué, Bélange and Niskanen 1974). While such a situation could be the result of a deliberate managerial choice, improper use of resources could also originate at lower levels of the organization. For this reason, a problem-solving approach often starts with a managerial search for available resources that can be used to identify and implement solutions (Salge 2011). If this search ends successfully, goals with low performance could be prioritized with no consequence for activities and performance levels on nonprioritized goals. In the other scenario, all resources are used to produce outcomes on the various goal dimensions, meaning that the allocation of resources to one goal reduces the amount of inputs available to produce other goals. Performance trade-offs are likely to occur in such a situation. However, they do not necessarily cancel each other out. A production-possibility frontier illustrates why by plotting the possible outcome combinations in the production of two goals. The frontier connects the goals in a concave relationship, reflecting the increasing opportunity cost of moving resources from one goal to the other. Substantially, the relationship reflects the fact that the most general resources are relocated at first, while specialized and less effective resources are being relocated in the movement along the frontier. Figure 1 shows a frontier for two goals, one with a high-performance level (y2) and one with a low-performance level (y1) in the starting point A. When resources are allocated to the goal with low performance, the organization moves to scenario B. Δ y2A – y2B expresses the performance loss from this reallocation, and Δ y1A – y1B expresses the performance gain. As seen in the figure, the concave relationship leads the allocation of resources to y1 to exceed the performance loss for y2. Figure 1. View largeDownload slide The Production-Possibility Frontier for Goals with High and Low Performance Figure 1. View largeDownload slide The Production-Possibility Frontier for Goals with High and Low Performance In sum, the two scenarios lead to an identical expectation for how prioritizing resources to goals with low performance will influence the organization’s performance level overall, namely that the reallocation will end in a surplus. This expectation is tested in hypothesis 5. H5 The overall performance changes from prioritizing goals with low performance will end in a surplus for the organization. Research Design and Data This section presents the main elements of the research design, namely the empirical setting, data, operationalization, identification strategy, and model specifications. Setting and Data: Performance Management in Danish Public Schools During the last 20 years, a number of reforms have sought to change the Danish public schools in line with the ideas of new public management (Nielsen 2014a). In the education system, the municipalities oversee the quality of the public schools and accordingly, the implementation of the initiatives was organized here. This flexibility created distinctive performance management regimes across municipalities, varying considerably in their expression and scope. In this study, the data comes from schools in a municipality with one of the most comprehensive systems. The system consists of two-year cycles. A cycle begins with the measurement of school performance on a large number of goal dimensions. Then, the schools receive a quality report that summarizes their achievements. After the release of the quality report, the principal meets with the stakeholders of the school (representatives of the parents, municipality, and teachers) to discuss the results. The meeting is organized as a learning forum (Moynihan 2005) in which each participant can express his or her interpretation of the results. At the end of the meeting, the principal decides on a number of priorities for the school. These priorities are areas of the school’s work that should be given special attention until the beginning of the next cycle. A school has from 12 to 18 months to work on the priorities before the measurement of performance is repeated. Figure 2 illustrates the cyclic nature of the system. Figure 2. View largeDownload slide The Cyclic Nature of the System Figure 2. View largeDownload slide The Cyclic Nature of the System The system has three attractive features that make it suitable for testing the hypotheses. First, the large number of performance indicators creates a situation where principals receive multiple information signals on how their school performs. Second, the principals’ decisions have a manifest expression because the priorities are written down and made publicly available. These features make it possible to connect information and decisions, thereby enabling a test of how managers react to differences in performance levels. Finally, and as shown in table 1 below, the schools in a municipality are—on average—quite similar to the rest of Denmark on a number of parameters. These parameters are school size, budget per student, the share of students in special classes, teaching time, student/teacher ratio, and competence-coverage. Thus, any functional aspect of performance management in this setting is more likely to be caused by the system than the characteristics of the schools. Table 1. Characteristics of the Municipality—The School Year 2012/2013 Characteristics Municipality Denmark Min Max Mean Mean School size (number of students) 124 1,021 602 428 Budget pr. student (in DKK) — — 62,746 61,065 Share of students in special classes 5% 7% 4.3% 5.2% Share of bilingual students 2.7% 99.7% 25.0% ~10% Share of teachers’ time used for teaching 25.2% 38.7% 32.3% 33.7% Student/teacher ratio 9 18 14 13 Competence-coveragea 33% 100% 70.0% 80.0% Characteristics Municipality Denmark Min Max Mean Mean School size (number of students) 124 1,021 602 428 Budget pr. student (in DKK) — — 62,746 61,065 Share of students in special classes 5% 7% 4.3% 5.2% Share of bilingual students 2.7% 99.7% 25.0% ~10% Share of teachers’ time used for teaching 25.2% 38.7% 32.3% 33.7% Student/teacher ratio 9 18 14 13 Competence-coveragea 33% 100% 70.0% 80.0% Note: The table is based on data from the quality reports and the Danish ministry of education. aThe share of classes being conducted by a teacher with a specialization in the subject. View Large Table 1. Characteristics of the Municipality—The School Year 2012/2013 Characteristics Municipality Denmark Min Max Mean Mean School size (number of students) 124 1,021 602 428 Budget pr. student (in DKK) — — 62,746 61,065 Share of students in special classes 5% 7% 4.3% 5.2% Share of bilingual students 2.7% 99.7% 25.0% ~10% Share of teachers’ time used for teaching 25.2% 38.7% 32.3% 33.7% Student/teacher ratio 9 18 14 13 Competence-coveragea 33% 100% 70.0% 80.0% Characteristics Municipality Denmark Min Max Mean Mean School size (number of students) 124 1,021 602 428 Budget pr. student (in DKK) — — 62,746 61,065 Share of students in special classes 5% 7% 4.3% 5.2% Share of bilingual students 2.7% 99.7% 25.0% ~10% Share of teachers’ time used for teaching 25.2% 38.7% 32.3% 33.7% Student/teacher ratio 9 18 14 13 Competence-coveragea 33% 100% 70.0% 80.0% Note: The table is based on data from the quality reports and the Danish ministry of education. aThe share of classes being conducted by a teacher with a specialization in the subject. View Large The data covers three cycles that begin in 2009, 2011, and 2013, respectively. The unit of analysis is goals with available performance information. Since each school is measured by a number of goals, the data have a multilevel structure with goals nested within a school. Over the years, the system has increased the number of performance indicators (and goals) from 33 in 2009 to 40 in 2013.1 There are between 51 and 48 schools in the municipality in the time span from 2009 to 2013, and therefore, the number of observations varies from 1500 to 1900. Operationalization A Goal’s Relative Performance In line with the theoretical expectations, when managers face variation in performance information across different organizational goals, it should lead to decisions that give priority to the goals with the lowest level of performance. Accordingly, a key feature in the empirical test of this theoretical argument is a variable that contains information on how well a school is performing across the various educational goals. This standardized measure of relative performance is created in the following way. First, the maximum level of performance for the goal is determined. This maximum is either 12 on the grading scale or 0/100 on a percentage scale. For example, for the goal of “parental satisfaction,” the maximum is 100%, while it is 0% for the goal of “student smoking.” With the maximum level of performance in place, the following formula calculates the relative performance of a goal: pist = pabistpmaxi×100 where pist is the measure of relative performance on goal i for school s in time t. pabist and pmaxi are measures of the school’s absolute level of performance and the maximum possible level of performance on the goal, respectively. Thus, higher values on the variable indicate better performance. An alternative way to create the variable for relative performance is to use the empirical maximum value in the sample instead of the theoretical maximum. This variable has been used as a robustness check to the results from model 4 in table A2. Results from the alternative specification are almost identical (the coefficient is −0.0066 and the t-statistic is −6.48). Goal Prioritization The variable prioritization is a dichotomous variable that contains information on whether a principal (for a given school at a given time) has chosen to prioritize a goal or not. The variable was created by matching a school’s priorities—those selected by the principal after meeting with stakeholders—with the performance indicators. A goal was coded 1 if there was a match to one (or more) of the school’s priorities and the value 0 if none of the priorities related to the goal. Since priorities are chosen in an open process, there are no specific demands on the number of priorities chosen, how specific they should be, or how closely they should relate to metrics. This way of organizing the process has two implications for the operationalization. First, some priorities do not match metrics. Instead, these priorities focus on themes such as technology, community collaboration, or the school’s economy. Second, there is an element of interpretation in the matching of priorities with the relevant goals, and therefore, there was a need for creating coding rules that ensure that the matching occurred in a valid and reliable manner. Some priorities were easily connected to a goal. For example, “Improve reading in pre-preparatory classes” matches the goal “Reading test: third grade.” Other priorities were not directly related to one specific goal but, rather, to a theme (such as learning, parental collaboration, student well-being, or health) with many underlying goals. Finally, some priorities were indirectly connected to either one or several goals. In the coding of these priorities, it was necessary to include additional information that could help uncover the meaning of the priority. For the majority of schools, this information was available in the school’s strategy plan and in the following quality report in which the school describes and reflects on the work with the priorities. Since it was rather consistent what principals meant by their indirect priorities, this meaning was used to code schools with indirect priorities but with no information available. Lastly, three points should be mentioned. First, even though the aim of the coding strategy was to connect priorities with intended goals, there will be instances where the match does not reflect the principal’s intention. Since there is no reason to expect mismatches to be systematic, the problem is a matter of noise in the measurement of the variable, which may have the disadvantage that it makes it more difficult to detect relationships in the estimations. Second, in relation to the validity of the variable, the coding has been reliability tested by having a research assistant code 10% of the goals. A comparison of the two codings gives a Krippendorffs alpha of 0.83, which is considered reliable. Third is the question of what kind of organizational changes a prioritization entails. Empirically, principals add resources (through their choice of strategy) to a goal in many different ways. For example, a strategy for improving parental satisfaction could be a newsletter, increasing the number of parent–teacher meetings, or by highlighting the importance of a good relationship with the parents. Thus, a prioritization could initiate change by providing material resources or by activating mental ones. Principals describe their strategy in a development plan (which is a document with the overall strategy for the school), and superiors in the administration approve the plan. Thus, while it is not possible to create a measure of how many and what kind of resources is entailed in a prioritization, the accountability mechanisms should ensure that improvements come about through functional processes. Identification Strategies and Model Specifications The Influence of Relative Performance on Principals’ Prioritizations Estimating the relationship between relative performance and principals’ prioritizations risks endogeneity problems because factors related to the school or the principal could influence both variables. To handle this problem, I use a fixed-effects model. Formally, the model is estimated as follows: Prist= β0+ β1Pist+β2'Xist+ γs+ϕt+ εist (1) Prist is the variable that indicates whether goal i is prioritized or not at school s at time t1. Pist is the measure of relative performance on a goal i for a school s in time t0. Xist is a vector of control variables, γs is school fixed effects, ϕt is time fixed effects, β0 is the constant, and εist is the error term. In the model, time-invariant factors at the school level are handled by including the fixed-effects term for the schools (Hesketh and Skrondal 2012, 95–7, 159). However, the model still needs to include factors at the goal level that could influence performance levels and the probability of a prioritization (included in the vector Xist). Two of these control variables relate to fundamental characteristics of a goal, namely, the number of years with a metric and an indicator variable that shows whether performance is measured through a survey. Furthermore, the political attention of a goal could influence principals’ prioritizations of resources and the daily management of the school (thereby influencing performance). To account for such attention, the model includes two indicator variables, measuring whether there is a political aspiration level (decided by the city council) on a goal and whether this aspiration level has been changed throughout the three cycles. Finally, the model includes five dummies for goals that have been given special political attention at the national level (grade point average [GPA], transition to further education, truancy, inclusion, and bullying). Even though the fixed-effects model makes it possible to handle part of the omitted variable problem, it cannot account for a problem related to the cyclic nature of the data. Within one cycle, there is a sequential relationship between performance information on relative performance (t0), prioritizations (t1/4), and relative performance (t1). To address this issue, the main analysis is supplemented with analysis split by cycle (in table A2), which allows an estimation of the effect of relative performance on principals’ prioritizations independent of the possible performance effect these prioritizations might have. Another possible confounding factor is principal turnover. Principal turnover could confound the estimates if the change is somehow systematically related to both relative performance and goal prioritization. To account for this problem, the analysis has been robustness checked by splitting the sample for schools with and without a principal turnover. This analysis yields substantial identical results to the ones presented in model 4 in table A2 (the coefficient and t-statistic are −0.0037 and −4.62 for schools with no turnover and −0.0069 and −5.59 for schools with turnover).2 Equation (1) has a binary outcome, which makes the models appropriate for both a linear probability model and a logit model (Wooldridge 2009, 246–50). Because the model includes school fixed effects, a substantial number of observations drops out of the estimation if there is no variation in the dependent variable within schools. Therefore, the models are estimated with a linear probability model.3 All models are estimated with cluster-robust standard errors to account for heteroscedasticity in the error term (Angrist and Pischke 2009, 47). The Effect of Goal Prioritizations on Performance The model that estimates whether prioritized goals influence performance in subsequent years needs to address two issues. First, different factors might influence both a goal’s likelihood of being prioritized and the performance level. Second, initial performance differences between nonprioritized and prioritized goals (as expected in H1) would blur the estimation in an ordinary regression model that does not properly capture changes over time. To handle these issues, I use a difference-in-differences model. The difference-in-differences setup compares developments in the performance trends for nonprioritized and prioritized goals, thereby taking the initial performance differences between the two types of goals into account. In line with the difference-in-differences design, nonprioritized goals act as a control group that captures time trends, and prioritized goals receive a treatment in the form of a prioritization. Formally, the model is given by: Pist= β0+β1Tt+δ(Prist x Tt) +β2'Xist+ γis+ εist (2) Pist is the outcome, which is the relative performance level for goal i at school s at time t. Prist is the variable that indicates whether a goal is prioritized (i.e., the treatment variable), Tt is the time measure, and Xist is a vector of control variables.4 This vector includes two indicator variables. The first indicator variable takes the value 1 if a political aspiration level was introduced on the performance dimension, the other takes the value 1 if there was a change in the aspiration level from the previous cycle. γis is goal fixed effects (that also capture time-invariant factors at the school level), β0 the constant, and εist is the error term. δ is the difference-in-differences estimate, consisting of the following terms in a two-period setup: δ=(Pprioritized, t1−Pprioritized, t0)−(Pnonprioritized, t1−Pnonprioritized, t0) The basic model is expanded in two ways. First, treatment is normally given at one point in time (Angrist and Pischke 2009). However, the data in this study allow for repeated treatment because the performance management system entails two decision moments. Figure 3 illustrates this point. Figure 3. View largeDownload slide Time Sequence in the Difference-in-Differences Setup Figure 3. View largeDownload slide Time Sequence in the Difference-in-Differences Setup As shown in the figure, a goal could be treated in either late 2009, late 2011, or in both moments. This feature of the data is handled by creating two treatment groups, namely, goals with prioritization in one cycle (in either 2009 or 2011) and goals with prioritizations in two cycles. The two treatment groups are not collapsed in the first cycle because at the first decision moment principals might decide on a long-term strategy for the goals that are prioritized in two cycles, meaning that these goals experience a different treatment and a unique performance path. Second, the design rests on the assumption that the control group and the treatment group share a common trend before the introduction of the treatment (Wooldridge 2009, 453–4). To test this assumption, the data must include at least two pretreatment data points for the outcome variable. Thus, the time variable should consist of at least three categories, namely, two pretreatment categories and one post-treatment category (Bertrand, Duflo, and Mullainathan 2004). To accommodate this requirement, I include outcome data for 2007, meaning that t in equation (2) = {2007; 2009; 2010; 2013}.5 In a classic difference-in-differences design, the pretreatment values reflect a situation where all units of analysis are untreated. Such a scenario does not hold in the present setting as performance management reforms were introduced in the Danish educational system already in the beginning of the 2000s, and accordingly, the schools could have worked with goal priorities before the first decision moment in 2009. This aspect of the setting poses a challenge to the design on two levels. First, it could induce bias in the estimations if a systematic component characterizes pre-2009 prioritizations. The most concerning scenario is if the group of goals that were prioritized in 2009 includes a higher share of pre-2009 prioritized goals. Such a situation would make it difficult to estimate treatment effects after 2009 because the prioritized goals were already on a positive trend (due to the pre-2009 prioritization). One way to assess the problem is the test for common trends as this test indicates whether there is a systematic difference in performance trends between prioritized and nonprioritized goals leading up to the first decision moment in 2009. As shown across the models in table A4, the common trend test does not point to any concerns, showing no significant pretreatment differences. To supplement this test, I have run two additional robustness checks. A number of goals did not have a performance indicator before 2009, and thus had a low probability of being prioritized before this year. By running the analysis solely on these goals, it is possible to obtain an estimate that is less influenced by the decisions that took place before 2009. Although effects are somewhat larger (compared to model 3 in table A4, the effect is 7.94 and the t-statistic 5.09) for goals with no data before 2009, the overall conclusion is the same. The second robustness check is a generalized difference-in-differences model.6 This model includes unit and time fixed effects and estimates how changes in treatment status (i.e., whether the goal is prioritized or not) within a unit influence outcomes over time. Since this model does not rely on variation across goals that were either prioritized or nonprioritized in 2009, it is not sensitive to systematic differences between these groups before 2009. The alternative difference-in-differences specification yields substantial identical results to the ones presented in model 3 table A4 with an effect of 4.13 (t-statistic 5.51). While the three robustness checks do not point to concerns with regard to the estimates, the pretreatment issue is still relevant when interpreting treatment effects. The reason is that effects could partly reflect the work that took place before 2009, which substantially means that positive performance increases are the result of more than one or two prioritizations. Naturally, this point is worth keeping in mind in the analysis. Finally, three aspects of the model specification and analytic strategy should be noted. First, the analysis consists of three main models (shown in table A4 in the appendix). In each model, nonprioritized goals act as the control group (i.e., the reference category). Models 1 and 3 test the effect of prioritizing a goal in one cycle (2009 and 2011, respectively), and model 2 tests the effect of repeated prioritizations. Second, in addition to these models, the analysis includes two models that are split by goals within a school that have either low or high performance (divided by the median). Third, the difference-in-differences models are estimated with the full set of observations. This model is robust to a completely balanced panel-model that only includes goals with data from all four time periods (compared to model 3 in table A4, the effect is 3.54 and t-statistic 2.85). Analysis Figure 4 illustrates the relationship between relative performance and principals’ tendency to prioritize a goal. The β1 in the figure is the estimated regression coefficient from the model specified in equation (1). Figure 4. View largeDownload slide How Relative Performance Influences Principals’ Prioritization of Goals Figure 4. View largeDownload slide How Relative Performance Influences Principals’ Prioritization of Goals Based on figure 4, it is possible to evaluate the expectation in H1 that public managers will use performance information to prioritize goals with the lowest level of performance. As evident in the figure, this expectation is confirmed. β1 is significantly negative, which means that goals with lower performance are more likely to be prioritized. In substantial terms, a performance difference of 10 percentage points between two goals will lead to a 5% increase in the probability that a principal prioritizes the goal with the lowest level of performance. For a standardized outcome, the effect is 0.17 standard deviations (SDs). This estimate underlines a substantial important aspect of the analysis, namely that even though relative performance influences principals’ prioritizations, the performance differences need to be quite substantial before the information offsets a managerial reaction. Such differences are rather common for the principals as they, on average, experience a difference of 63 percentage points between the highest and lowest performing goals in their school. Across such large performance differences, the probability of prioritizing the goals with low performance is 40% higher. The last part of the analysis concerns the influence of priorities on performance in subsequent years. First is the question of whether prioritizations are merely symbolic (H2), or they actually create improvements but primarily for goals with a low initial performance level (H3). The analysis also includes a test of H4, which claims that a long-term strategy (consisting of repeated prioritizations) increases the probability of performance improvements. These hypotheses are tested in difference-in-differences models, and the results are summarized in figure 5 (the full models are shown in table A4 in the appendix). Figure 5. View largeDownload slide How Prioritization of Goals Influences Performance in Subsequent Years Figure 5. View largeDownload slide How Prioritization of Goals Influences Performance in Subsequent Years The results confirm the expectations in H3 and H4. As evident by the figure, goals that have been prioritized in two cycles experience a performance increase in the second cycle. The effect is around four scale points on the performance scale. In relation to H3, the analysis split for goals with low- and high-performance levels confirms the expectation that improvements occur for goals with a low initial level of performance (the result is statistically significant, p < .01, in a three-way interaction). Figure 5 also shows that short-term prioritizations do not create improvements. Goals that are prioritized in one cycle experience a small (insignificant) increase in performance (0.56 for goals that are prioritized in 2009 and 0.51 for goals that are prioritized in 2011) in the first cycle; however, they do not gain the most substantial performance increase because they are not reprioritized. At first sight, the effect of around four scale points might not seem overwhelming in substantial terms as the scale of relative performance ranges from 19 to 100. However, when standardizing the coefficient, the effect of 0.25 SDs presents itself as quite substantial. In relation to other studies that focus on interventions in public schools, Rockoff et al. (2012) find that providing principals with reports on teachers’ performance increases students’ math achievements by 0.053 SDs. Andersen, Humlum, and Nandrup (2016) show that increasing instruction time in schools improves students’ reading by 0.15 SDs. In relation to these findings, it is worth noting that the effect covers different strategies for creating improvements. In some instances, the schools have made major changes in their organization, facilities, and curriculum. However, improvements also reflect minor initiatives that aim specifically at the particular goal, for example, to create a newsletter to the parents. The last part of the analysis concerns H5 and the expectation that a prioritization of goals with low performance will end in a performance surplus for the organization over the years. Performance trends for nonprioritized goals are estimated by the time variable (i.e., the effect of time when treatment is at the reference category) in models 1, 2, and 3 in table A4. H5 is tested by summarizing these coefficients and comparing them to the treatment effects. When using 2009 as the reference point, the performance for nonprioritized goals increases over time, meaning that the results confirm the expectation that principals are able to improve performance on goals with low performance without substantial performance trade-offs. Conclusion This article examines how the performance of various organizational goals influence managers’ tendency to prioritize the goals and whether the prioritization of certain goals improves performance in subsequent years. The results show that principals in a municipality in Denmark prioritize the educational goals where their schools are performing the worst. However, the information has the most profound influence on prioritizations when there is a substantial discrepancy between goals with low and high performance. After a goal has been made a priority for several years, the performance level increases. As nonprioritized goals do not experience a drop in performance, the schools end in an overall performance surplus over the years, suggesting minimal performance trade-offs. These results are noteworthy in themselves, but they also have broader implications for our understanding of how performance management fits in the public sector. First, as the results show that public managers are able to engage in functional performance management practices, they naturally raise the question of what drives the purposeful use of performance information. Understanding when performance management efforts make a difference is essential to move beyond the simplistic claims that performance management does or does not work (Gerrish 2016). One apparent contextual factor in the present setting is the learning forums where managers discuss the information with stakeholders. While previous research suggests that such forums support performance information use (Moynihan 2005), there is not much knowledge of what makes them functional. The forums in this study are interesting because they strike a balance between outside involvement and managerial autonomy. The deliberation between stakeholders might help overcome some of the biases that have been shown in the individual processing of performance information (Nielsen and Moynihan 2017; Olsen 2017) while at the same time allowing managers to feel ownership of the process. This point also relates to a general trait of the system, namely, the balance between autonomy and coercion. One the one side, the system includes various checkpoints for principals, forcing them to make a decision and develop a strategy for improvements. However, they are also given autonomy in their choice of priorities and strategy. In this way, the findings point to an important balance between finding the right means to engage managers but leaving sufficient room for management (Nielsen 2014a). Finally, it is also worth mentioning that the context is just as interesting from the perspective of what is not present; there are no financial incentives at stake for the principals. Thus, the results suggest that it is possible to create a functional management process in the absence of incentives but with the presence of deliberative learning forums. A second contribution is the idea of managers as biased problem solvers. More specifically, I not only document that the level of performance affects managers’ decisions; I also document that such decisions still facilitate problem-solving by directing attention to goals where the greatest returns to investment may occur. The principals’ emphasis on negative information (low performance) is identified for one specific decision, namely, the prioritization of organizational resources and attention. It remains to be seen how managers react to the content of performance information in other settings, for example, when the information is used in the management of employees (Behn 2003). Third, an important result from this study is that performance gains require ongoing persistence. This result has two practical implications. First, there are no quick fixes in public organizations. However, a political environment may demand new initiatives each year and grow impatient if a strategy does not yield a fast payoff. Such rapid changes are likely to undercut long-term thinking and strategies that need time to bear fruit. Second, there is an inherent risk in performance management systems that attempts at improving performance are ruined by too little and too superficial attention. As the results show that one cycle (one and a half year) is probably not enough to improve performance, too frequent shifts in priorities will most likely result in wasted resources with no performance gains. If performance strategies are to pay off, they need to be deployed with a medium- and long-term perspective. I am grateful to Simon Calmar Andersen, Mads Leth Jakobsen, and Donald P. Moynihan, as well as three anonymous reviewers, for their helpful comments on drafts of this article. Earlier versions of the article were presented at the 20th International Research Society for Public Management Conference, April 2016, in Hong Kong and at the Public Management Research Conference, June 2016, in Aarhus. Footnotes 1 See table A1 in the appendix for a complete list of goals and descriptive statistics. 2 Another possible confounding factor is incrementalism in principals’ decisions (Lindblom 1959) such that decisions from the first cycle influence decisions in the following cycles. However, a lagged dependent variable in a fixed effects model is by nature endogenous and likely to cause biased estimates (Wooldridge 2002). Lagged dependent variable models are therefore used as a robustness check to the main model. The model does no point to any concern compared the results in model 4 in table A2 (the effect is −0.0047 and the t-statistic −6.92). 3 The results from a logit model are reported in table A3 in the appendix. 4 The influence of principal turnover has also been tested for this model. Compared to the effect in model 3 in table A4, the effect is 3.83 (t-statistic 3.57) for schools with no turnover and 4.51 (t-statistic 3.51) for schools with a turnover. 5 Performance data is not available for all goals in 2007. Data are available for nine goals that are related to learning in terms of grades and tests scores, parents’ perceptions of student wellbeing, and other aspects of parental satisfaction and collaboration with the school. 6 Formally described as Pit= β0+β1prioritizationit+ β2'Xist+ γi+ ϕt +εist Appendix Table A1. Descriptive Statistics for Goals Goal [min; max] Meana SDa Prioritizedb School GPA [19; 71] 54.81 8.63 0.35 School GPA—lowest fourth [59; 83] 70.73 5.37 0.54 School GPA—highest fourth [48; 98] 83.42 9.05 0.47 Test—math (3rd grade) [44; 100] 88.71 10.30 0.29 Test—reading (3rd grade) [71; 100] 94.23 5.7 0.35 Test—reading (8th grade) [38; 100] 90.71 8.75 0.35 Transition: 3 months [45; 100] 93.62 6.43 0.21 Transition: 15 months [30; 100] 82.61 14.84 0.21 Pc. satisfaction: transition [39; 84] 61.44 9.50 0.20 Sd. satisfaction: boredom [55; 96] 89.42 4.00 0.41 Pc. satisfaction: academic challenge [35; 100] 58.85 11.04 0.39 Sd. satisfaction: friends [79; 100] 96.35 4.44 0.16 Sd. satisfaction: co-determination [54; 89] 72.11 6.23 0.16 Sd. satisfaction: victim of bullying [63; 100] 93.98 4.22 0.21 Pc. satisfaction: ability to stop bullying [47;81] 69.26 8.98 0.21 Pc. satisfaction: fellowship [37; 84] 70.95 8.83 0.16 Pc. satisfaction: equal opportunities [35; 84] 51.79 10.54 0.24 Sd. satisfaction: like the class [47; 100] 94.50 8.11 0.23 Pc. satisfaction: bullying occurs [63; 97] 88.67 5.96 0.20 Truancy [53; 96] 90.03 7.81 0.21 Worrisome truancy [50; 100] 85.26 11.25 0.24 Sd. satisfaction: happiness in general [47; 100] 96.26 6.00 0.25 Sd. satisfaction: happy at the time [47; 100] 98.10 3.54 0.27 Sd. satisfaction: like the school [70; 100] 91.41 3.58 0.27 Sd. satisfaction: recognition [81; 100] 93.32 2.39 0.25 Pc. satisfaction: child’s well-being [70; 92] 83.70 4.60 0.24 Student exercise [27; 100] 75.83 22.61 0.29 Student overweight [51; 100] 84.76 7.30 0.30 Student smoking [67; 100] 95.65 4.35 0.14 Student drunkenness [64; 100] 92.29 5.60 0.16 Quality of teeth: filling [0; 79.8] 56.16 11.78 0.17 Quality of teeth: cavity [50; 100] 92.22 7.10 0.17 Pc. satisfaction: collaboration [52; 92] 75.86 6.45 0.52 Pc. satisfaction: own contribution [60; 88] 75.31 7.51 0.51 Pc. satisfaction: expectations [32; 69] 51.70 6.66 0.51 Pc. satisfaction: school satisfaction [51; 97] 77.47 8.39 0.51 Pc. satisfaction: youth center satisfaction [44; 98] 78.03 10.76 0.53 Pc. satisfaction: daily contact [43; 90] 65.73 8.73 0.51 Pc. satisfaction: involvement [23; 86] 49.15 10.19 0.51 Total 82.25 16.13 0.31 Goal [min; max] Meana SDa Prioritizedb School GPA [19; 71] 54.81 8.63 0.35 School GPA—lowest fourth [59; 83] 70.73 5.37 0.54 School GPA—highest fourth [48; 98] 83.42 9.05 0.47 Test—math (3rd grade) [44; 100] 88.71 10.30 0.29 Test—reading (3rd grade) [71; 100] 94.23 5.7 0.35 Test—reading (8th grade) [38; 100] 90.71 8.75 0.35 Transition: 3 months [45; 100] 93.62 6.43 0.21 Transition: 15 months [30; 100] 82.61 14.84 0.21 Pc. satisfaction: transition [39; 84] 61.44 9.50 0.20 Sd. satisfaction: boredom [55; 96] 89.42 4.00 0.41 Pc. satisfaction: academic challenge [35; 100] 58.85 11.04 0.39 Sd. satisfaction: friends [79; 100] 96.35 4.44 0.16 Sd. satisfaction: co-determination [54; 89] 72.11 6.23 0.16 Sd. satisfaction: victim of bullying [63; 100] 93.98 4.22 0.21 Pc. satisfaction: ability to stop bullying [47;81] 69.26 8.98 0.21 Pc. satisfaction: fellowship [37; 84] 70.95 8.83 0.16 Pc. satisfaction: equal opportunities [35; 84] 51.79 10.54 0.24 Sd. satisfaction: like the class [47; 100] 94.50 8.11 0.23 Pc. satisfaction: bullying occurs [63; 97] 88.67 5.96 0.20 Truancy [53; 96] 90.03 7.81 0.21 Worrisome truancy [50; 100] 85.26 11.25 0.24 Sd. satisfaction: happiness in general [47; 100] 96.26 6.00 0.25 Sd. satisfaction: happy at the time [47; 100] 98.10 3.54 0.27 Sd. satisfaction: like the school [70; 100] 91.41 3.58 0.27 Sd. satisfaction: recognition [81; 100] 93.32 2.39 0.25 Pc. satisfaction: child’s well-being [70; 92] 83.70 4.60 0.24 Student exercise [27; 100] 75.83 22.61 0.29 Student overweight [51; 100] 84.76 7.30 0.30 Student smoking [67; 100] 95.65 4.35 0.14 Student drunkenness [64; 100] 92.29 5.60 0.16 Quality of teeth: filling [0; 79.8] 56.16 11.78 0.17 Quality of teeth: cavity [50; 100] 92.22 7.10 0.17 Pc. satisfaction: collaboration [52; 92] 75.86 6.45 0.52 Pc. satisfaction: own contribution [60; 88] 75.31 7.51 0.51 Pc. satisfaction: expectations [32; 69] 51.70 6.66 0.51 Pc. satisfaction: school satisfaction [51; 97] 77.47 8.39 0.51 Pc. satisfaction: youth center satisfaction [44; 98] 78.03 10.76 0.53 Pc. satisfaction: daily contact [43; 90] 65.73 8.73 0.51 Pc. satisfaction: involvement [23; 86] 49.15 10.19 0.51 Total 82.25 16.13 0.31 aRelative performance. bThe relative number of prioritizations, across all years. cParental. dStudent. Table A1. Descriptive Statistics for Goals Goal [min; max] Meana SDa Prioritizedb School GPA [19; 71] 54.81 8.63 0.35 School GPA—lowest fourth [59; 83] 70.73 5.37 0.54 School GPA—highest fourth [48; 98] 83.42 9.05 0.47 Test—math (3rd grade) [44; 100] 88.71 10.30 0.29 Test—reading (3rd grade) [71; 100] 94.23 5.7 0.35 Test—reading (8th grade) [38; 100] 90.71 8.75 0.35 Transition: 3 months [45; 100] 93.62 6.43 0.21 Transition: 15 months [30; 100] 82.61 14.84 0.21 Pc. satisfaction: transition [39; 84] 61.44 9.50 0.20 Sd. satisfaction: boredom [55; 96] 89.42 4.00 0.41 Pc. satisfaction: academic challenge [35; 100] 58.85 11.04 0.39 Sd. satisfaction: friends [79; 100] 96.35 4.44 0.16 Sd. satisfaction: co-determination [54; 89] 72.11 6.23 0.16 Sd. satisfaction: victim of bullying [63; 100] 93.98 4.22 0.21 Pc. satisfaction: ability to stop bullying [47;81] 69.26 8.98 0.21 Pc. satisfaction: fellowship [37; 84] 70.95 8.83 0.16 Pc. satisfaction: equal opportunities [35; 84] 51.79 10.54 0.24 Sd. satisfaction: like the class [47; 100] 94.50 8.11 0.23 Pc. satisfaction: bullying occurs [63; 97] 88.67 5.96 0.20 Truancy [53; 96] 90.03 7.81 0.21 Worrisome truancy [50; 100] 85.26 11.25 0.24 Sd. satisfaction: happiness in general [47; 100] 96.26 6.00 0.25 Sd. satisfaction: happy at the time [47; 100] 98.10 3.54 0.27 Sd. satisfaction: like the school [70; 100] 91.41 3.58 0.27 Sd. satisfaction: recognition [81; 100] 93.32 2.39 0.25 Pc. satisfaction: child’s well-being [70; 92] 83.70 4.60 0.24 Student exercise [27; 100] 75.83 22.61 0.29 Student overweight [51; 100] 84.76 7.30 0.30 Student smoking [67; 100] 95.65 4.35 0.14 Student drunkenness [64; 100] 92.29 5.60 0.16 Quality of teeth: filling [0; 79.8] 56.16 11.78 0.17 Quality of teeth: cavity [50; 100] 92.22 7.10 0.17 Pc. satisfaction: collaboration [52; 92] 75.86 6.45 0.52 Pc. satisfaction: own contribution [60; 88] 75.31 7.51 0.51 Pc. satisfaction: expectations [32; 69] 51.70 6.66 0.51 Pc. satisfaction: school satisfaction [51; 97] 77.47 8.39 0.51 Pc. satisfaction: youth center satisfaction [44; 98] 78.03 10.76 0.53 Pc. satisfaction: daily contact [43; 90] 65.73 8.73 0.51 Pc. satisfaction: involvement [23; 86] 49.15 10.19 0.51 Total 82.25 16.13 0.31 Goal [min; max] Meana SDa Prioritizedb School GPA [19; 71] 54.81 8.63 0.35 School GPA—lowest fourth [59; 83] 70.73 5.37 0.54 School GPA—highest fourth [48; 98] 83.42 9.05 0.47 Test—math (3rd grade) [44; 100] 88.71 10.30 0.29 Test—reading (3rd grade) [71; 100] 94.23 5.7 0.35 Test—reading (8th grade) [38; 100] 90.71 8.75 0.35 Transition: 3 months [45; 100] 93.62 6.43 0.21 Transition: 15 months [30; 100] 82.61 14.84 0.21 Pc. satisfaction: transition [39; 84] 61.44 9.50 0.20 Sd. satisfaction: boredom [55; 96] 89.42 4.00 0.41 Pc. satisfaction: academic challenge [35; 100] 58.85 11.04 0.39 Sd. satisfaction: friends [79; 100] 96.35 4.44 0.16 Sd. satisfaction: co-determination [54; 89] 72.11 6.23 0.16 Sd. satisfaction: victim of bullying [63; 100] 93.98 4.22 0.21 Pc. satisfaction: ability to stop bullying [47;81] 69.26 8.98 0.21 Pc. satisfaction: fellowship [37; 84] 70.95 8.83 0.16 Pc. satisfaction: equal opportunities [35; 84] 51.79 10.54 0.24 Sd. satisfaction: like the class [47; 100] 94.50 8.11 0.23 Pc. satisfaction: bullying occurs [63; 97] 88.67 5.96 0.20 Truancy [53; 96] 90.03 7.81 0.21 Worrisome truancy [50; 100] 85.26 11.25 0.24 Sd. satisfaction: happiness in general [47; 100] 96.26 6.00 0.25 Sd. satisfaction: happy at the time [47; 100] 98.10 3.54 0.27 Sd. satisfaction: like the school [70; 100] 91.41 3.58 0.27 Sd. satisfaction: recognition [81; 100] 93.32 2.39 0.25 Pc. satisfaction: child’s well-being [70; 92] 83.70 4.60 0.24 Student exercise [27; 100] 75.83 22.61 0.29 Student overweight [51; 100] 84.76 7.30 0.30 Student smoking [67; 100] 95.65 4.35 0.14 Student drunkenness [64; 100] 92.29 5.60 0.16 Quality of teeth: filling [0; 79.8] 56.16 11.78 0.17 Quality of teeth: cavity [50; 100] 92.22 7.10 0.17 Pc. satisfaction: collaboration [52; 92] 75.86 6.45 0.52 Pc. satisfaction: own contribution [60; 88] 75.31 7.51 0.51 Pc. satisfaction: expectations [32; 69] 51.70 6.66 0.51 Pc. satisfaction: school satisfaction [51; 97] 77.47 8.39 0.51 Pc. satisfaction: youth center satisfaction [44; 98] 78.03 10.76 0.53 Pc. satisfaction: daily contact [43; 90] 65.73 8.73 0.51 Pc. satisfaction: involvement [23; 86] 49.15 10.19 0.51 Total 82.25 16.13 0.31 aRelative performance. bThe relative number of prioritizations, across all years. cParental. dStudent. Table A2. How Relative Performance Influences Principals’ Prioritization of Goals Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.0035* (−2.66) −0.0055*** (−4.92) −0.0043*** (−4.48) −0.0050*** (−7.06) Survey indicator 0.086* (2.17) −0.030 (−0.67) −0.12** (−3.18) −0.013 (−0.48) Number of years measured 0.030** (2.80) 0.026** (2.91) 0.012 (1.75) 0.029*** (5.29) Political aspiration level 0.0021 (0.09) −0.14* (−2.44) −0.12** (−3.01) −0.069* (−2.07) Change in aspiration level 0.028 (0.61) −0.029 (−0.82) −0.046* (−2.51) Goal: inclusion −0.22*** (−4.01) 0.073 (1.06) −0.039 (−0.77) −0.078* (−2.36) Goal: GPA −0.096 (−1.37) −0.18*** (−3.86) −0.057 (−0.91) −0.13** (−3.00) Goal: transition to further education −0.23*** (−4.19) −0.0082 (−0.09) −0.13 (−1.81) −0.14** (−3.16) Goal: bullying −0.33*** (−4.41) −0.12 (−1.49) −0.19*** (−3.82) −0.18*** (−5.20) Goal: truancy 0.039 (0.50) −0.21** (−3.12) −0.19* (−2.13) −0.10* (−2.04) Cycle: 2009 Reference Cycle: 2011 0.10** (2.80) Cycle: 2013 0.081 (1.94) Constant 0.43*** (4.46) 0.83*** (7.96) 0.83*** (7.52) 0.64*** (10.12) N (schools) 49 50 47 53 N (goals) 1,519 1,839 1,763 5,121 F-test 8.76*** 8.12*** 6.24*** 11.81*** Adjusted R2 0.062 0.054 0.033 0.036 Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.0035* (−2.66) −0.0055*** (−4.92) −0.0043*** (−4.48) −0.0050*** (−7.06) Survey indicator 0.086* (2.17) −0.030 (−0.67) −0.12** (−3.18) −0.013 (−0.48) Number of years measured 0.030** (2.80) 0.026** (2.91) 0.012 (1.75) 0.029*** (5.29) Political aspiration level 0.0021 (0.09) −0.14* (−2.44) −0.12** (−3.01) −0.069* (−2.07) Change in aspiration level 0.028 (0.61) −0.029 (−0.82) −0.046* (−2.51) Goal: inclusion −0.22*** (−4.01) 0.073 (1.06) −0.039 (−0.77) −0.078* (−2.36) Goal: GPA −0.096 (−1.37) −0.18*** (−3.86) −0.057 (−0.91) −0.13** (−3.00) Goal: transition to further education −0.23*** (−4.19) −0.0082 (−0.09) −0.13 (−1.81) −0.14** (−3.16) Goal: bullying −0.33*** (−4.41) −0.12 (−1.49) −0.19*** (−3.82) −0.18*** (−5.20) Goal: truancy 0.039 (0.50) −0.21** (−3.12) −0.19* (−2.13) −0.10* (−2.04) Cycle: 2009 Reference Cycle: 2011 0.10** (2.80) Cycle: 2013 0.081 (1.94) Constant 0.43*** (4.46) 0.83*** (7.96) 0.83*** (7.52) 0.64*** (10.12) N (schools) 49 50 47 53 N (goals) 1,519 1,839 1,763 5,121 F-test 8.76*** 8.12*** 6.24*** 11.81*** Adjusted R2 0.062 0.054 0.033 0.036 Note: OLS—schools fixed effects. Clustered standard errors. t-Statistics in parenthesis. *p < .05; **p < .01; ***p < .001 (two-sided test). Table A2. How Relative Performance Influences Principals’ Prioritization of Goals Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.0035* (−2.66) −0.0055*** (−4.92) −0.0043*** (−4.48) −0.0050*** (−7.06) Survey indicator 0.086* (2.17) −0.030 (−0.67) −0.12** (−3.18) −0.013 (−0.48) Number of years measured 0.030** (2.80) 0.026** (2.91) 0.012 (1.75) 0.029*** (5.29) Political aspiration level 0.0021 (0.09) −0.14* (−2.44) −0.12** (−3.01) −0.069* (−2.07) Change in aspiration level 0.028 (0.61) −0.029 (−0.82) −0.046* (−2.51) Goal: inclusion −0.22*** (−4.01) 0.073 (1.06) −0.039 (−0.77) −0.078* (−2.36) Goal: GPA −0.096 (−1.37) −0.18*** (−3.86) −0.057 (−0.91) −0.13** (−3.00) Goal: transition to further education −0.23*** (−4.19) −0.0082 (−0.09) −0.13 (−1.81) −0.14** (−3.16) Goal: bullying −0.33*** (−4.41) −0.12 (−1.49) −0.19*** (−3.82) −0.18*** (−5.20) Goal: truancy 0.039 (0.50) −0.21** (−3.12) −0.19* (−2.13) −0.10* (−2.04) Cycle: 2009 Reference Cycle: 2011 0.10** (2.80) Cycle: 2013 0.081 (1.94) Constant 0.43*** (4.46) 0.83*** (7.96) 0.83*** (7.52) 0.64*** (10.12) N (schools) 49 50 47 53 N (goals) 1,519 1,839 1,763 5,121 F-test 8.76*** 8.12*** 6.24*** 11.81*** Adjusted R2 0.062 0.054 0.033 0.036 Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.0035* (−2.66) −0.0055*** (−4.92) −0.0043*** (−4.48) −0.0050*** (−7.06) Survey indicator 0.086* (2.17) −0.030 (−0.67) −0.12** (−3.18) −0.013 (−0.48) Number of years measured 0.030** (2.80) 0.026** (2.91) 0.012 (1.75) 0.029*** (5.29) Political aspiration level 0.0021 (0.09) −0.14* (−2.44) −0.12** (−3.01) −0.069* (−2.07) Change in aspiration level 0.028 (0.61) −0.029 (−0.82) −0.046* (−2.51) Goal: inclusion −0.22*** (−4.01) 0.073 (1.06) −0.039 (−0.77) −0.078* (−2.36) Goal: GPA −0.096 (−1.37) −0.18*** (−3.86) −0.057 (−0.91) −0.13** (−3.00) Goal: transition to further education −0.23*** (−4.19) −0.0082 (−0.09) −0.13 (−1.81) −0.14** (−3.16) Goal: bullying −0.33*** (−4.41) −0.12 (−1.49) −0.19*** (−3.82) −0.18*** (−5.20) Goal: truancy 0.039 (0.50) −0.21** (−3.12) −0.19* (−2.13) −0.10* (−2.04) Cycle: 2009 Reference Cycle: 2011 0.10** (2.80) Cycle: 2013 0.081 (1.94) Constant 0.43*** (4.46) 0.83*** (7.96) 0.83*** (7.52) 0.64*** (10.12) N (schools) 49 50 47 53 N (goals) 1,519 1,839 1,763 5,121 F-test 8.76*** 8.12*** 6.24*** 11.81*** Adjusted R2 0.062 0.054 0.033 0.036 Note: OLS—schools fixed effects. Clustered standard errors. t-Statistics in parenthesis. *p < .05; **p < .01; ***p < .001 (two-sided test). Table A3. How Relative Performance Influence Principals’ Prioritization of Goals Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.019*** (−4.04) −0.029*** (−6.36) −0.022*** (−4.50) −0.024*** (−9.44) Survey indicator 0.50** (3.11) −0.16 (−1.06) −0.63*** (−4.31) −0.061 (−0.75) Number of years measured 0.18** (2.73) 0.14** (2.65) 0.061 (1.24) 0.15*** (4.94) Political aspiration level −0.027 (−0.10) −0.71*** (−4.50) −0.60*** (−3.40) −0.31** (−3.15) Change in aspiration level 0.16 (0.60) −0.14 (−0.80) −0.23* (−2.55) Goal: inclusion −2.25** (−3.02) 0.39 (1.07) −0.32 (−0.76) −0.53* (−2.20) Goal: GPA −0.45 (−0.98) −0.93* (−2.29) −0.33 (−0.80) −0.65** (−2.73) Goal: transition to further education −2.21*** (−4.05) −0.028 (−0.10) −0.70* (−2.36) −0.77*** (−4.35) Goal: bullying −2.07*** (−4.03) −0.64 (−1.61) −1.15* (−2.55) −0.96*** (−4.18) Goal: truancy 0.24 (0.83) −1.35*** (−3.75) −1.13** (−3.28) −0.56** (−3.06) Cycle: 2009 Reference Cycle: 2011 0.49*** (4.68) Cycle: 2013 0.39*** (3.54) N (schools) 45 49 45 52 N (goals) 1,402 1,801 1,687 5,083 Log likelihood −637.39 −884.47 −852.23 −2770.02 Likelihood ratio 112.76*** 108.44*** 67.48*** 194.53*** Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.019*** (−4.04) −0.029*** (−6.36) −0.022*** (−4.50) −0.024*** (−9.44) Survey indicator 0.50** (3.11) −0.16 (−1.06) −0.63*** (−4.31) −0.061 (−0.75) Number of years measured 0.18** (2.73) 0.14** (2.65) 0.061 (1.24) 0.15*** (4.94) Political aspiration level −0.027 (−0.10) −0.71*** (−4.50) −0.60*** (−3.40) −0.31** (−3.15) Change in aspiration level 0.16 (0.60) −0.14 (−0.80) −0.23* (−2.55) Goal: inclusion −2.25** (−3.02) 0.39 (1.07) −0.32 (−0.76) −0.53* (−2.20) Goal: GPA −0.45 (−0.98) −0.93* (−2.29) −0.33 (−0.80) −0.65** (−2.73) Goal: transition to further education −2.21*** (−4.05) −0.028 (−0.10) −0.70* (−2.36) −0.77*** (−4.35) Goal: bullying −2.07*** (−4.03) −0.64 (−1.61) −1.15* (−2.55) −0.96*** (−4.18) Goal: truancy 0.24 (0.83) −1.35*** (−3.75) −1.13** (−3.28) −0.56** (−3.06) Cycle: 2009 Reference Cycle: 2011 0.49*** (4.68) Cycle: 2013 0.39*** (3.54) N (schools) 45 49 45 52 N (goals) 1,402 1,801 1,687 5,083 Log likelihood −637.39 −884.47 −852.23 −2770.02 Likelihood ratio 112.76*** 108.44*** 67.48*** 194.53*** Note: Logistic regression—schools fixed effects. t-Statistics in parenthesis. *p < .05; **p < .01; ***p < .001 (two-sided test). Table A3. How Relative Performance Influence Principals’ Prioritization of Goals Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.019*** (−4.04) −0.029*** (−6.36) −0.022*** (−4.50) −0.024*** (−9.44) Survey indicator 0.50** (3.11) −0.16 (−1.06) −0.63*** (−4.31) −0.061 (−0.75) Number of years measured 0.18** (2.73) 0.14** (2.65) 0.061 (1.24) 0.15*** (4.94) Political aspiration level −0.027 (−0.10) −0.71*** (−4.50) −0.60*** (−3.40) −0.31** (−3.15) Change in aspiration level 0.16 (0.60) −0.14 (−0.80) −0.23* (−2.55) Goal: inclusion −2.25** (−3.02) 0.39 (1.07) −0.32 (−0.76) −0.53* (−2.20) Goal: GPA −0.45 (−0.98) −0.93* (−2.29) −0.33 (−0.80) −0.65** (−2.73) Goal: transition to further education −2.21*** (−4.05) −0.028 (−0.10) −0.70* (−2.36) −0.77*** (−4.35) Goal: bullying −2.07*** (−4.03) −0.64 (−1.61) −1.15* (−2.55) −0.96*** (−4.18) Goal: truancy 0.24 (0.83) −1.35*** (−3.75) −1.13** (−3.28) −0.56** (−3.06) Cycle: 2009 Reference Cycle: 2011 0.49*** (4.68) Cycle: 2013 0.39*** (3.54) N (schools) 45 49 45 52 N (goals) 1,402 1,801 1,687 5,083 Log likelihood −637.39 −884.47 −852.23 −2770.02 Likelihood ratio 112.76*** 108.44*** 67.48*** 194.53*** Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.019*** (−4.04) −0.029*** (−6.36) −0.022*** (−4.50) −0.024*** (−9.44) Survey indicator 0.50** (3.11) −0.16 (−1.06) −0.63*** (−4.31) −0.061 (−0.75) Number of years measured 0.18** (2.73) 0.14** (2.65) 0.061 (1.24) 0.15*** (4.94) Political aspiration level −0.027 (−0.10) −0.71*** (−4.50) −0.60*** (−3.40) −0.31** (−3.15) Change in aspiration level 0.16 (0.60) −0.14 (−0.80) −0.23* (−2.55) Goal: inclusion −2.25** (−3.02) 0.39 (1.07) −0.32 (−0.76) −0.53* (−2.20) Goal: GPA −0.45 (−0.98) −0.93* (−2.29) −0.33 (−0.80) −0.65** (−2.73) Goal: transition to further education −2.21*** (−4.05) −0.028 (−0.10) −0.70* (−2.36) −0.77*** (−4.35) Goal: bullying −2.07*** (−4.03) −0.64 (−1.61) −1.15* (−2.55) −0.96*** (−4.18) Goal: truancy 0.24 (0.83) −1.35*** (−3.75) −1.13** (−3.28) −0.56** (−3.06) Cycle: 2009 Reference Cycle: 2011 0.49*** (4.68) Cycle: 2013 0.39*** (3.54) N (schools) 45 49 45 52 N (goals) 1,402 1,801 1,687 5,083 Log likelihood −637.39 −884.47 −852.23 −2770.02 Likelihood ratio 112.76*** 108.44*** 67.48*** 194.53*** Note: Logistic regression—schools fixed effects. t-Statistics in parenthesis. *p < .05; **p < .01; ***p < .001 (two-sided test). Table A4. How Prioritization of Goals Influences Performance in Subsequent Years Model 1 2 3 3.1 3.2 Goals are prioritized in One cycle (09–11) One cycle (11–13) Two cycles (09–13) Two cycles (09–13) Two cycles (09–13) Sample All goals All goals All goals Low-performance goals High-performance goals Time  2007 −0.72 (−1.17) −1.97** (−3.02) −0.66 (−1.14) −2.24** (−2.85) −4.21*** (−9.62)  2009 Ref. (.) −1.27*** (−3.33) Ref. (.) Ref. (.) Ref. (.)  2011 1.36*** (3.58) Ref. (.) 1.14** (3.17) 2.43*** (5.03) −0.23 (−0.85)  2013 0.21 (0.57) −1.03* (−2.53) 0.28 (0.80) 4.42*** (8.16) −0.31 (−1.35) Diff-in-diffs estimates  Prioritization × 2007 −1.83 (−1.46) 0.52 (0.47) −1.19 (−1.11) 0.46 (0.37) 1.07 (1.07)  Prioritization × 2009 Ref. (.) 0.78 (1.16) Ref. (.) Ref. (.) Ref. (.)  Prioritization × 2011 0.56 (0.75) Ref. (.) 1.10 (1.38) 1.34 (1.47) 0.012 (0.02)  Prioritization × 2013 0.21 (0.27) 0.51 (0.78) 4.07*** (4.97) 3.10** (3.25) 1.07 (1.48) Constant 81.9*** (349.23) 82.3*** (270.62) 80.6*** (353.92) 66.1*** (191.91) 93.6*** (615.52) Controls Yes Yes Yes Yes Yes N 4,165 4,491 4,028 1,958 2,070 F-test 5.04*** 2.56** 10.38*** 29.04*** 15.27*** Adjusted R2 0.014 0.0040 0.006 0.014 0.11 Model 1 2 3 3.1 3.2 Goals are prioritized in One cycle (09–11) One cycle (11–13) Two cycles (09–13) Two cycles (09–13) Two cycles (09–13) Sample All goals All goals All goals Low-performance goals High-performance goals Time  2007 −0.72 (−1.17) −1.97** (−3.02) −0.66 (−1.14) −2.24** (−2.85) −4.21*** (−9.62)  2009 Ref. (.) −1.27*** (−3.33) Ref. (.) Ref. (.) Ref. (.)  2011 1.36*** (3.58) Ref. (.) 1.14** (3.17) 2.43*** (5.03) −0.23 (−0.85)  2013 0.21 (0.57) −1.03* (−2.53) 0.28 (0.80) 4.42*** (8.16) −0.31 (−1.35) Diff-in-diffs estimates  Prioritization × 2007 −1.83 (−1.46) 0.52 (0.47) −1.19 (−1.11) 0.46 (0.37) 1.07 (1.07)  Prioritization × 2009 Ref. (.) 0.78 (1.16) Ref. (.) Ref. (.) Ref. (.)  Prioritization × 2011 0.56 (0.75) Ref. (.) 1.10 (1.38) 1.34 (1.47) 0.012 (0.02)  Prioritization × 2013 0.21 (0.27) 0.51 (0.78) 4.07*** (4.97) 3.10** (3.25) 1.07 (1.48) Constant 81.9*** (349.23) 82.3*** (270.62) 80.6*** (353.92) 66.1*** (191.91) 93.6*** (615.52) Controls Yes Yes Yes Yes Yes N 4,165 4,491 4,028 1,958 2,070 F-test 5.04*** 2.56** 10.38*** 29.04*** 15.27*** Adjusted R2 0.014 0.0040 0.006 0.014 0.11 Note: Difference-in-differences estimates—OLS (goal fixed effects). t-Statistics in parenthesis. Coefficients that test for a common trend are in italics. The models include controls for whether a political aspiration level was introduced on the performance dimension and whether this level changed. *p < .05; **p < .01; ***p < .001 (two-sided test). Table A4. How Prioritization of Goals Influences Performance in Subsequent Years Model 1 2 3 3.1 3.2 Goals are prioritized in One cycle (09–11) One cycle (11–13) Two cycles (09–13) Two cycles (09–13) Two cycles (09–13) Sample All goals All goals All goals Low-performance goals High-performance goals Time  2007 −0.72 (−1.17) −1.97** (−3.02) −0.66 (−1.14) −2.24** (−2.85) −4.21*** (−9.62)  2009 Ref. (.) −1.27*** (−3.33) Ref. (.) Ref. (.) Ref. (.)  2011 1.36*** (3.58) Ref. (.) 1.14** (3.17) 2.43*** (5.03) −0.23 (−0.85)  2013 0.21 (0.57) −1.03* (−2.53) 0.28 (0.80) 4.42*** (8.16) −0.31 (−1.35) Diff-in-diffs estimates  Prioritization × 2007 −1.83 (−1.46) 0.52 (0.47) −1.19 (−1.11) 0.46 (0.37) 1.07 (1.07)  Prioritization × 2009 Ref. (.) 0.78 (1.16) Ref. (.) Ref. (.) Ref. (.)  Prioritization × 2011 0.56 (0.75) Ref. (.) 1.10 (1.38) 1.34 (1.47) 0.012 (0.02)  Prioritization × 2013 0.21 (0.27) 0.51 (0.78) 4.07*** (4.97) 3.10** (3.25) 1.07 (1.48) Constant 81.9*** (349.23) 82.3*** (270.62) 80.6*** (353.92) 66.1*** (191.91) 93.6*** (615.52) Controls Yes Yes Yes Yes Yes N 4,165 4,491 4,028 1,958 2,070 F-test 5.04*** 2.56** 10.38*** 29.04*** 15.27*** Adjusted R2 0.014 0.0040 0.006 0.014 0.11 Model 1 2 3 3.1 3.2 Goals are prioritized in One cycle (09–11) One cycle (11–13) Two cycles (09–13) Two cycles (09–13) Two cycles (09–13) Sample All goals All goals All goals Low-performance goals High-performance goals Time  2007 −0.72 (−1.17) −1.97** (−3.02) −0.66 (−1.14) −2.24** (−2.85) −4.21*** (−9.62)  2009 Ref. (.) −1.27*** (−3.33) Ref. (.) Ref. (.) Ref. (.)  2011 1.36*** (3.58) Ref. (.) 1.14** (3.17) 2.43*** (5.03) −0.23 (−0.85)  2013 0.21 (0.57) −1.03* (−2.53) 0.28 (0.80) 4.42*** (8.16) −0.31 (−1.35) Diff-in-diffs estimates  Prioritization × 2007 −1.83 (−1.46) 0.52 (0.47) −1.19 (−1.11) 0.46 (0.37) 1.07 (1.07)  Prioritization × 2009 Ref. (.) 0.78 (1.16) Ref. (.) Ref. (.) Ref. (.)  Prioritization × 2011 0.56 (0.75) Ref. (.) 1.10 (1.38) 1.34 (1.47) 0.012 (0.02)  Prioritization × 2013 0.21 (0.27) 0.51 (0.78) 4.07*** (4.97) 3.10** (3.25) 1.07 (1.48) Constant 81.9*** (349.23) 82.3*** (270.62) 80.6*** (353.92) 66.1*** (191.91) 93.6*** (615.52) Controls Yes Yes Yes Yes Yes N 4,165 4,491 4,028 1,958 2,070 F-test 5.04*** 2.56** 10.38*** 29.04*** 15.27*** Adjusted R2 0.014 0.0040 0.006 0.014 0.11 Note: Difference-in-differences estimates—OLS (goal fixed effects). t-Statistics in parenthesis. Coefficients that test for a common trend are in italics. The models include controls for whether a political aspiration level was introduced on the performance dimension and whether this level changed. *p < .05; **p < .01; ***p < .001 (two-sided test). References Andersen , Simon Calmar . 2008 . The impact of public management reforms on student performance in Danish schools . Public Administration 86 : 541 – 58 . Google Scholar CrossRef Search ADS Andersen , Simon Calmar , and Peter B . Mortensen . 2010 . “ Policy Stability and Organizational Performance: Is There a Relationship ?” Journal of Public Administration Research and Theory 20 : 1 – 22 . Google Scholar CrossRef Search ADS Andersen , Simon Calmar , and Donald P. Moynihan . 2016 . How leaders respond to diversity: The moderating role of organizational culture on performance information use . Journal of Public Administration Research and Theory 26 : 448 – 60 . Google Scholar CrossRef Search ADS Andersen , Simon Calmar , Maria K. Humlum , and Anne B. Nandrup . 2016 . Increasing instruction time in school does increase learning . Proceedings of the National Academy of Sciences of the United States of America 113 : 7481 – 4 . Google Scholar CrossRef Search ADS PubMed Angrist , Joshua D. , and Jörn-Steffen Pischke . 2009 . Mostly harmless econometrics: An empiricists companion . Princeton : Princeton Univ. Press . Askim , Jostein , Åge Johnsen , and Knut-Andreas Christophersen . 2008 . Factors behind organizational learning from benchmarking: Experiences from Norwegian municipal benchmarking networks . Journal of Public Administration Research and Theory 18 : 297 – 320 . Google Scholar CrossRef Search ADS Audia , Pino G. , and Sebastian Brion . 2007 . Reluctant to change: Self-enhancing responses to diverging performance measures . Organizational Behavior and Human Decision Processes 102 : 255 – 69 . Google Scholar CrossRef Search ADS Audia , Pino G. , Sebastian Brion , and Heinrich R. Greve . 2015 . Self-assessment, self-enhancement, and the choice of comparison organizations for evaluating organizational performance . Advances in Strategic Management: Cognition and Strategy 32 : 89 – 118 . Google Scholar CrossRef Search ADS Banerjee , Abhijit , Raghabendra Chattopadhyay , Esther Duflo , Daniel Keniston , and Nina Singh . 2014 . Can institutions be reformed from within? Evidence from a randomized experiment with the Rajasthan Police . Working paper. Baum , Joel A. , and Kristina B. Dahlin . 2007 . Aspiration performance and railroads’ patterns of learning from train wrecks and crashes . Organization Science 18 : 368 – 85 . Google Scholar CrossRef Search ADS Behn , Robert D . 2003 . Why measure performance? Different purposes require different measures . Public Administration Review 63 : 586 – 606 . Google Scholar CrossRef Search ADS Bertrand , Marianne , Esther Duflo , and Sendhil Mullainathan . 2004 . How much should we trust differences-in-differences estimates ? Quarterly Journal of Economics 119 : 249 – 75 . Google Scholar CrossRef Search ADS Bevan , Gwyn , and Christopher Hood . 2006 . What’s measured is what matters: Targets and gaming in the English public healthcare system . Public Administration 84 : 517 – 38 . Google Scholar CrossRef Search ADS Bourdeaux , Carolyn , and Grace Chikoto . 2008 . Legislative influences on performance management reform . Public Administration Review 68 : 253 – 65 . Google Scholar CrossRef Search ADS Boyne , George A . 2002 . Concepts and indicators of local authority performance: An evaluation of the statutory frameworks in England and Wales . Public Money and Management 22 : 17 – 24 . Google Scholar CrossRef Search ADS ———. 2003 . What is public service improvement ? Public Administration 81 : 211 – 27 . CrossRef Search ADS Busuioc , Madalina , and Martin Lodge . 2017 . Reputation and accountability relationships: Managing accountability expectations through reputation . Public Administration Review 77 : 91 – 100 . Google Scholar CrossRef Search ADS Carpenter , Daniel . 2001 . The forging of bureaucratic autonomy: Reputations, networks, and policy innovation in executive agencies . Princeton, NJ : Princeton Univ. Press ———. 2010 . Reputation and power: Organizational image and pharmaceutical regulation at the FDA . Princeton, NJ: Princeton Univ. Press Christensen , Julian , Casper Mondrup Dahlmann , Asbjørn Hovgaard Mathiasen , Donald P. Moynihan , and Niels Bjørn Grund Petersen . 2018 . How do elected officials evaluate performance? Goal preferences, governance preferences, and the process of goal reprioritization . Journal of Public Administration Research and Theory , 28:197–211. Cyert , Ricard M. , and James G. March . 1963 . A behavioral theory of the firm . Englewood Cliffs, NJ : Prentice-Hall . Diwas , K. C. , Bradley R. Staats , and Francesca Gino . 2013 . Learning from my success and from others’ failure: Evidence from minimally invasive cardiac surgery . Management Science 59 : 2435 – 49 . Google Scholar CrossRef Search ADS Feldman , Martha S. , and James G. March . 1981 . Information in organizations as signal and symbol . Administrative Science Quarterly 26 : 171 – 86 . Google Scholar CrossRef Search ADS Gerrish , ED . 2016 . The impact of performance management on performance in public organizations: A meta-analysis . Public Administration Review 76 : 48 – 66 . Google Scholar CrossRef Search ADS Gilmour , John B. , and David E. Lewis . 2006 . Does performance budgeting work? An examination of the office of management and budget’s PART scores . Public Administration Review 66 : 742 – 52 . Google Scholar CrossRef Search ADS Hanushek , Eric A . 2003 . The failure of input-based schooling policies . The Economic Journal 113 : 64 – 98 . Google Scholar CrossRef Search ADS Heckman , James J . 2000 . Policies to foster human capital . Research in Economics 54 : 3 – 56 . Google Scholar CrossRef Search ADS Heinrich , Carolyn J . 1999 . Do government bureaucrats make effective use of performance management information ? Journal of Public Administration Research and Theory 9 : 363 – 93 . Google Scholar CrossRef Search ADS ———. 2010 . Third-party governance under no child left behind: Accountability and performance management challenges in supplemental educational services provision . Journal of Public Administration Research and Theory 20 : 59 – 80 . CrossRef Search ADS Hesketh , Sophia Rabe , and Anders Skrondal . 2012 . Multilevel and longitudinal modeling using Stata , 3rd ed . College Station, TX : Stata Press Publication . Hood , Christopher . 2011 . The blame game: Spin, bureaucracy, and self-preservation in government . Princeton, NJ : Princeton Univ. Press . Jordan , Alexander H. , and Pino G. Audia . 2012 . Self-enhancement and learning from performance feedback . Academy of Management Review 37 : 211 – 31 . Google Scholar CrossRef Search ADS Kelman , Steven , and John N. Friedman . 2009 . Performance improvement and performance dysfunction: An empirical examination of distortionary impacts of the emergency room wait-time target in the English National Health Service . Journal of Public Administration Research and Theory 19 : 917 – 46 . Google Scholar CrossRef Search ADS Krause , George A . 2003 . Coping with uncertainty: Analyzing risk propensities of SEC budgetary decisions, 1949–1997 . American Political Science Review 97 : 171 – 88 . Google Scholar CrossRef Search ADS Kroll , Alexander . 2013 . The other type of performance information: Nonroutine feedback, its relevance and use . Public Administration Review 73 : 265 – 76 . Google Scholar CrossRef Search ADS ———. 2015a . Exploring the link between performance information use and organizational performance: A contingency approach . Public Performance and Management Review 39 : 7 – 32 . CrossRef Search ADS ———. 2015b . Explaining the use of performance information by public managers: A planned-behavior approach . American Review of Public Administration 45 : 201 – 15 . CrossRef Search ADS Lavertu , Stéphane , David E. Lewis , and Donald P. Moynihan . 2013 . Government reform, political ideology, and administrative burden: The case of performance management in the Bush administration . Public Administration Review 73 : 845 – 57 . Google Scholar CrossRef Search ADS Lindblom , Charles E . 1959 . The science of “muddling through” . Public Administration Review 19 : 79 – 88 . Google Scholar CrossRef Search ADS MacDonald , Jason A. , and William W. Franko . 2007 . Bureaucratic capacity and bureaucratic discretion: Does congress tie policy authority to performance ? American Politics Research 35 : 790 – 807 . Google Scholar CrossRef Search ADS Maor , Moshe , Sharon Gilad , and Pazit Ben-Nun Bloom . 2013 . Organizational reputation, regulatory talk, and strategic silence . Journal of Public Administration Research and Theory 23 : 581 – 608 . Google Scholar CrossRef Search ADS March , James G . 1991 . Exploration and exploitation in organizational learning . Organization Science 2 : 71 – 87 . Google Scholar CrossRef Search ADS Meier , Kenneth J. , and Laurence J. O’Toole Jr . 2013 . Subjective organizational performance and measurement error: Common source bias and spurious relationships . Journal of Public Administration Research and Theory 23 : 429 – 56 . Google Scholar CrossRef Search ADS ———. 2008 . Management theory and Occam’s razor: How public organizations buffer the environment . Administration and Society 39 : 931 – 58 . CrossRef Search ADS Meier , Kenneth J. , Nathan Favero , and Ling Zhu . 2015 . Performance gaps and managerial decisions: A Bayesian decision theory of managerial action . Journal of Public Administration Research and Theory 25 : 1221 – 46 . Google Scholar CrossRef Search ADS Melkers , Julia , and Kathrine Willoughby . 2005 . Models of performance-measurement use in local governments: Understanding budgeting, communication, and lasting effects . Public Administration Review 65 : 180 – 90 . Google Scholar CrossRef Search ADS Migue , Jean-Luc , Gérard Belanger , and William A. Niskanen . 1974 . Towards a general theory of managerial discretion . Public Choice 17 : 27 – 51 . Google Scholar CrossRef Search ADS Mintzberg , Henry , Duru Raisinghani , and Andre Théorêt . 1976 . The structure of “unstructured” decision processes . Administrative Science Quarterly 21 : 246 – 75 . Google Scholar CrossRef Search ADS Moynihan , Donald P . 2005 . Goal-based learning and the future of performance management . Public Administration Review 65 : 203 – 16 . Google Scholar CrossRef Search ADS ———. 2008 . Dynamics of performance management: Constructing information and reform . Washington, DC: Georgetown Univ. Press . ——— . 2009 . “ Through a Glass, Darkly: Understanding the Effects of Performance Regimes .” Public Performance & Management Review 32 : 592 – 603 . Google Scholar CrossRef Search ADS Moynihan , Donald P. , Sanjay K. Pandey , and Bradley E. Wright . 2011a . Setting the table: How transformational leadership fosters performance information use . Journal of Public Administration Research and Theory 22 : 143 – 64 . Google Scholar CrossRef Search ADS Moynihan , Donald P. , Sergio Fernandez , Soonhee Kim , Kelly LeRoux , Suzanne J. Piotrowski , Bradley Wright , and Kaifeng Yang . 2011b . Performance regimes amidst governance complexity . Journal of Public Administration Research and Theory 21 ( Suppl 1 ): i141 – 55 . Google Scholar CrossRef Search ADS Moynihan , Donald P. , and Stephane Lavertu . 2012 . Does involvement in performance management routines encourage performance information use? Evaluating GPRA and PART . Public Administration Review 72 : 592 – 602 . Google Scholar CrossRef Search ADS Nielsen , Poul A . 2014a . Performance management, managerial authority, and public service performance . Journal of Public Administration Research and Theory 24 : 431 – 58 . Google Scholar CrossRef Search ADS ———. 2014b . Learning from performance feedback: Performance information, aspiration levels, and managerial priorities . Public Administration 92 : 142 – 60 . CrossRef Search ADS Nielsen , Poul A. , and Donald P. Moynihan . 2017 . How do politicians attribute bureaucratic responsibility for performance? Negativity bias and interest group advocacy . Journal of Public Administration Research and Theory . 27:269–283. O’Toole , Laurence , and Kenneth J. Meier . 2011 . Public management: Organizations, governance, and performance . Cambridge : Cambridge Univ. Press . Google Scholar CrossRef Search ADS Olsen , Asmus . 2017 . Compared to what? Experimental evidence on social and historical reference points in performance evaluation . Journal of Public Administration Research and Theory . 27:562–580. Ostrom , Elinor , Roger B. Parks , Gordon P. Whitaker , and Stephen L. Percy . 1978 . The public service production process: A framework for analyzing police services . Policy Studies Journal 7 : 381 – 9 . Google Scholar CrossRef Search ADS Poister , Theodore H . 2010 . The future of strategic planning in the public sector: Linking strategic management and performance . Public Administration Review 70 : 246 – 54 . Google Scholar CrossRef Search ADS Pollitt , Christopher . 2013 . The logics of performance management . Evaluation 19 : 346 – 63 . Google Scholar CrossRef Search ADS Pritchett , Lant , and Deon Filmer . 1999 . What education production functions really show: A positive theory of education expenditure . Economics of Education Review 18 : 223 – 39 . Google Scholar CrossRef Search ADS Rich , Robert F . 1997 . Measuring knowledge utilization: Process and outcomes . Knowledge and Policy 10 : 11 – 24 . Google Scholar CrossRef Search ADS Rockoff , Jonah E. , Douglas O. Staiger , Thomas J. Kane , and Eric S. Taylor . 2012 . Information and employee evaluation: Evidence from a randomized intervention in public schools . American Economic Review 102 : 3184 – 213 . Google Scholar CrossRef Search ADS Rutherford , Amanda , and Kenneth J. Meier . 2015 . Managerial goals in a performance-driven system: Theory and empirical tests in higher education . Public Administration 93 : 17 – 33 . Google Scholar CrossRef Search ADS Salge , Torsten O . 2011 . A behavioral model of innovative search: Evidence from public hospital services . Journal of Public Administration Research and Theory 21 : 181 – 210 . Google Scholar CrossRef Search ADS Shinkle , George A . 2012 . “ Organizational Aspirations, Reference Points, and Goals: Building on the Past and Aiming for the Future .” Journal of Management 38 : 415 – 55 . Google Scholar CrossRef Search ADS Snyder , Jeffrey W. , Andrew Saultz , and Rebecca Jacobsen . 2018 . Antipolitics and the hindrance of performance management in education . Journal of Public Administration Research and Theory . Forthcoming. Soss , Joe , Richard Fording , and Sanford F. Schram . 2011 . The organization of discipline: From performance management to perversity and punishment . Journal of Public Administration Research and Theory 21 : 203 – 32 . Google Scholar CrossRef Search ADS Spence , Michael . 2002 . Signaling in retrospect and the informational structure of markets . American Economic Review 92 : 434 – 59 . Google Scholar CrossRef Search ADS Taylor , Jeannette . 2009 . Strengthening the link between performance measurement and decision making . Public Administration 87 : 853 – 71 . Google Scholar CrossRef Search ADS Vakkuri , Jarmo . 2010 . Struggling with ambiguity: Public managers as users of NPM-oriented management instruments . Public Administration 88 : 999 – 1024 . Google Scholar CrossRef Search ADS van Helden , Jan G. , Åge Johnsen , and Jarmo Vakkuri . 2012 . The life cycle approach to performance management: Implications for public management and evaluation . Evaluation 18 : 159 – 75 . Google Scholar CrossRef Search ADS Wooldridge , Jeffrey M . 2002 . Econometric analysis of cross section and panel data . 4th edn. Mason; Cambridge, MA : MIT Press . ———. 2009 . Introductory econometrics—A modern approach . South Western College Pub . Yang , Kaifeng , and Jun Yi Hsieh . 2007 . Managerial effectiveness of government performance measurement: Testing a middle-range model . Public Administration Review 67 : 861 – 79 . Google Scholar CrossRef Search ADS © The Author(s) 2018. Published by Oxford University Press on behalf of the Public Management Research Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Public Administration Research and Theory Oxford University Press

Successful Problem Solvers? Managerial Performance Information Use to Improve Low Organizational Performance

Loading next page...
 
/lp/ou_press/successful-problem-solvers-managerial-performance-information-use-to-1xN000KeEF
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the Public Management Research Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
ISSN
1053-1858
eISSN
1477-9803
D.O.I.
10.1093/jopart/muy017
Publisher site
See Article on Publisher Site

Abstract

Abstract Performance management is increasingly the norm for public organizations. Although there is an emergent literature on performance information use, we still know little on how managers engage in functional performance management practices. At the same time, growing evidence suggests that managers face pressure to improve low performance because of a negativity bias in the political environment. However, in managerial performance information use, the negativity bias might be reconsidered as a prioritization heuristic with positive performance attributes, directing attention to organizational goals with a favorable return of investment. I test this argument with data from public schools. A fixed-effect estimation is used to analyze how principals prioritize when they are provided with performance information on a number of different educational goals. Furthermore, a difference-in-differences model tests whether the prioritizations of certain goals have performance-enhancing effects over time. The analysis shows that principals prioritize goals with low performance and that prioritizations result in performance increase. The improvements primarily occur for goals that have a low performance level and that are repeatedly prioritized. Introduction During the last decades, one of the most widespread trends in the governing of public organizations has been the introduction of performance management. The core of performance management systems is performance information, which should inform managers’ decisions and thereby improve organizational performance (Andersen and Moynihan 2016; Heinrich 1999; Kroll 2015a; Nielsen 2014a; Snyder, Saultz, and Jacobsen et al. 2018). However, evidence of success is scarce. This is partly because managers show reluctance to integrate performance information in their routines and decisions (Kroll 2013; Melkers and Willoughby 2005; Moynihan and Lavertu 2012) and partly due to the inherent difficulty in identifying examples of purposeful use. Taken together, these circumstances leave us without much empirical and theoretical understanding of functional performance management practices (for an exception, see Kelman and Friedman 2009). So, how could managers use performance information to create organizational improvements? While the literature on performance management points to informed decision-making as the key to this process (e.g., Moynihan 2008; Taylor 2009; Vakkuri 2010), it offers little guidance on how data could be transformed into meaningful messages that guide organizational change. This study addresses this shortcoming by presenting two ways managers could utilize information: Either by reacting to signals of failure and initiating problem solving (Cyert and March 1963; Meier, Favero, and Zhu 2015), or by exploiting successes to develop expertise and specialize (March 1991). To provide insights into which strategy managers pursue, this study examines how performance on different organizational goals influences managers’ willingness to prioritize certain goals. While previous research has provided important insights into the ways to increase performance information use (Moynihan, Pandey, and Wright 2011a; Yang and Hsieh 2007), most studies rely on self-reports from managers to evaluate their use. By testing the influence of information on a tangible decision outcome, this study offers one of the first accounts of performance information use in actual decision-making. A key argument in the article is that prioritizations do not only concern rationales for improvements, signaling effects to the political environment also play a part (Bourdeaux and Chikoto 2008; Lavertu, Lewis, and Moynihan 2013). The study extends previous work in this area by testing the idea that the strong emphasis on failures (i.e., a negativity bias) in the political environment (Hood 2011; Nielsen and Moynihan 2017; Olsen 2017) encourages managers to prioritize goals with low performance. The distinction between internal and external rationales is not only important for understanding a manager’s choice of priorities; it also has implications for the outcome of this decision. Priorities that are chosen solely for their signaling value to political principals might end up being nothing more than symbolic (Feldman and March 1981). However, even when the decision initiates functional organizational change, success is no guarantee. I argue that prioritizations influence outcomes with a diminishing return, meaning that they primarily lead to performance increases for the most problematic goals. In this way, managers’ tendency to focus on performance failures may actually have positive attributes, directing scarce attention to goals with the highest potential for improvement. Other than being one of the first studies to address these questions, the article contributes to the literature on performance management in three general ways. First, I provide an empirical example of how managers engage in functional performance management practices, namely by using performance information to solve problems. The context for this finding is Danish public schools where principals prioritize the educational goals (for instance grades, truancy, wellbeing, parental satisfaction, and student overweight) where their school is performing the worst. After a goal have been made a priority for several years, the performance level increases. While it is not possible to attribute these improvements directly to the use of information, the analysis shows that performance primarily increases for goals with a low initial performance level, thereby indicating that the underlying mechanism is managerial problem solving. On this basis, the study provides an important perspective to the many studies that tell a pessimistic story on performance management in the public sector (e.g., Andersen 2008; Gerrish 2016; Heinrich 2010). In addition to the empirical account, the article contributes theoretically to the debate on how managers could use performance information to make decisions that improve organizational performance (Askim, Johnsen, and Christophersen 2008; Kroll 2015a; Moynihan 2009). More specifically, I develop a comprehensive theoretical framework that theorizes on the rationales for utilizing performance information in strategic decision-making, but that also considers the pitfalls when these decisions unfold as attempts to improve performance. Finally, a valid test of the research questions requires addressing the methodological challenges in the literature. To do this, I operationalize managers’ use of performance information in a novel way. Since the principals make an explicit prioritization, it is possible to couple their decisions with the performance level of each goal. This way of structuring data avoids the problem of common source bias (Meier and O’Toole 2013) and allows the use of a fixed-effect model and a difference-in-differences setup that can address the problem of omitted variable bias more exhaustingly than previous research. Purposeful Performance Information Use The theory consists of the following sections. The first section outlines the conceptual framework for understanding performance information use in strategic decision-making. In the next section, I present the internal and external managerial considerations for prioritizing goals with low- or high-performance levels. The following sections concern the effects of goal prioritizations on performance. Here, I consider situations with symbolic prioritizations, as well as instances where prioritizations initiate functional organization change. Expectations for the latter situation are developed by theorizing on the production process, including the possibility of diminishing returns, the importance of sustained attention, and performance trade-offs between goals. Strategic Performance Management This article focuses on managerial use of performance information in strategic decision-making, which broadly concerns a manager’s choice of strategies and actions to ensure goal achievement (Poister 2010). To theorize on this process, two shortcomings in the literature on performance management need consideration. The first relates to goal multiplicity. Public organizations often pursue multiple goals, and performance indicators at least partly reflect this multiplicity (Christensen et al. 2018; Moynihan et al. 2011b). However, studies of performance information use rarely consider this point, neither theoretically nor empirically. Performance is either treated as an abstract concept (in studies based on self-reports), or only one dimension of performance is examined (e.g., students’ final grades). Second, research on performance management needs a more precise conceptual understanding of what “use” entails. Previous research relies on self-reports from managers to solve this issue, thereby leaving what constitutes use to be a subjective matter. This study takes a more independent approach and conceptualizes use as the process in which performance information—consciously or unconsciously—influences managers and thereby contributes to a specific thought, a decision or action (Rich 1997, 15). In relation to these points, performance information use in strategic decision-making concerns the relationship between performance levels across organizational goals and a manager’s willingness to make certain goals a priority. Simply put, this means that managers can use the information to direct attention to goals with a (relatively) low-performance level or goals with a (relatively) high-performance level. However, there is also a possibility that managers do not let their prioritizations influence by the results at all. Such a tendency would reflect the many studies that find managerial reluctance to inform decisions and practices with performance information (Kroll 2015b; Moynihan and Lavertu 2012; Moynihan, Pandey, and Wright 2011a; Taylor 2009). Even though the studies do not examine use in relation to strategic prioritizations, the findings suggest that managers’ decisions in performance management systems often reflect other factors than the content of the information. Rationales for Using Performance Information to Prioritize Goals In the situation where managers actually consult performance information when deciding on organizational goals, where do they turn their attention? An important distinction in answering this question is between internal and external considerations (O’Toole and Meier 2011). The internal dimension concerns management in relation to processes that take place within the organization with the objective of creating performance improvements (Meier, Favero, and Zhu 2015). While external management could also concern improvements, this aspect of public management has a more intermediate objective, namely to maintain autonomy and support for the organization (Carpenter 2001). Internal Considerations: Problem-Solving and Specialization An obvious starting point for the internal consideration is the research tradition that originates from the behavioral model developed in the work by Cyert and March (1963). A key idea in this model is that managers improve performance through three sequential phases, namely (1) identification of problems, (2) search for solutions, and (3) implementation of the most appropriate solution (Cyert and March 1963, 34). The identification phase starts with a stimulus that indicates a problem and a need for managerial action (Mintzberg, Raisinghani, and Théorêt 1976). Performance information is a key component in this process, quantifying the results of activities and thereby easing the identification of problematic areas in the organization’s work. The main theoretical developments and empirical tests of the problem-solving model have taken place in the private sector (for a review, see Shinkle 2012). However, recently, a number of studies have used the framework to explain public managers’ response to performance data. For example, Salge (2011) finds that performance shortfalls lead public organizations to increase search for innovative solutions, and Nielsen (2014b) and Rutherford and Meier (2015) show that managers redirect attention to various organizational outcomes in response to performance declines and negative social feedback. While these studies have provided vital first steps in adapting the behavioral model to the public sector, their conceptualization of performance shortfalls does not capture the complexity in modern public management completely. The underlying assumption in the studies is that managers evaluate performance shortfalls and successes either along a single performance dimension or in serial processing between multiple dimensions, comparing absolute results to an aspiration level. Thus, the performance hierarchy between goals is not taken into consideration, which leaves out an important aspect of how managers form opinions about what constitutes low and high performance. On a theoretical level, the idea of a problem-solving manager has also been subject to discussion. On the one side is the line of research that draws on the psychological literature on self-affirmation and points to self-enhancement as a motive when managers process performance information (Jordan and Audia 2012). Empirical support for this idea is found in the studies that show that managers adapt various strategies to cope with performance shortfalls, for instance, by choosing social and historical reference points that portrait performance in a positive light (Audia, Brion, and Greve 2015) or shifting focus to favorable performance indicators (Audia and Brion 2007). The last finding is particularly interesting as it connects to another research tradition that focuses on how managers create improvements from successes (e.g., Baum and Dahlin 2007; Diwas, Staats, and Gino 2013). A key point in this literature is that an organization could build competence and gain improvement through exploitation and utilization of tacit knowledge about functional organizational activities (March 1991). Knowledge of such activities could be employed in two different ways to create further improvements. One way is to dismantle information on the functional activities throughout the organization in an attempt to create best-practice learning (Askim, Johnsen, and Christophersen 2008). Another way is to strengthen the work on the successful goal dimension. Here, the knowledge would enter as competence building in an effort to specialize activities, for example, by reorganizing resources to enhance the practices that have proven functional. External Considerations: Buffering and Reputation Building Managerial reactions to performance information do not take place in an environmental vacuum; rather, they are a vital part of hierarchical accountability relations. This point is evident in many studies that show that an engaging political environment significantly increases the likelihood of managers responding to the information (Bourdeaux and Chikoto 2008; Lavertu, Lewis, and Moynihan 2013). While the findings suggest that managers pay attention to their environment in performance information use, there is little theoretical basis for understanding how political considerations influence the decisions made in the process. One important thing to note about the decision to prioritize certain organizational goals is the signaling effect to the environment (Spence 2002). This signaling effect occurs is particularly relevant when priorities have a clear connection to a performance result (Poister 2010). For goals with low performance, a prioritization signalizes responsiveness to organizational problems; for goals with high performance, a prioritization signalizes competence building. Both signals could be used strategically by managers in their quest for political support and autonomy. Prioritization of goals with low performance is a managerial strategy to buffer the organization from political interventions (Meier and O’Toole 2008). Politicians have a strong electorate incentive to avoid blame for poor performance results (Krause 2003), which could lead them to disclaim responsibility and instead attribute control over outcomes to managers (Nielsen and Moynihan 2017). In addition, low performance could offset a political reaction, for instance, in terms reduced autonomy (MacDonald and Franko 2007) and funding (Gilmour and Lewis 2006). Managers can use their prioritizations to avoid such a reaction by signalizing to the political environment that the problem is recognized and action is being taken. Prioritization of goals with high performance is a strategy to improve the organization’s reputation. Since a reputation describes the stakeholders’ perceptions of the unique capacities, roles, and obligations of the organization, it is a valuable political asset (Carpenter 2010). For this reason, managers spend time creating, maintaining, and protecting their organization’s reputation (Maor, Gilad, and Bloom 2013). One way to achieve this objective is to highlight the organization’s contribution to the public good by appearing successful and distinct from other organizations (Busuick and Lodge 2017). A prioritization of a goal with high performance can serve to fulfill this purpose because the signal should draw political attention to well-functioning areas of the organization’s work, thereby highlighting the organization’s impact on society. How Do Managers Weight the Different Considerations? The previous sections have identified both internal and external rationales for how managers could integrate performance information in their prioritizations of organizational goals. So, how should we expect managers to weight these different considerations? One aspect that could factor into this decision is accountability relations. In this regard, an important feature of political environment is a negativity bias, which creates an environment where politicians and citizens evaluate performance information with a strong emphasis on failures (Hood 2011; Nielsen and Moynihan 2017; Olsen 2017). Because of these tendencies, managers might perceive the problem solving and buffering strategies as particularly beneficial to them. This expectation is tested in the first hypothesis. H1 Public managers will react to performance information on multiple goals by prioritizing goals with the lowest performance level. Effects of Goal Prioritization Symbolic Decisions The distinction between external and internal rationales for prioritizing certain organizational goals is particularly important when considering the potential performance effects of the decision. The reason is that prioritizations based solely on external considerations risk being symbolic use of performance information (Feldman and March 1981). Symbolic use occurs in a situation where managers decide in a way that reflects the content of the information, but they do nothing more than making the decision publicly available. Such perverse engagement in performance management systems has been found in a number of studies, which show that public servants perceive the systems as irrelevant for their work (Lavertu, Lewis, and Moynihan 2013; Soss, Fording, and Schram 2011) and as a structural obstacle to game (Bevan and Hood 2006). If this type of behavior also characterizes managers’ goal prioritizations, we should not expect any organizational change and therefore no performance improvements. This scenario is tested in hypothesis 2. H2 Public managers’ prioritization of goals with low performance is symbolic and does not lead to performance improvements. Performance Potentials and Diminishing Returns While the hypothesis above expects managers to engage in performance management practices with a minimum of effort, a prioritization could also entail a real effort to create performance improvements. In this situation, a prioritization initiates a process that adds resources to the particular goal dimension. The resource allocation should be understood broadly, encompassing both material resources (for instance, new equipment, better facilities, or supporting staff) and activation of mental resources (for instance, motivation and abilities). No matter the specific content of the prioritization, the manifestation should follow the internal management rationales for making the prioritization. This means that prioritizations of goals with low performance start a process where the organization uses resources to analyze the reasons for the shortcomings and correct the problems (Cyert and March 1963). For goals with high performance, resources add to a process where successes are scrutinized to create knowledge that builds competences and commence specialization efforts. One thing to note about both processes is the risk that the investment does not yield performance payoff. Research on both financial (Andersen and Mortensen 2010; Hanushek 2003) and human resources (e.g., Banerjee et al. 2012) have consistently shown that the mere addition of resources to achieve a goal does not equal success. Instead, there are good reasons to believe the marginal return of resources should be taken into consideration. To understand why we need to consult the production function that describes how resources transform into outcomes on a goal dimension. This relationship is often characterized using the Cobb–Douglas production function. Here, resources have a diminishing return, meaning that addition of resources to the production increases outputs at a diminishing rate. The reason is that the benefits of the capital stock are being exploited as production increases, thereby lowering the product of each additional worker. Because of this exploration, the marginal return of resources is highest when few resources are used in the production and the corresponding output level is low. Although resources’ diminishing return typically is used to explain how companies can maximize their production of material outputs, the idea has also been applied to describe production processes in the public sector, concerning service delivery and the creation of humane outcomes (Heckman 2000). The theoretical argument for making this application is that production processes in public organizations also risk exploration of the capital stock, thereby resembling those in the private sector (Pritchett and Filmer 1999, 223–4). Substantially, the argument provides an explanation for why resources do not always yield a performance payoff. The reason is that the marginal product trends toward zero, or even becomes negative, at high production levels because the many resources cause inefficiency. While the argument for resources’ diminishing return focuses on the input side of a production, the equation could also be turned around, implying that output or outcome levels could be used to assess a production’s potential for improvement. Following this idea, goals with low performance have a great potential because the production is at a low level and accordingly, the marginal product of adding resources should be high. This expectation is tested in the following hypothesis. H3 Public managers’ prioritization of goals will improve performance, primarily for goals with an initial low level of performance. The Importance of Sustained Attention A standard way to define performance management is as an ongoing system that is bound by performance information creation, decision-making, and outcomes (Moynihan 2008). A system approach, by definition, includes a temporal component, as repeated measurements produce new information over time (van Helden, Johnsen, and Vakkuri 2012). However, this temporal aspect is a rather unexplored aspect of performance management, especially in relation to managerial decision-making. The cyclic aspect of performance systems means that an initial decision (in the first cycle) can be revisited in later cycles (Pollitt 2013). In the situation where managers prioritize resources, they face a choice in the second cycle of either reprioritizing the same goals as in the first cycle or choosing to prioritize a new set of goals. To understand the implications of this decision, it is necessary to open the black box, which has been labeled the “production process” so far. Here, a relevant point of departure is the classical input-output model (Boyne 2002). In this model, organizational activities are the key mechanism that transforms resources into performance (Boyne 2003). A prioritization could aim at improving existing activities or developing new activities. An example is the principle that prioritizes the goal “student health.” At the school, a certain amount of resources is already used on student health, such as the provision of gym classes and green areas with playgrounds. When the goal is made a priority, the principal can increase the number of gym classes or develop a new initiative. Neither solutions quick fixes because they cannot be implemented and provide a return within a day, week, or month. In this way, the example highlights two general traits of production processes in public organizations, namely that (1) they consist of a long chain of complex subprocesses and (2) there could be a substantial time lag before addition of input to a goal actually has an influence on the outcome (Ostrom et al. 1978). When these considerations are related to the cyclic aspect of performance management, it becomes evident why managers are facing an important decision in later cycles of the system. In a situation where a goal is only prioritized in one cycle, it is unlikely that the prioritization will lead to improvements, simply because the input to production processes lack the sufficient time to transform into improvements. If the manager chooses to prioritize a completely new set of goals, there is a risk that the initial prioritization will only have superficial attention without any real improvement as a result. This argument is tested in hypothesis 4. H4 A long-term strategy (repeated prioritizations of a goal) increases the likelihood of performance improvements. Performance Trade-Offs An important aspect of hypothesis 2.1 and 2.2 is what happens to the performance level for nonprioritized goals. Part of the answer to this question depends on the level of available resources in the organization. In one scenario, the organization finds itself in a situation where resources are used for activities that do not benefit the users (i.e., slack) (Migué, Bélange and Niskanen 1974). While such a situation could be the result of a deliberate managerial choice, improper use of resources could also originate at lower levels of the organization. For this reason, a problem-solving approach often starts with a managerial search for available resources that can be used to identify and implement solutions (Salge 2011). If this search ends successfully, goals with low performance could be prioritized with no consequence for activities and performance levels on nonprioritized goals. In the other scenario, all resources are used to produce outcomes on the various goal dimensions, meaning that the allocation of resources to one goal reduces the amount of inputs available to produce other goals. Performance trade-offs are likely to occur in such a situation. However, they do not necessarily cancel each other out. A production-possibility frontier illustrates why by plotting the possible outcome combinations in the production of two goals. The frontier connects the goals in a concave relationship, reflecting the increasing opportunity cost of moving resources from one goal to the other. Substantially, the relationship reflects the fact that the most general resources are relocated at first, while specialized and less effective resources are being relocated in the movement along the frontier. Figure 1 shows a frontier for two goals, one with a high-performance level (y2) and one with a low-performance level (y1) in the starting point A. When resources are allocated to the goal with low performance, the organization moves to scenario B. Δ y2A – y2B expresses the performance loss from this reallocation, and Δ y1A – y1B expresses the performance gain. As seen in the figure, the concave relationship leads the allocation of resources to y1 to exceed the performance loss for y2. Figure 1. View largeDownload slide The Production-Possibility Frontier for Goals with High and Low Performance Figure 1. View largeDownload slide The Production-Possibility Frontier for Goals with High and Low Performance In sum, the two scenarios lead to an identical expectation for how prioritizing resources to goals with low performance will influence the organization’s performance level overall, namely that the reallocation will end in a surplus. This expectation is tested in hypothesis 5. H5 The overall performance changes from prioritizing goals with low performance will end in a surplus for the organization. Research Design and Data This section presents the main elements of the research design, namely the empirical setting, data, operationalization, identification strategy, and model specifications. Setting and Data: Performance Management in Danish Public Schools During the last 20 years, a number of reforms have sought to change the Danish public schools in line with the ideas of new public management (Nielsen 2014a). In the education system, the municipalities oversee the quality of the public schools and accordingly, the implementation of the initiatives was organized here. This flexibility created distinctive performance management regimes across municipalities, varying considerably in their expression and scope. In this study, the data comes from schools in a municipality with one of the most comprehensive systems. The system consists of two-year cycles. A cycle begins with the measurement of school performance on a large number of goal dimensions. Then, the schools receive a quality report that summarizes their achievements. After the release of the quality report, the principal meets with the stakeholders of the school (representatives of the parents, municipality, and teachers) to discuss the results. The meeting is organized as a learning forum (Moynihan 2005) in which each participant can express his or her interpretation of the results. At the end of the meeting, the principal decides on a number of priorities for the school. These priorities are areas of the school’s work that should be given special attention until the beginning of the next cycle. A school has from 12 to 18 months to work on the priorities before the measurement of performance is repeated. Figure 2 illustrates the cyclic nature of the system. Figure 2. View largeDownload slide The Cyclic Nature of the System Figure 2. View largeDownload slide The Cyclic Nature of the System The system has three attractive features that make it suitable for testing the hypotheses. First, the large number of performance indicators creates a situation where principals receive multiple information signals on how their school performs. Second, the principals’ decisions have a manifest expression because the priorities are written down and made publicly available. These features make it possible to connect information and decisions, thereby enabling a test of how managers react to differences in performance levels. Finally, and as shown in table 1 below, the schools in a municipality are—on average—quite similar to the rest of Denmark on a number of parameters. These parameters are school size, budget per student, the share of students in special classes, teaching time, student/teacher ratio, and competence-coverage. Thus, any functional aspect of performance management in this setting is more likely to be caused by the system than the characteristics of the schools. Table 1. Characteristics of the Municipality—The School Year 2012/2013 Characteristics Municipality Denmark Min Max Mean Mean School size (number of students) 124 1,021 602 428 Budget pr. student (in DKK) — — 62,746 61,065 Share of students in special classes 5% 7% 4.3% 5.2% Share of bilingual students 2.7% 99.7% 25.0% ~10% Share of teachers’ time used for teaching 25.2% 38.7% 32.3% 33.7% Student/teacher ratio 9 18 14 13 Competence-coveragea 33% 100% 70.0% 80.0% Characteristics Municipality Denmark Min Max Mean Mean School size (number of students) 124 1,021 602 428 Budget pr. student (in DKK) — — 62,746 61,065 Share of students in special classes 5% 7% 4.3% 5.2% Share of bilingual students 2.7% 99.7% 25.0% ~10% Share of teachers’ time used for teaching 25.2% 38.7% 32.3% 33.7% Student/teacher ratio 9 18 14 13 Competence-coveragea 33% 100% 70.0% 80.0% Note: The table is based on data from the quality reports and the Danish ministry of education. aThe share of classes being conducted by a teacher with a specialization in the subject. View Large Table 1. Characteristics of the Municipality—The School Year 2012/2013 Characteristics Municipality Denmark Min Max Mean Mean School size (number of students) 124 1,021 602 428 Budget pr. student (in DKK) — — 62,746 61,065 Share of students in special classes 5% 7% 4.3% 5.2% Share of bilingual students 2.7% 99.7% 25.0% ~10% Share of teachers’ time used for teaching 25.2% 38.7% 32.3% 33.7% Student/teacher ratio 9 18 14 13 Competence-coveragea 33% 100% 70.0% 80.0% Characteristics Municipality Denmark Min Max Mean Mean School size (number of students) 124 1,021 602 428 Budget pr. student (in DKK) — — 62,746 61,065 Share of students in special classes 5% 7% 4.3% 5.2% Share of bilingual students 2.7% 99.7% 25.0% ~10% Share of teachers’ time used for teaching 25.2% 38.7% 32.3% 33.7% Student/teacher ratio 9 18 14 13 Competence-coveragea 33% 100% 70.0% 80.0% Note: The table is based on data from the quality reports and the Danish ministry of education. aThe share of classes being conducted by a teacher with a specialization in the subject. View Large The data covers three cycles that begin in 2009, 2011, and 2013, respectively. The unit of analysis is goals with available performance information. Since each school is measured by a number of goals, the data have a multilevel structure with goals nested within a school. Over the years, the system has increased the number of performance indicators (and goals) from 33 in 2009 to 40 in 2013.1 There are between 51 and 48 schools in the municipality in the time span from 2009 to 2013, and therefore, the number of observations varies from 1500 to 1900. Operationalization A Goal’s Relative Performance In line with the theoretical expectations, when managers face variation in performance information across different organizational goals, it should lead to decisions that give priority to the goals with the lowest level of performance. Accordingly, a key feature in the empirical test of this theoretical argument is a variable that contains information on how well a school is performing across the various educational goals. This standardized measure of relative performance is created in the following way. First, the maximum level of performance for the goal is determined. This maximum is either 12 on the grading scale or 0/100 on a percentage scale. For example, for the goal of “parental satisfaction,” the maximum is 100%, while it is 0% for the goal of “student smoking.” With the maximum level of performance in place, the following formula calculates the relative performance of a goal: pist = pabistpmaxi×100 where pist is the measure of relative performance on goal i for school s in time t. pabist and pmaxi are measures of the school’s absolute level of performance and the maximum possible level of performance on the goal, respectively. Thus, higher values on the variable indicate better performance. An alternative way to create the variable for relative performance is to use the empirical maximum value in the sample instead of the theoretical maximum. This variable has been used as a robustness check to the results from model 4 in table A2. Results from the alternative specification are almost identical (the coefficient is −0.0066 and the t-statistic is −6.48). Goal Prioritization The variable prioritization is a dichotomous variable that contains information on whether a principal (for a given school at a given time) has chosen to prioritize a goal or not. The variable was created by matching a school’s priorities—those selected by the principal after meeting with stakeholders—with the performance indicators. A goal was coded 1 if there was a match to one (or more) of the school’s priorities and the value 0 if none of the priorities related to the goal. Since priorities are chosen in an open process, there are no specific demands on the number of priorities chosen, how specific they should be, or how closely they should relate to metrics. This way of organizing the process has two implications for the operationalization. First, some priorities do not match metrics. Instead, these priorities focus on themes such as technology, community collaboration, or the school’s economy. Second, there is an element of interpretation in the matching of priorities with the relevant goals, and therefore, there was a need for creating coding rules that ensure that the matching occurred in a valid and reliable manner. Some priorities were easily connected to a goal. For example, “Improve reading in pre-preparatory classes” matches the goal “Reading test: third grade.” Other priorities were not directly related to one specific goal but, rather, to a theme (such as learning, parental collaboration, student well-being, or health) with many underlying goals. Finally, some priorities were indirectly connected to either one or several goals. In the coding of these priorities, it was necessary to include additional information that could help uncover the meaning of the priority. For the majority of schools, this information was available in the school’s strategy plan and in the following quality report in which the school describes and reflects on the work with the priorities. Since it was rather consistent what principals meant by their indirect priorities, this meaning was used to code schools with indirect priorities but with no information available. Lastly, three points should be mentioned. First, even though the aim of the coding strategy was to connect priorities with intended goals, there will be instances where the match does not reflect the principal’s intention. Since there is no reason to expect mismatches to be systematic, the problem is a matter of noise in the measurement of the variable, which may have the disadvantage that it makes it more difficult to detect relationships in the estimations. Second, in relation to the validity of the variable, the coding has been reliability tested by having a research assistant code 10% of the goals. A comparison of the two codings gives a Krippendorffs alpha of 0.83, which is considered reliable. Third is the question of what kind of organizational changes a prioritization entails. Empirically, principals add resources (through their choice of strategy) to a goal in many different ways. For example, a strategy for improving parental satisfaction could be a newsletter, increasing the number of parent–teacher meetings, or by highlighting the importance of a good relationship with the parents. Thus, a prioritization could initiate change by providing material resources or by activating mental ones. Principals describe their strategy in a development plan (which is a document with the overall strategy for the school), and superiors in the administration approve the plan. Thus, while it is not possible to create a measure of how many and what kind of resources is entailed in a prioritization, the accountability mechanisms should ensure that improvements come about through functional processes. Identification Strategies and Model Specifications The Influence of Relative Performance on Principals’ Prioritizations Estimating the relationship between relative performance and principals’ prioritizations risks endogeneity problems because factors related to the school or the principal could influence both variables. To handle this problem, I use a fixed-effects model. Formally, the model is estimated as follows: Prist= β0+ β1Pist+β2'Xist+ γs+ϕt+ εist (1) Prist is the variable that indicates whether goal i is prioritized or not at school s at time t1. Pist is the measure of relative performance on a goal i for a school s in time t0. Xist is a vector of control variables, γs is school fixed effects, ϕt is time fixed effects, β0 is the constant, and εist is the error term. In the model, time-invariant factors at the school level are handled by including the fixed-effects term for the schools (Hesketh and Skrondal 2012, 95–7, 159). However, the model still needs to include factors at the goal level that could influence performance levels and the probability of a prioritization (included in the vector Xist). Two of these control variables relate to fundamental characteristics of a goal, namely, the number of years with a metric and an indicator variable that shows whether performance is measured through a survey. Furthermore, the political attention of a goal could influence principals’ prioritizations of resources and the daily management of the school (thereby influencing performance). To account for such attention, the model includes two indicator variables, measuring whether there is a political aspiration level (decided by the city council) on a goal and whether this aspiration level has been changed throughout the three cycles. Finally, the model includes five dummies for goals that have been given special political attention at the national level (grade point average [GPA], transition to further education, truancy, inclusion, and bullying). Even though the fixed-effects model makes it possible to handle part of the omitted variable problem, it cannot account for a problem related to the cyclic nature of the data. Within one cycle, there is a sequential relationship between performance information on relative performance (t0), prioritizations (t1/4), and relative performance (t1). To address this issue, the main analysis is supplemented with analysis split by cycle (in table A2), which allows an estimation of the effect of relative performance on principals’ prioritizations independent of the possible performance effect these prioritizations might have. Another possible confounding factor is principal turnover. Principal turnover could confound the estimates if the change is somehow systematically related to both relative performance and goal prioritization. To account for this problem, the analysis has been robustness checked by splitting the sample for schools with and without a principal turnover. This analysis yields substantial identical results to the ones presented in model 4 in table A2 (the coefficient and t-statistic are −0.0037 and −4.62 for schools with no turnover and −0.0069 and −5.59 for schools with turnover).2 Equation (1) has a binary outcome, which makes the models appropriate for both a linear probability model and a logit model (Wooldridge 2009, 246–50). Because the model includes school fixed effects, a substantial number of observations drops out of the estimation if there is no variation in the dependent variable within schools. Therefore, the models are estimated with a linear probability model.3 All models are estimated with cluster-robust standard errors to account for heteroscedasticity in the error term (Angrist and Pischke 2009, 47). The Effect of Goal Prioritizations on Performance The model that estimates whether prioritized goals influence performance in subsequent years needs to address two issues. First, different factors might influence both a goal’s likelihood of being prioritized and the performance level. Second, initial performance differences between nonprioritized and prioritized goals (as expected in H1) would blur the estimation in an ordinary regression model that does not properly capture changes over time. To handle these issues, I use a difference-in-differences model. The difference-in-differences setup compares developments in the performance trends for nonprioritized and prioritized goals, thereby taking the initial performance differences between the two types of goals into account. In line with the difference-in-differences design, nonprioritized goals act as a control group that captures time trends, and prioritized goals receive a treatment in the form of a prioritization. Formally, the model is given by: Pist= β0+β1Tt+δ(Prist x Tt) +β2'Xist+ γis+ εist (2) Pist is the outcome, which is the relative performance level for goal i at school s at time t. Prist is the variable that indicates whether a goal is prioritized (i.e., the treatment variable), Tt is the time measure, and Xist is a vector of control variables.4 This vector includes two indicator variables. The first indicator variable takes the value 1 if a political aspiration level was introduced on the performance dimension, the other takes the value 1 if there was a change in the aspiration level from the previous cycle. γis is goal fixed effects (that also capture time-invariant factors at the school level), β0 the constant, and εist is the error term. δ is the difference-in-differences estimate, consisting of the following terms in a two-period setup: δ=(Pprioritized, t1−Pprioritized, t0)−(Pnonprioritized, t1−Pnonprioritized, t0) The basic model is expanded in two ways. First, treatment is normally given at one point in time (Angrist and Pischke 2009). However, the data in this study allow for repeated treatment because the performance management system entails two decision moments. Figure 3 illustrates this point. Figure 3. View largeDownload slide Time Sequence in the Difference-in-Differences Setup Figure 3. View largeDownload slide Time Sequence in the Difference-in-Differences Setup As shown in the figure, a goal could be treated in either late 2009, late 2011, or in both moments. This feature of the data is handled by creating two treatment groups, namely, goals with prioritization in one cycle (in either 2009 or 2011) and goals with prioritizations in two cycles. The two treatment groups are not collapsed in the first cycle because at the first decision moment principals might decide on a long-term strategy for the goals that are prioritized in two cycles, meaning that these goals experience a different treatment and a unique performance path. Second, the design rests on the assumption that the control group and the treatment group share a common trend before the introduction of the treatment (Wooldridge 2009, 453–4). To test this assumption, the data must include at least two pretreatment data points for the outcome variable. Thus, the time variable should consist of at least three categories, namely, two pretreatment categories and one post-treatment category (Bertrand, Duflo, and Mullainathan 2004). To accommodate this requirement, I include outcome data for 2007, meaning that t in equation (2) = {2007; 2009; 2010; 2013}.5 In a classic difference-in-differences design, the pretreatment values reflect a situation where all units of analysis are untreated. Such a scenario does not hold in the present setting as performance management reforms were introduced in the Danish educational system already in the beginning of the 2000s, and accordingly, the schools could have worked with goal priorities before the first decision moment in 2009. This aspect of the setting poses a challenge to the design on two levels. First, it could induce bias in the estimations if a systematic component characterizes pre-2009 prioritizations. The most concerning scenario is if the group of goals that were prioritized in 2009 includes a higher share of pre-2009 prioritized goals. Such a situation would make it difficult to estimate treatment effects after 2009 because the prioritized goals were already on a positive trend (due to the pre-2009 prioritization). One way to assess the problem is the test for common trends as this test indicates whether there is a systematic difference in performance trends between prioritized and nonprioritized goals leading up to the first decision moment in 2009. As shown across the models in table A4, the common trend test does not point to any concerns, showing no significant pretreatment differences. To supplement this test, I have run two additional robustness checks. A number of goals did not have a performance indicator before 2009, and thus had a low probability of being prioritized before this year. By running the analysis solely on these goals, it is possible to obtain an estimate that is less influenced by the decisions that took place before 2009. Although effects are somewhat larger (compared to model 3 in table A4, the effect is 7.94 and the t-statistic 5.09) for goals with no data before 2009, the overall conclusion is the same. The second robustness check is a generalized difference-in-differences model.6 This model includes unit and time fixed effects and estimates how changes in treatment status (i.e., whether the goal is prioritized or not) within a unit influence outcomes over time. Since this model does not rely on variation across goals that were either prioritized or nonprioritized in 2009, it is not sensitive to systematic differences between these groups before 2009. The alternative difference-in-differences specification yields substantial identical results to the ones presented in model 3 table A4 with an effect of 4.13 (t-statistic 5.51). While the three robustness checks do not point to concerns with regard to the estimates, the pretreatment issue is still relevant when interpreting treatment effects. The reason is that effects could partly reflect the work that took place before 2009, which substantially means that positive performance increases are the result of more than one or two prioritizations. Naturally, this point is worth keeping in mind in the analysis. Finally, three aspects of the model specification and analytic strategy should be noted. First, the analysis consists of three main models (shown in table A4 in the appendix). In each model, nonprioritized goals act as the control group (i.e., the reference category). Models 1 and 3 test the effect of prioritizing a goal in one cycle (2009 and 2011, respectively), and model 2 tests the effect of repeated prioritizations. Second, in addition to these models, the analysis includes two models that are split by goals within a school that have either low or high performance (divided by the median). Third, the difference-in-differences models are estimated with the full set of observations. This model is robust to a completely balanced panel-model that only includes goals with data from all four time periods (compared to model 3 in table A4, the effect is 3.54 and t-statistic 2.85). Analysis Figure 4 illustrates the relationship between relative performance and principals’ tendency to prioritize a goal. The β1 in the figure is the estimated regression coefficient from the model specified in equation (1). Figure 4. View largeDownload slide How Relative Performance Influences Principals’ Prioritization of Goals Figure 4. View largeDownload slide How Relative Performance Influences Principals’ Prioritization of Goals Based on figure 4, it is possible to evaluate the expectation in H1 that public managers will use performance information to prioritize goals with the lowest level of performance. As evident in the figure, this expectation is confirmed. β1 is significantly negative, which means that goals with lower performance are more likely to be prioritized. In substantial terms, a performance difference of 10 percentage points between two goals will lead to a 5% increase in the probability that a principal prioritizes the goal with the lowest level of performance. For a standardized outcome, the effect is 0.17 standard deviations (SDs). This estimate underlines a substantial important aspect of the analysis, namely that even though relative performance influences principals’ prioritizations, the performance differences need to be quite substantial before the information offsets a managerial reaction. Such differences are rather common for the principals as they, on average, experience a difference of 63 percentage points between the highest and lowest performing goals in their school. Across such large performance differences, the probability of prioritizing the goals with low performance is 40% higher. The last part of the analysis concerns the influence of priorities on performance in subsequent years. First is the question of whether prioritizations are merely symbolic (H2), or they actually create improvements but primarily for goals with a low initial performance level (H3). The analysis also includes a test of H4, which claims that a long-term strategy (consisting of repeated prioritizations) increases the probability of performance improvements. These hypotheses are tested in difference-in-differences models, and the results are summarized in figure 5 (the full models are shown in table A4 in the appendix). Figure 5. View largeDownload slide How Prioritization of Goals Influences Performance in Subsequent Years Figure 5. View largeDownload slide How Prioritization of Goals Influences Performance in Subsequent Years The results confirm the expectations in H3 and H4. As evident by the figure, goals that have been prioritized in two cycles experience a performance increase in the second cycle. The effect is around four scale points on the performance scale. In relation to H3, the analysis split for goals with low- and high-performance levels confirms the expectation that improvements occur for goals with a low initial level of performance (the result is statistically significant, p < .01, in a three-way interaction). Figure 5 also shows that short-term prioritizations do not create improvements. Goals that are prioritized in one cycle experience a small (insignificant) increase in performance (0.56 for goals that are prioritized in 2009 and 0.51 for goals that are prioritized in 2011) in the first cycle; however, they do not gain the most substantial performance increase because they are not reprioritized. At first sight, the effect of around four scale points might not seem overwhelming in substantial terms as the scale of relative performance ranges from 19 to 100. However, when standardizing the coefficient, the effect of 0.25 SDs presents itself as quite substantial. In relation to other studies that focus on interventions in public schools, Rockoff et al. (2012) find that providing principals with reports on teachers’ performance increases students’ math achievements by 0.053 SDs. Andersen, Humlum, and Nandrup (2016) show that increasing instruction time in schools improves students’ reading by 0.15 SDs. In relation to these findings, it is worth noting that the effect covers different strategies for creating improvements. In some instances, the schools have made major changes in their organization, facilities, and curriculum. However, improvements also reflect minor initiatives that aim specifically at the particular goal, for example, to create a newsletter to the parents. The last part of the analysis concerns H5 and the expectation that a prioritization of goals with low performance will end in a performance surplus for the organization over the years. Performance trends for nonprioritized goals are estimated by the time variable (i.e., the effect of time when treatment is at the reference category) in models 1, 2, and 3 in table A4. H5 is tested by summarizing these coefficients and comparing them to the treatment effects. When using 2009 as the reference point, the performance for nonprioritized goals increases over time, meaning that the results confirm the expectation that principals are able to improve performance on goals with low performance without substantial performance trade-offs. Conclusion This article examines how the performance of various organizational goals influence managers’ tendency to prioritize the goals and whether the prioritization of certain goals improves performance in subsequent years. The results show that principals in a municipality in Denmark prioritize the educational goals where their schools are performing the worst. However, the information has the most profound influence on prioritizations when there is a substantial discrepancy between goals with low and high performance. After a goal has been made a priority for several years, the performance level increases. As nonprioritized goals do not experience a drop in performance, the schools end in an overall performance surplus over the years, suggesting minimal performance trade-offs. These results are noteworthy in themselves, but they also have broader implications for our understanding of how performance management fits in the public sector. First, as the results show that public managers are able to engage in functional performance management practices, they naturally raise the question of what drives the purposeful use of performance information. Understanding when performance management efforts make a difference is essential to move beyond the simplistic claims that performance management does or does not work (Gerrish 2016). One apparent contextual factor in the present setting is the learning forums where managers discuss the information with stakeholders. While previous research suggests that such forums support performance information use (Moynihan 2005), there is not much knowledge of what makes them functional. The forums in this study are interesting because they strike a balance between outside involvement and managerial autonomy. The deliberation between stakeholders might help overcome some of the biases that have been shown in the individual processing of performance information (Nielsen and Moynihan 2017; Olsen 2017) while at the same time allowing managers to feel ownership of the process. This point also relates to a general trait of the system, namely, the balance between autonomy and coercion. One the one side, the system includes various checkpoints for principals, forcing them to make a decision and develop a strategy for improvements. However, they are also given autonomy in their choice of priorities and strategy. In this way, the findings point to an important balance between finding the right means to engage managers but leaving sufficient room for management (Nielsen 2014a). Finally, it is also worth mentioning that the context is just as interesting from the perspective of what is not present; there are no financial incentives at stake for the principals. Thus, the results suggest that it is possible to create a functional management process in the absence of incentives but with the presence of deliberative learning forums. A second contribution is the idea of managers as biased problem solvers. More specifically, I not only document that the level of performance affects managers’ decisions; I also document that such decisions still facilitate problem-solving by directing attention to goals where the greatest returns to investment may occur. The principals’ emphasis on negative information (low performance) is identified for one specific decision, namely, the prioritization of organizational resources and attention. It remains to be seen how managers react to the content of performance information in other settings, for example, when the information is used in the management of employees (Behn 2003). Third, an important result from this study is that performance gains require ongoing persistence. This result has two practical implications. First, there are no quick fixes in public organizations. However, a political environment may demand new initiatives each year and grow impatient if a strategy does not yield a fast payoff. Such rapid changes are likely to undercut long-term thinking and strategies that need time to bear fruit. Second, there is an inherent risk in performance management systems that attempts at improving performance are ruined by too little and too superficial attention. As the results show that one cycle (one and a half year) is probably not enough to improve performance, too frequent shifts in priorities will most likely result in wasted resources with no performance gains. If performance strategies are to pay off, they need to be deployed with a medium- and long-term perspective. I am grateful to Simon Calmar Andersen, Mads Leth Jakobsen, and Donald P. Moynihan, as well as three anonymous reviewers, for their helpful comments on drafts of this article. Earlier versions of the article were presented at the 20th International Research Society for Public Management Conference, April 2016, in Hong Kong and at the Public Management Research Conference, June 2016, in Aarhus. Footnotes 1 See table A1 in the appendix for a complete list of goals and descriptive statistics. 2 Another possible confounding factor is incrementalism in principals’ decisions (Lindblom 1959) such that decisions from the first cycle influence decisions in the following cycles. However, a lagged dependent variable in a fixed effects model is by nature endogenous and likely to cause biased estimates (Wooldridge 2002). Lagged dependent variable models are therefore used as a robustness check to the main model. The model does no point to any concern compared the results in model 4 in table A2 (the effect is −0.0047 and the t-statistic −6.92). 3 The results from a logit model are reported in table A3 in the appendix. 4 The influence of principal turnover has also been tested for this model. Compared to the effect in model 3 in table A4, the effect is 3.83 (t-statistic 3.57) for schools with no turnover and 4.51 (t-statistic 3.51) for schools with a turnover. 5 Performance data is not available for all goals in 2007. Data are available for nine goals that are related to learning in terms of grades and tests scores, parents’ perceptions of student wellbeing, and other aspects of parental satisfaction and collaboration with the school. 6 Formally described as Pit= β0+β1prioritizationit+ β2'Xist+ γi+ ϕt +εist Appendix Table A1. Descriptive Statistics for Goals Goal [min; max] Meana SDa Prioritizedb School GPA [19; 71] 54.81 8.63 0.35 School GPA—lowest fourth [59; 83] 70.73 5.37 0.54 School GPA—highest fourth [48; 98] 83.42 9.05 0.47 Test—math (3rd grade) [44; 100] 88.71 10.30 0.29 Test—reading (3rd grade) [71; 100] 94.23 5.7 0.35 Test—reading (8th grade) [38; 100] 90.71 8.75 0.35 Transition: 3 months [45; 100] 93.62 6.43 0.21 Transition: 15 months [30; 100] 82.61 14.84 0.21 Pc. satisfaction: transition [39; 84] 61.44 9.50 0.20 Sd. satisfaction: boredom [55; 96] 89.42 4.00 0.41 Pc. satisfaction: academic challenge [35; 100] 58.85 11.04 0.39 Sd. satisfaction: friends [79; 100] 96.35 4.44 0.16 Sd. satisfaction: co-determination [54; 89] 72.11 6.23 0.16 Sd. satisfaction: victim of bullying [63; 100] 93.98 4.22 0.21 Pc. satisfaction: ability to stop bullying [47;81] 69.26 8.98 0.21 Pc. satisfaction: fellowship [37; 84] 70.95 8.83 0.16 Pc. satisfaction: equal opportunities [35; 84] 51.79 10.54 0.24 Sd. satisfaction: like the class [47; 100] 94.50 8.11 0.23 Pc. satisfaction: bullying occurs [63; 97] 88.67 5.96 0.20 Truancy [53; 96] 90.03 7.81 0.21 Worrisome truancy [50; 100] 85.26 11.25 0.24 Sd. satisfaction: happiness in general [47; 100] 96.26 6.00 0.25 Sd. satisfaction: happy at the time [47; 100] 98.10 3.54 0.27 Sd. satisfaction: like the school [70; 100] 91.41 3.58 0.27 Sd. satisfaction: recognition [81; 100] 93.32 2.39 0.25 Pc. satisfaction: child’s well-being [70; 92] 83.70 4.60 0.24 Student exercise [27; 100] 75.83 22.61 0.29 Student overweight [51; 100] 84.76 7.30 0.30 Student smoking [67; 100] 95.65 4.35 0.14 Student drunkenness [64; 100] 92.29 5.60 0.16 Quality of teeth: filling [0; 79.8] 56.16 11.78 0.17 Quality of teeth: cavity [50; 100] 92.22 7.10 0.17 Pc. satisfaction: collaboration [52; 92] 75.86 6.45 0.52 Pc. satisfaction: own contribution [60; 88] 75.31 7.51 0.51 Pc. satisfaction: expectations [32; 69] 51.70 6.66 0.51 Pc. satisfaction: school satisfaction [51; 97] 77.47 8.39 0.51 Pc. satisfaction: youth center satisfaction [44; 98] 78.03 10.76 0.53 Pc. satisfaction: daily contact [43; 90] 65.73 8.73 0.51 Pc. satisfaction: involvement [23; 86] 49.15 10.19 0.51 Total 82.25 16.13 0.31 Goal [min; max] Meana SDa Prioritizedb School GPA [19; 71] 54.81 8.63 0.35 School GPA—lowest fourth [59; 83] 70.73 5.37 0.54 School GPA—highest fourth [48; 98] 83.42 9.05 0.47 Test—math (3rd grade) [44; 100] 88.71 10.30 0.29 Test—reading (3rd grade) [71; 100] 94.23 5.7 0.35 Test—reading (8th grade) [38; 100] 90.71 8.75 0.35 Transition: 3 months [45; 100] 93.62 6.43 0.21 Transition: 15 months [30; 100] 82.61 14.84 0.21 Pc. satisfaction: transition [39; 84] 61.44 9.50 0.20 Sd. satisfaction: boredom [55; 96] 89.42 4.00 0.41 Pc. satisfaction: academic challenge [35; 100] 58.85 11.04 0.39 Sd. satisfaction: friends [79; 100] 96.35 4.44 0.16 Sd. satisfaction: co-determination [54; 89] 72.11 6.23 0.16 Sd. satisfaction: victim of bullying [63; 100] 93.98 4.22 0.21 Pc. satisfaction: ability to stop bullying [47;81] 69.26 8.98 0.21 Pc. satisfaction: fellowship [37; 84] 70.95 8.83 0.16 Pc. satisfaction: equal opportunities [35; 84] 51.79 10.54 0.24 Sd. satisfaction: like the class [47; 100] 94.50 8.11 0.23 Pc. satisfaction: bullying occurs [63; 97] 88.67 5.96 0.20 Truancy [53; 96] 90.03 7.81 0.21 Worrisome truancy [50; 100] 85.26 11.25 0.24 Sd. satisfaction: happiness in general [47; 100] 96.26 6.00 0.25 Sd. satisfaction: happy at the time [47; 100] 98.10 3.54 0.27 Sd. satisfaction: like the school [70; 100] 91.41 3.58 0.27 Sd. satisfaction: recognition [81; 100] 93.32 2.39 0.25 Pc. satisfaction: child’s well-being [70; 92] 83.70 4.60 0.24 Student exercise [27; 100] 75.83 22.61 0.29 Student overweight [51; 100] 84.76 7.30 0.30 Student smoking [67; 100] 95.65 4.35 0.14 Student drunkenness [64; 100] 92.29 5.60 0.16 Quality of teeth: filling [0; 79.8] 56.16 11.78 0.17 Quality of teeth: cavity [50; 100] 92.22 7.10 0.17 Pc. satisfaction: collaboration [52; 92] 75.86 6.45 0.52 Pc. satisfaction: own contribution [60; 88] 75.31 7.51 0.51 Pc. satisfaction: expectations [32; 69] 51.70 6.66 0.51 Pc. satisfaction: school satisfaction [51; 97] 77.47 8.39 0.51 Pc. satisfaction: youth center satisfaction [44; 98] 78.03 10.76 0.53 Pc. satisfaction: daily contact [43; 90] 65.73 8.73 0.51 Pc. satisfaction: involvement [23; 86] 49.15 10.19 0.51 Total 82.25 16.13 0.31 aRelative performance. bThe relative number of prioritizations, across all years. cParental. dStudent. Table A1. Descriptive Statistics for Goals Goal [min; max] Meana SDa Prioritizedb School GPA [19; 71] 54.81 8.63 0.35 School GPA—lowest fourth [59; 83] 70.73 5.37 0.54 School GPA—highest fourth [48; 98] 83.42 9.05 0.47 Test—math (3rd grade) [44; 100] 88.71 10.30 0.29 Test—reading (3rd grade) [71; 100] 94.23 5.7 0.35 Test—reading (8th grade) [38; 100] 90.71 8.75 0.35 Transition: 3 months [45; 100] 93.62 6.43 0.21 Transition: 15 months [30; 100] 82.61 14.84 0.21 Pc. satisfaction: transition [39; 84] 61.44 9.50 0.20 Sd. satisfaction: boredom [55; 96] 89.42 4.00 0.41 Pc. satisfaction: academic challenge [35; 100] 58.85 11.04 0.39 Sd. satisfaction: friends [79; 100] 96.35 4.44 0.16 Sd. satisfaction: co-determination [54; 89] 72.11 6.23 0.16 Sd. satisfaction: victim of bullying [63; 100] 93.98 4.22 0.21 Pc. satisfaction: ability to stop bullying [47;81] 69.26 8.98 0.21 Pc. satisfaction: fellowship [37; 84] 70.95 8.83 0.16 Pc. satisfaction: equal opportunities [35; 84] 51.79 10.54 0.24 Sd. satisfaction: like the class [47; 100] 94.50 8.11 0.23 Pc. satisfaction: bullying occurs [63; 97] 88.67 5.96 0.20 Truancy [53; 96] 90.03 7.81 0.21 Worrisome truancy [50; 100] 85.26 11.25 0.24 Sd. satisfaction: happiness in general [47; 100] 96.26 6.00 0.25 Sd. satisfaction: happy at the time [47; 100] 98.10 3.54 0.27 Sd. satisfaction: like the school [70; 100] 91.41 3.58 0.27 Sd. satisfaction: recognition [81; 100] 93.32 2.39 0.25 Pc. satisfaction: child’s well-being [70; 92] 83.70 4.60 0.24 Student exercise [27; 100] 75.83 22.61 0.29 Student overweight [51; 100] 84.76 7.30 0.30 Student smoking [67; 100] 95.65 4.35 0.14 Student drunkenness [64; 100] 92.29 5.60 0.16 Quality of teeth: filling [0; 79.8] 56.16 11.78 0.17 Quality of teeth: cavity [50; 100] 92.22 7.10 0.17 Pc. satisfaction: collaboration [52; 92] 75.86 6.45 0.52 Pc. satisfaction: own contribution [60; 88] 75.31 7.51 0.51 Pc. satisfaction: expectations [32; 69] 51.70 6.66 0.51 Pc. satisfaction: school satisfaction [51; 97] 77.47 8.39 0.51 Pc. satisfaction: youth center satisfaction [44; 98] 78.03 10.76 0.53 Pc. satisfaction: daily contact [43; 90] 65.73 8.73 0.51 Pc. satisfaction: involvement [23; 86] 49.15 10.19 0.51 Total 82.25 16.13 0.31 Goal [min; max] Meana SDa Prioritizedb School GPA [19; 71] 54.81 8.63 0.35 School GPA—lowest fourth [59; 83] 70.73 5.37 0.54 School GPA—highest fourth [48; 98] 83.42 9.05 0.47 Test—math (3rd grade) [44; 100] 88.71 10.30 0.29 Test—reading (3rd grade) [71; 100] 94.23 5.7 0.35 Test—reading (8th grade) [38; 100] 90.71 8.75 0.35 Transition: 3 months [45; 100] 93.62 6.43 0.21 Transition: 15 months [30; 100] 82.61 14.84 0.21 Pc. satisfaction: transition [39; 84] 61.44 9.50 0.20 Sd. satisfaction: boredom [55; 96] 89.42 4.00 0.41 Pc. satisfaction: academic challenge [35; 100] 58.85 11.04 0.39 Sd. satisfaction: friends [79; 100] 96.35 4.44 0.16 Sd. satisfaction: co-determination [54; 89] 72.11 6.23 0.16 Sd. satisfaction: victim of bullying [63; 100] 93.98 4.22 0.21 Pc. satisfaction: ability to stop bullying [47;81] 69.26 8.98 0.21 Pc. satisfaction: fellowship [37; 84] 70.95 8.83 0.16 Pc. satisfaction: equal opportunities [35; 84] 51.79 10.54 0.24 Sd. satisfaction: like the class [47; 100] 94.50 8.11 0.23 Pc. satisfaction: bullying occurs [63; 97] 88.67 5.96 0.20 Truancy [53; 96] 90.03 7.81 0.21 Worrisome truancy [50; 100] 85.26 11.25 0.24 Sd. satisfaction: happiness in general [47; 100] 96.26 6.00 0.25 Sd. satisfaction: happy at the time [47; 100] 98.10 3.54 0.27 Sd. satisfaction: like the school [70; 100] 91.41 3.58 0.27 Sd. satisfaction: recognition [81; 100] 93.32 2.39 0.25 Pc. satisfaction: child’s well-being [70; 92] 83.70 4.60 0.24 Student exercise [27; 100] 75.83 22.61 0.29 Student overweight [51; 100] 84.76 7.30 0.30 Student smoking [67; 100] 95.65 4.35 0.14 Student drunkenness [64; 100] 92.29 5.60 0.16 Quality of teeth: filling [0; 79.8] 56.16 11.78 0.17 Quality of teeth: cavity [50; 100] 92.22 7.10 0.17 Pc. satisfaction: collaboration [52; 92] 75.86 6.45 0.52 Pc. satisfaction: own contribution [60; 88] 75.31 7.51 0.51 Pc. satisfaction: expectations [32; 69] 51.70 6.66 0.51 Pc. satisfaction: school satisfaction [51; 97] 77.47 8.39 0.51 Pc. satisfaction: youth center satisfaction [44; 98] 78.03 10.76 0.53 Pc. satisfaction: daily contact [43; 90] 65.73 8.73 0.51 Pc. satisfaction: involvement [23; 86] 49.15 10.19 0.51 Total 82.25 16.13 0.31 aRelative performance. bThe relative number of prioritizations, across all years. cParental. dStudent. Table A2. How Relative Performance Influences Principals’ Prioritization of Goals Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.0035* (−2.66) −0.0055*** (−4.92) −0.0043*** (−4.48) −0.0050*** (−7.06) Survey indicator 0.086* (2.17) −0.030 (−0.67) −0.12** (−3.18) −0.013 (−0.48) Number of years measured 0.030** (2.80) 0.026** (2.91) 0.012 (1.75) 0.029*** (5.29) Political aspiration level 0.0021 (0.09) −0.14* (−2.44) −0.12** (−3.01) −0.069* (−2.07) Change in aspiration level 0.028 (0.61) −0.029 (−0.82) −0.046* (−2.51) Goal: inclusion −0.22*** (−4.01) 0.073 (1.06) −0.039 (−0.77) −0.078* (−2.36) Goal: GPA −0.096 (−1.37) −0.18*** (−3.86) −0.057 (−0.91) −0.13** (−3.00) Goal: transition to further education −0.23*** (−4.19) −0.0082 (−0.09) −0.13 (−1.81) −0.14** (−3.16) Goal: bullying −0.33*** (−4.41) −0.12 (−1.49) −0.19*** (−3.82) −0.18*** (−5.20) Goal: truancy 0.039 (0.50) −0.21** (−3.12) −0.19* (−2.13) −0.10* (−2.04) Cycle: 2009 Reference Cycle: 2011 0.10** (2.80) Cycle: 2013 0.081 (1.94) Constant 0.43*** (4.46) 0.83*** (7.96) 0.83*** (7.52) 0.64*** (10.12) N (schools) 49 50 47 53 N (goals) 1,519 1,839 1,763 5,121 F-test 8.76*** 8.12*** 6.24*** 11.81*** Adjusted R2 0.062 0.054 0.033 0.036 Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.0035* (−2.66) −0.0055*** (−4.92) −0.0043*** (−4.48) −0.0050*** (−7.06) Survey indicator 0.086* (2.17) −0.030 (−0.67) −0.12** (−3.18) −0.013 (−0.48) Number of years measured 0.030** (2.80) 0.026** (2.91) 0.012 (1.75) 0.029*** (5.29) Political aspiration level 0.0021 (0.09) −0.14* (−2.44) −0.12** (−3.01) −0.069* (−2.07) Change in aspiration level 0.028 (0.61) −0.029 (−0.82) −0.046* (−2.51) Goal: inclusion −0.22*** (−4.01) 0.073 (1.06) −0.039 (−0.77) −0.078* (−2.36) Goal: GPA −0.096 (−1.37) −0.18*** (−3.86) −0.057 (−0.91) −0.13** (−3.00) Goal: transition to further education −0.23*** (−4.19) −0.0082 (−0.09) −0.13 (−1.81) −0.14** (−3.16) Goal: bullying −0.33*** (−4.41) −0.12 (−1.49) −0.19*** (−3.82) −0.18*** (−5.20) Goal: truancy 0.039 (0.50) −0.21** (−3.12) −0.19* (−2.13) −0.10* (−2.04) Cycle: 2009 Reference Cycle: 2011 0.10** (2.80) Cycle: 2013 0.081 (1.94) Constant 0.43*** (4.46) 0.83*** (7.96) 0.83*** (7.52) 0.64*** (10.12) N (schools) 49 50 47 53 N (goals) 1,519 1,839 1,763 5,121 F-test 8.76*** 8.12*** 6.24*** 11.81*** Adjusted R2 0.062 0.054 0.033 0.036 Note: OLS—schools fixed effects. Clustered standard errors. t-Statistics in parenthesis. *p < .05; **p < .01; ***p < .001 (two-sided test). Table A2. How Relative Performance Influences Principals’ Prioritization of Goals Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.0035* (−2.66) −0.0055*** (−4.92) −0.0043*** (−4.48) −0.0050*** (−7.06) Survey indicator 0.086* (2.17) −0.030 (−0.67) −0.12** (−3.18) −0.013 (−0.48) Number of years measured 0.030** (2.80) 0.026** (2.91) 0.012 (1.75) 0.029*** (5.29) Political aspiration level 0.0021 (0.09) −0.14* (−2.44) −0.12** (−3.01) −0.069* (−2.07) Change in aspiration level 0.028 (0.61) −0.029 (−0.82) −0.046* (−2.51) Goal: inclusion −0.22*** (−4.01) 0.073 (1.06) −0.039 (−0.77) −0.078* (−2.36) Goal: GPA −0.096 (−1.37) −0.18*** (−3.86) −0.057 (−0.91) −0.13** (−3.00) Goal: transition to further education −0.23*** (−4.19) −0.0082 (−0.09) −0.13 (−1.81) −0.14** (−3.16) Goal: bullying −0.33*** (−4.41) −0.12 (−1.49) −0.19*** (−3.82) −0.18*** (−5.20) Goal: truancy 0.039 (0.50) −0.21** (−3.12) −0.19* (−2.13) −0.10* (−2.04) Cycle: 2009 Reference Cycle: 2011 0.10** (2.80) Cycle: 2013 0.081 (1.94) Constant 0.43*** (4.46) 0.83*** (7.96) 0.83*** (7.52) 0.64*** (10.12) N (schools) 49 50 47 53 N (goals) 1,519 1,839 1,763 5,121 F-test 8.76*** 8.12*** 6.24*** 11.81*** Adjusted R2 0.062 0.054 0.033 0.036 Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.0035* (−2.66) −0.0055*** (−4.92) −0.0043*** (−4.48) −0.0050*** (−7.06) Survey indicator 0.086* (2.17) −0.030 (−0.67) −0.12** (−3.18) −0.013 (−0.48) Number of years measured 0.030** (2.80) 0.026** (2.91) 0.012 (1.75) 0.029*** (5.29) Political aspiration level 0.0021 (0.09) −0.14* (−2.44) −0.12** (−3.01) −0.069* (−2.07) Change in aspiration level 0.028 (0.61) −0.029 (−0.82) −0.046* (−2.51) Goal: inclusion −0.22*** (−4.01) 0.073 (1.06) −0.039 (−0.77) −0.078* (−2.36) Goal: GPA −0.096 (−1.37) −0.18*** (−3.86) −0.057 (−0.91) −0.13** (−3.00) Goal: transition to further education −0.23*** (−4.19) −0.0082 (−0.09) −0.13 (−1.81) −0.14** (−3.16) Goal: bullying −0.33*** (−4.41) −0.12 (−1.49) −0.19*** (−3.82) −0.18*** (−5.20) Goal: truancy 0.039 (0.50) −0.21** (−3.12) −0.19* (−2.13) −0.10* (−2.04) Cycle: 2009 Reference Cycle: 2011 0.10** (2.80) Cycle: 2013 0.081 (1.94) Constant 0.43*** (4.46) 0.83*** (7.96) 0.83*** (7.52) 0.64*** (10.12) N (schools) 49 50 47 53 N (goals) 1,519 1,839 1,763 5,121 F-test 8.76*** 8.12*** 6.24*** 11.81*** Adjusted R2 0.062 0.054 0.033 0.036 Note: OLS—schools fixed effects. Clustered standard errors. t-Statistics in parenthesis. *p < .05; **p < .01; ***p < .001 (two-sided test). Table A3. How Relative Performance Influence Principals’ Prioritization of Goals Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.019*** (−4.04) −0.029*** (−6.36) −0.022*** (−4.50) −0.024*** (−9.44) Survey indicator 0.50** (3.11) −0.16 (−1.06) −0.63*** (−4.31) −0.061 (−0.75) Number of years measured 0.18** (2.73) 0.14** (2.65) 0.061 (1.24) 0.15*** (4.94) Political aspiration level −0.027 (−0.10) −0.71*** (−4.50) −0.60*** (−3.40) −0.31** (−3.15) Change in aspiration level 0.16 (0.60) −0.14 (−0.80) −0.23* (−2.55) Goal: inclusion −2.25** (−3.02) 0.39 (1.07) −0.32 (−0.76) −0.53* (−2.20) Goal: GPA −0.45 (−0.98) −0.93* (−2.29) −0.33 (−0.80) −0.65** (−2.73) Goal: transition to further education −2.21*** (−4.05) −0.028 (−0.10) −0.70* (−2.36) −0.77*** (−4.35) Goal: bullying −2.07*** (−4.03) −0.64 (−1.61) −1.15* (−2.55) −0.96*** (−4.18) Goal: truancy 0.24 (0.83) −1.35*** (−3.75) −1.13** (−3.28) −0.56** (−3.06) Cycle: 2009 Reference Cycle: 2011 0.49*** (4.68) Cycle: 2013 0.39*** (3.54) N (schools) 45 49 45 52 N (goals) 1,402 1,801 1,687 5,083 Log likelihood −637.39 −884.47 −852.23 −2770.02 Likelihood ratio 112.76*** 108.44*** 67.48*** 194.53*** Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.019*** (−4.04) −0.029*** (−6.36) −0.022*** (−4.50) −0.024*** (−9.44) Survey indicator 0.50** (3.11) −0.16 (−1.06) −0.63*** (−4.31) −0.061 (−0.75) Number of years measured 0.18** (2.73) 0.14** (2.65) 0.061 (1.24) 0.15*** (4.94) Political aspiration level −0.027 (−0.10) −0.71*** (−4.50) −0.60*** (−3.40) −0.31** (−3.15) Change in aspiration level 0.16 (0.60) −0.14 (−0.80) −0.23* (−2.55) Goal: inclusion −2.25** (−3.02) 0.39 (1.07) −0.32 (−0.76) −0.53* (−2.20) Goal: GPA −0.45 (−0.98) −0.93* (−2.29) −0.33 (−0.80) −0.65** (−2.73) Goal: transition to further education −2.21*** (−4.05) −0.028 (−0.10) −0.70* (−2.36) −0.77*** (−4.35) Goal: bullying −2.07*** (−4.03) −0.64 (−1.61) −1.15* (−2.55) −0.96*** (−4.18) Goal: truancy 0.24 (0.83) −1.35*** (−3.75) −1.13** (−3.28) −0.56** (−3.06) Cycle: 2009 Reference Cycle: 2011 0.49*** (4.68) Cycle: 2013 0.39*** (3.54) N (schools) 45 49 45 52 N (goals) 1,402 1,801 1,687 5,083 Log likelihood −637.39 −884.47 −852.23 −2770.02 Likelihood ratio 112.76*** 108.44*** 67.48*** 194.53*** Note: Logistic regression—schools fixed effects. t-Statistics in parenthesis. *p < .05; **p < .01; ***p < .001 (two-sided test). Table A3. How Relative Performance Influence Principals’ Prioritization of Goals Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.019*** (−4.04) −0.029*** (−6.36) −0.022*** (−4.50) −0.024*** (−9.44) Survey indicator 0.50** (3.11) −0.16 (−1.06) −0.63*** (−4.31) −0.061 (−0.75) Number of years measured 0.18** (2.73) 0.14** (2.65) 0.061 (1.24) 0.15*** (4.94) Political aspiration level −0.027 (−0.10) −0.71*** (−4.50) −0.60*** (−3.40) −0.31** (−3.15) Change in aspiration level 0.16 (0.60) −0.14 (−0.80) −0.23* (−2.55) Goal: inclusion −2.25** (−3.02) 0.39 (1.07) −0.32 (−0.76) −0.53* (−2.20) Goal: GPA −0.45 (−0.98) −0.93* (−2.29) −0.33 (−0.80) −0.65** (−2.73) Goal: transition to further education −2.21*** (−4.05) −0.028 (−0.10) −0.70* (−2.36) −0.77*** (−4.35) Goal: bullying −2.07*** (−4.03) −0.64 (−1.61) −1.15* (−2.55) −0.96*** (−4.18) Goal: truancy 0.24 (0.83) −1.35*** (−3.75) −1.13** (−3.28) −0.56** (−3.06) Cycle: 2009 Reference Cycle: 2011 0.49*** (4.68) Cycle: 2013 0.39*** (3.54) N (schools) 45 49 45 52 N (goals) 1,402 1,801 1,687 5,083 Log likelihood −637.39 −884.47 −852.23 −2770.02 Likelihood ratio 112.76*** 108.44*** 67.48*** 194.53*** Model 1 2 3 4 Cycle 2009 2011 2013 All Relative performance −0.019*** (−4.04) −0.029*** (−6.36) −0.022*** (−4.50) −0.024*** (−9.44) Survey indicator 0.50** (3.11) −0.16 (−1.06) −0.63*** (−4.31) −0.061 (−0.75) Number of years measured 0.18** (2.73) 0.14** (2.65) 0.061 (1.24) 0.15*** (4.94) Political aspiration level −0.027 (−0.10) −0.71*** (−4.50) −0.60*** (−3.40) −0.31** (−3.15) Change in aspiration level 0.16 (0.60) −0.14 (−0.80) −0.23* (−2.55) Goal: inclusion −2.25** (−3.02) 0.39 (1.07) −0.32 (−0.76) −0.53* (−2.20) Goal: GPA −0.45 (−0.98) −0.93* (−2.29) −0.33 (−0.80) −0.65** (−2.73) Goal: transition to further education −2.21*** (−4.05) −0.028 (−0.10) −0.70* (−2.36) −0.77*** (−4.35) Goal: bullying −2.07*** (−4.03) −0.64 (−1.61) −1.15* (−2.55) −0.96*** (−4.18) Goal: truancy 0.24 (0.83) −1.35*** (−3.75) −1.13** (−3.28) −0.56** (−3.06) Cycle: 2009 Reference Cycle: 2011 0.49*** (4.68) Cycle: 2013 0.39*** (3.54) N (schools) 45 49 45 52 N (goals) 1,402 1,801 1,687 5,083 Log likelihood −637.39 −884.47 −852.23 −2770.02 Likelihood ratio 112.76*** 108.44*** 67.48*** 194.53*** Note: Logistic regression—schools fixed effects. t-Statistics in parenthesis. *p < .05; **p < .01; ***p < .001 (two-sided test). Table A4. How Prioritization of Goals Influences Performance in Subsequent Years Model 1 2 3 3.1 3.2 Goals are prioritized in One cycle (09–11) One cycle (11–13) Two cycles (09–13) Two cycles (09–13) Two cycles (09–13) Sample All goals All goals All goals Low-performance goals High-performance goals Time  2007 −0.72 (−1.17) −1.97** (−3.02) −0.66 (−1.14) −2.24** (−2.85) −4.21*** (−9.62)  2009 Ref. (.) −1.27*** (−3.33) Ref. (.) Ref. (.) Ref. (.)  2011 1.36*** (3.58) Ref. (.) 1.14** (3.17) 2.43*** (5.03) −0.23 (−0.85)  2013 0.21 (0.57) −1.03* (−2.53) 0.28 (0.80) 4.42*** (8.16) −0.31 (−1.35) Diff-in-diffs estimates  Prioritization × 2007 −1.83 (−1.46) 0.52 (0.47) −1.19 (−1.11) 0.46 (0.37) 1.07 (1.07)  Prioritization × 2009 Ref. (.) 0.78 (1.16) Ref. (.) Ref. (.) Ref. (.)  Prioritization × 2011 0.56 (0.75) Ref. (.) 1.10 (1.38) 1.34 (1.47) 0.012 (0.02)  Prioritization × 2013 0.21 (0.27) 0.51 (0.78) 4.07*** (4.97) 3.10** (3.25) 1.07 (1.48) Constant 81.9*** (349.23) 82.3*** (270.62) 80.6*** (353.92) 66.1*** (191.91) 93.6*** (615.52) Controls Yes Yes Yes Yes Yes N 4,165 4,491 4,028 1,958 2,070 F-test 5.04*** 2.56** 10.38*** 29.04*** 15.27*** Adjusted R2 0.014 0.0040 0.006 0.014 0.11 Model 1 2 3 3.1 3.2 Goals are prioritized in One cycle (09–11) One cycle (11–13) Two cycles (09–13) Two cycles (09–13) Two cycles (09–13) Sample All goals All goals All goals Low-performance goals High-performance goals Time  2007 −0.72 (−1.17) −1.97** (−3.02) −0.66 (−1.14) −2.24** (−2.85) −4.21*** (−9.62)  2009 Ref. (.) −1.27*** (−3.33) Ref. (.) Ref. (.) Ref. (.)  2011 1.36*** (3.58) Ref. (.) 1.14** (3.17) 2.43*** (5.03) −0.23 (−0.85)  2013 0.21 (0.57) −1.03* (−2.53) 0.28 (0.80) 4.42*** (8.16) −0.31 (−1.35) Diff-in-diffs estimates  Prioritization × 2007 −1.83 (−1.46) 0.52 (0.47) −1.19 (−1.11) 0.46 (0.37) 1.07 (1.07)  Prioritization × 2009 Ref. (.) 0.78 (1.16) Ref. (.) Ref. (.) Ref. (.)  Prioritization × 2011 0.56 (0.75) Ref. (.) 1.10 (1.38) 1.34 (1.47) 0.012 (0.02)  Prioritization × 2013 0.21 (0.27) 0.51 (0.78) 4.07*** (4.97) 3.10** (3.25) 1.07 (1.48) Constant 81.9*** (349.23) 82.3*** (270.62) 80.6*** (353.92) 66.1*** (191.91) 93.6*** (615.52) Controls Yes Yes Yes Yes Yes N 4,165 4,491 4,028 1,958 2,070 F-test 5.04*** 2.56** 10.38*** 29.04*** 15.27*** Adjusted R2 0.014 0.0040 0.006 0.014 0.11 Note: Difference-in-differences estimates—OLS (goal fixed effects). t-Statistics in parenthesis. Coefficients that test for a common trend are in italics. The models include controls for whether a political aspiration level was introduced on the performance dimension and whether this level changed. *p < .05; **p < .01; ***p < .001 (two-sided test). Table A4. How Prioritization of Goals Influences Performance in Subsequent Years Model 1 2 3 3.1 3.2 Goals are prioritized in One cycle (09–11) One cycle (11–13) Two cycles (09–13) Two cycles (09–13) Two cycles (09–13) Sample All goals All goals All goals Low-performance goals High-performance goals Time  2007 −0.72 (−1.17) −1.97** (−3.02) −0.66 (−1.14) −2.24** (−2.85) −4.21*** (−9.62)  2009 Ref. (.) −1.27*** (−3.33) Ref. (.) Ref. (.) Ref. (.)  2011 1.36*** (3.58) Ref. (.) 1.14** (3.17) 2.43*** (5.03) −0.23 (−0.85)  2013 0.21 (0.57) −1.03* (−2.53) 0.28 (0.80) 4.42*** (8.16) −0.31 (−1.35) Diff-in-diffs estimates  Prioritization × 2007 −1.83 (−1.46) 0.52 (0.47) −1.19 (−1.11) 0.46 (0.37) 1.07 (1.07)  Prioritization × 2009 Ref. (.) 0.78 (1.16) Ref. (.) Ref. (.) Ref. (.)  Prioritization × 2011 0.56 (0.75) Ref. (.) 1.10 (1.38) 1.34 (1.47) 0.012 (0.02)  Prioritization × 2013 0.21 (0.27) 0.51 (0.78) 4.07*** (4.97) 3.10** (3.25) 1.07 (1.48) Constant 81.9*** (349.23) 82.3*** (270.62) 80.6*** (353.92) 66.1*** (191.91) 93.6*** (615.52) Controls Yes Yes Yes Yes Yes N 4,165 4,491 4,028 1,958 2,070 F-test 5.04*** 2.56** 10.38*** 29.04*** 15.27*** Adjusted R2 0.014 0.0040 0.006 0.014 0.11 Model 1 2 3 3.1 3.2 Goals are prioritized in One cycle (09–11) One cycle (11–13) Two cycles (09–13) Two cycles (09–13) Two cycles (09–13) Sample All goals All goals All goals Low-performance goals High-performance goals Time  2007 −0.72 (−1.17) −1.97** (−3.02) −0.66 (−1.14) −2.24** (−2.85) −4.21*** (−9.62)  2009 Ref. (.) −1.27*** (−3.33) Ref. (.) Ref. (.) Ref. (.)  2011 1.36*** (3.58) Ref. (.) 1.14** (3.17) 2.43*** (5.03) −0.23 (−0.85)  2013 0.21 (0.57) −1.03* (−2.53) 0.28 (0.80) 4.42*** (8.16) −0.31 (−1.35) Diff-in-diffs estimates  Prioritization × 2007 −1.83 (−1.46) 0.52 (0.47) −1.19 (−1.11) 0.46 (0.37) 1.07 (1.07)  Prioritization × 2009 Ref. (.) 0.78 (1.16) Ref. (.) Ref. (.) Ref. (.)  Prioritization × 2011 0.56 (0.75) Ref. (.) 1.10 (1.38) 1.34 (1.47) 0.012 (0.02)  Prioritization × 2013 0.21 (0.27) 0.51 (0.78) 4.07*** (4.97) 3.10** (3.25) 1.07 (1.48) Constant 81.9*** (349.23) 82.3*** (270.62) 80.6*** (353.92) 66.1*** (191.91) 93.6*** (615.52) Controls Yes Yes Yes Yes Yes N 4,165 4,491 4,028 1,958 2,070 F-test 5.04*** 2.56** 10.38*** 29.04*** 15.27*** Adjusted R2 0.014 0.0040 0.006 0.014 0.11 Note: Difference-in-differences estimates—OLS (goal fixed effects). t-Statistics in parenthesis. Coefficients that test for a common trend are in italics. The models include controls for whether a political aspiration level was introduced on the performance dimension and whether this level changed. *p < .05; **p < .01; ***p < .001 (two-sided test). References Andersen , Simon Calmar . 2008 . The impact of public management reforms on student performance in Danish schools . Public Administration 86 : 541 – 58 . Google Scholar CrossRef Search ADS Andersen , Simon Calmar , and Peter B . Mortensen . 2010 . “ Policy Stability and Organizational Performance: Is There a Relationship ?” Journal of Public Administration Research and Theory 20 : 1 – 22 . Google Scholar CrossRef Search ADS Andersen , Simon Calmar , and Donald P. Moynihan . 2016 . How leaders respond to diversity: The moderating role of organizational culture on performance information use . Journal of Public Administration Research and Theory 26 : 448 – 60 . Google Scholar CrossRef Search ADS Andersen , Simon Calmar , Maria K. Humlum , and Anne B. Nandrup . 2016 . Increasing instruction time in school does increase learning . Proceedings of the National Academy of Sciences of the United States of America 113 : 7481 – 4 . Google Scholar CrossRef Search ADS PubMed Angrist , Joshua D. , and Jörn-Steffen Pischke . 2009 . Mostly harmless econometrics: An empiricists companion . Princeton : Princeton Univ. Press . Askim , Jostein , Åge Johnsen , and Knut-Andreas Christophersen . 2008 . Factors behind organizational learning from benchmarking: Experiences from Norwegian municipal benchmarking networks . Journal of Public Administration Research and Theory 18 : 297 – 320 . Google Scholar CrossRef Search ADS Audia , Pino G. , and Sebastian Brion . 2007 . Reluctant to change: Self-enhancing responses to diverging performance measures . Organizational Behavior and Human Decision Processes 102 : 255 – 69 . Google Scholar CrossRef Search ADS Audia , Pino G. , Sebastian Brion , and Heinrich R. Greve . 2015 . Self-assessment, self-enhancement, and the choice of comparison organizations for evaluating organizational performance . Advances in Strategic Management: Cognition and Strategy 32 : 89 – 118 . Google Scholar CrossRef Search ADS Banerjee , Abhijit , Raghabendra Chattopadhyay , Esther Duflo , Daniel Keniston , and Nina Singh . 2014 . Can institutions be reformed from within? Evidence from a randomized experiment with the Rajasthan Police . Working paper. Baum , Joel A. , and Kristina B. Dahlin . 2007 . Aspiration performance and railroads’ patterns of learning from train wrecks and crashes . Organization Science 18 : 368 – 85 . Google Scholar CrossRef Search ADS Behn , Robert D . 2003 . Why measure performance? Different purposes require different measures . Public Administration Review 63 : 586 – 606 . Google Scholar CrossRef Search ADS Bertrand , Marianne , Esther Duflo , and Sendhil Mullainathan . 2004 . How much should we trust differences-in-differences estimates ? Quarterly Journal of Economics 119 : 249 – 75 . Google Scholar CrossRef Search ADS Bevan , Gwyn , and Christopher Hood . 2006 . What’s measured is what matters: Targets and gaming in the English public healthcare system . Public Administration 84 : 517 – 38 . Google Scholar CrossRef Search ADS Bourdeaux , Carolyn , and Grace Chikoto . 2008 . Legislative influences on performance management reform . Public Administration Review 68 : 253 – 65 . Google Scholar CrossRef Search ADS Boyne , George A . 2002 . Concepts and indicators of local authority performance: An evaluation of the statutory frameworks in England and Wales . Public Money and Management 22 : 17 – 24 . Google Scholar CrossRef Search ADS ———. 2003 . What is public service improvement ? Public Administration 81 : 211 – 27 . CrossRef Search ADS Busuioc , Madalina , and Martin Lodge . 2017 . Reputation and accountability relationships: Managing accountability expectations through reputation . Public Administration Review 77 : 91 – 100 . Google Scholar CrossRef Search ADS Carpenter , Daniel . 2001 . The forging of bureaucratic autonomy: Reputations, networks, and policy innovation in executive agencies . Princeton, NJ : Princeton Univ. Press ———. 2010 . Reputation and power: Organizational image and pharmaceutical regulation at the FDA . Princeton, NJ: Princeton Univ. Press Christensen , Julian , Casper Mondrup Dahlmann , Asbjørn Hovgaard Mathiasen , Donald P. Moynihan , and Niels Bjørn Grund Petersen . 2018 . How do elected officials evaluate performance? Goal preferences, governance preferences, and the process of goal reprioritization . Journal of Public Administration Research and Theory , 28:197–211. Cyert , Ricard M. , and James G. March . 1963 . A behavioral theory of the firm . Englewood Cliffs, NJ : Prentice-Hall . Diwas , K. C. , Bradley R. Staats , and Francesca Gino . 2013 . Learning from my success and from others’ failure: Evidence from minimally invasive cardiac surgery . Management Science 59 : 2435 – 49 . Google Scholar CrossRef Search ADS Feldman , Martha S. , and James G. March . 1981 . Information in organizations as signal and symbol . Administrative Science Quarterly 26 : 171 – 86 . Google Scholar CrossRef Search ADS Gerrish , ED . 2016 . The impact of performance management on performance in public organizations: A meta-analysis . Public Administration Review 76 : 48 – 66 . Google Scholar CrossRef Search ADS Gilmour , John B. , and David E. Lewis . 2006 . Does performance budgeting work? An examination of the office of management and budget’s PART scores . Public Administration Review 66 : 742 – 52 . Google Scholar CrossRef Search ADS Hanushek , Eric A . 2003 . The failure of input-based schooling policies . The Economic Journal 113 : 64 – 98 . Google Scholar CrossRef Search ADS Heckman , James J . 2000 . Policies to foster human capital . Research in Economics 54 : 3 – 56 . Google Scholar CrossRef Search ADS Heinrich , Carolyn J . 1999 . Do government bureaucrats make effective use of performance management information ? Journal of Public Administration Research and Theory 9 : 363 – 93 . Google Scholar CrossRef Search ADS ———. 2010 . Third-party governance under no child left behind: Accountability and performance management challenges in supplemental educational services provision . Journal of Public Administration Research and Theory 20 : 59 – 80 . CrossRef Search ADS Hesketh , Sophia Rabe , and Anders Skrondal . 2012 . Multilevel and longitudinal modeling using Stata , 3rd ed . College Station, TX : Stata Press Publication . Hood , Christopher . 2011 . The blame game: Spin, bureaucracy, and self-preservation in government . Princeton, NJ : Princeton Univ. Press . Jordan , Alexander H. , and Pino G. Audia . 2012 . Self-enhancement and learning from performance feedback . Academy of Management Review 37 : 211 – 31 . Google Scholar CrossRef Search ADS Kelman , Steven , and John N. Friedman . 2009 . Performance improvement and performance dysfunction: An empirical examination of distortionary impacts of the emergency room wait-time target in the English National Health Service . Journal of Public Administration Research and Theory 19 : 917 – 46 . Google Scholar CrossRef Search ADS Krause , George A . 2003 . Coping with uncertainty: Analyzing risk propensities of SEC budgetary decisions, 1949–1997 . American Political Science Review 97 : 171 – 88 . Google Scholar CrossRef Search ADS Kroll , Alexander . 2013 . The other type of performance information: Nonroutine feedback, its relevance and use . Public Administration Review 73 : 265 – 76 . Google Scholar CrossRef Search ADS ———. 2015a . Exploring the link between performance information use and organizational performance: A contingency approach . Public Performance and Management Review 39 : 7 – 32 . CrossRef Search ADS ———. 2015b . Explaining the use of performance information by public managers: A planned-behavior approach . American Review of Public Administration 45 : 201 – 15 . CrossRef Search ADS Lavertu , Stéphane , David E. Lewis , and Donald P. Moynihan . 2013 . Government reform, political ideology, and administrative burden: The case of performance management in the Bush administration . Public Administration Review 73 : 845 – 57 . Google Scholar CrossRef Search ADS Lindblom , Charles E . 1959 . The science of “muddling through” . Public Administration Review 19 : 79 – 88 . Google Scholar CrossRef Search ADS MacDonald , Jason A. , and William W. Franko . 2007 . Bureaucratic capacity and bureaucratic discretion: Does congress tie policy authority to performance ? American Politics Research 35 : 790 – 807 . Google Scholar CrossRef Search ADS Maor , Moshe , Sharon Gilad , and Pazit Ben-Nun Bloom . 2013 . Organizational reputation, regulatory talk, and strategic silence . Journal of Public Administration Research and Theory 23 : 581 – 608 . Google Scholar CrossRef Search ADS March , James G . 1991 . Exploration and exploitation in organizational learning . Organization Science 2 : 71 – 87 . Google Scholar CrossRef Search ADS Meier , Kenneth J. , and Laurence J. O’Toole Jr . 2013 . Subjective organizational performance and measurement error: Common source bias and spurious relationships . Journal of Public Administration Research and Theory 23 : 429 – 56 . Google Scholar CrossRef Search ADS ———. 2008 . Management theory and Occam’s razor: How public organizations buffer the environment . Administration and Society 39 : 931 – 58 . CrossRef Search ADS Meier , Kenneth J. , Nathan Favero , and Ling Zhu . 2015 . Performance gaps and managerial decisions: A Bayesian decision theory of managerial action . Journal of Public Administration Research and Theory 25 : 1221 – 46 . Google Scholar CrossRef Search ADS Melkers , Julia , and Kathrine Willoughby . 2005 . Models of performance-measurement use in local governments: Understanding budgeting, communication, and lasting effects . Public Administration Review 65 : 180 – 90 . Google Scholar CrossRef Search ADS Migue , Jean-Luc , Gérard Belanger , and William A. Niskanen . 1974 . Towards a general theory of managerial discretion . Public Choice 17 : 27 – 51 . Google Scholar CrossRef Search ADS Mintzberg , Henry , Duru Raisinghani , and Andre Théorêt . 1976 . The structure of “unstructured” decision processes . Administrative Science Quarterly 21 : 246 – 75 . Google Scholar CrossRef Search ADS Moynihan , Donald P . 2005 . Goal-based learning and the future of performance management . Public Administration Review 65 : 203 – 16 . Google Scholar CrossRef Search ADS ———. 2008 . Dynamics of performance management: Constructing information and reform . Washington, DC: Georgetown Univ. Press . ——— . 2009 . “ Through a Glass, Darkly: Understanding the Effects of Performance Regimes .” Public Performance & Management Review 32 : 592 – 603 . Google Scholar CrossRef Search ADS Moynihan , Donald P. , Sanjay K. Pandey , and Bradley E. Wright . 2011a . Setting the table: How transformational leadership fosters performance information use . Journal of Public Administration Research and Theory 22 : 143 – 64 . Google Scholar CrossRef Search ADS Moynihan , Donald P. , Sergio Fernandez , Soonhee Kim , Kelly LeRoux , Suzanne J. Piotrowski , Bradley Wright , and Kaifeng Yang . 2011b . Performance regimes amidst governance complexity . Journal of Public Administration Research and Theory 21 ( Suppl 1 ): i141 – 55 . Google Scholar CrossRef Search ADS Moynihan , Donald P. , and Stephane Lavertu . 2012 . Does involvement in performance management routines encourage performance information use? Evaluating GPRA and PART . Public Administration Review 72 : 592 – 602 . Google Scholar CrossRef Search ADS Nielsen , Poul A . 2014a . Performance management, managerial authority, and public service performance . Journal of Public Administration Research and Theory 24 : 431 – 58 . Google Scholar CrossRef Search ADS ———. 2014b . Learning from performance feedback: Performance information, aspiration levels, and managerial priorities . Public Administration 92 : 142 – 60 . CrossRef Search ADS Nielsen , Poul A. , and Donald P. Moynihan . 2017 . How do politicians attribute bureaucratic responsibility for performance? Negativity bias and interest group advocacy . Journal of Public Administration Research and Theory . 27:269–283. O’Toole , Laurence , and Kenneth J. Meier . 2011 . Public management: Organizations, governance, and performance . Cambridge : Cambridge Univ. Press . Google Scholar CrossRef Search ADS Olsen , Asmus . 2017 . Compared to what? Experimental evidence on social and historical reference points in performance evaluation . Journal of Public Administration Research and Theory . 27:562–580. Ostrom , Elinor , Roger B. Parks , Gordon P. Whitaker , and Stephen L. Percy . 1978 . The public service production process: A framework for analyzing police services . Policy Studies Journal 7 : 381 – 9 . Google Scholar CrossRef Search ADS Poister , Theodore H . 2010 . The future of strategic planning in the public sector: Linking strategic management and performance . Public Administration Review 70 : 246 – 54 . Google Scholar CrossRef Search ADS Pollitt , Christopher . 2013 . The logics of performance management . Evaluation 19 : 346 – 63 . Google Scholar CrossRef Search ADS Pritchett , Lant , and Deon Filmer . 1999 . What education production functions really show: A positive theory of education expenditure . Economics of Education Review 18 : 223 – 39 . Google Scholar CrossRef Search ADS Rich , Robert F . 1997 . Measuring knowledge utilization: Process and outcomes . Knowledge and Policy 10 : 11 – 24 . Google Scholar CrossRef Search ADS Rockoff , Jonah E. , Douglas O. Staiger , Thomas J. Kane , and Eric S. Taylor . 2012 . Information and employee evaluation: Evidence from a randomized intervention in public schools . American Economic Review 102 : 3184 – 213 . Google Scholar CrossRef Search ADS Rutherford , Amanda , and Kenneth J. Meier . 2015 . Managerial goals in a performance-driven system: Theory and empirical tests in higher education . Public Administration 93 : 17 – 33 . Google Scholar CrossRef Search ADS Salge , Torsten O . 2011 . A behavioral model of innovative search: Evidence from public hospital services . Journal of Public Administration Research and Theory 21 : 181 – 210 . Google Scholar CrossRef Search ADS Shinkle , George A . 2012 . “ Organizational Aspirations, Reference Points, and Goals: Building on the Past and Aiming for the Future .” Journal of Management 38 : 415 – 55 . Google Scholar CrossRef Search ADS Snyder , Jeffrey W. , Andrew Saultz , and Rebecca Jacobsen . 2018 . Antipolitics and the hindrance of performance management in education . Journal of Public Administration Research and Theory . Forthcoming. Soss , Joe , Richard Fording , and Sanford F. Schram . 2011 . The organization of discipline: From performance management to perversity and punishment . Journal of Public Administration Research and Theory 21 : 203 – 32 . Google Scholar CrossRef Search ADS Spence , Michael . 2002 . Signaling in retrospect and the informational structure of markets . American Economic Review 92 : 434 – 59 . Google Scholar CrossRef Search ADS Taylor , Jeannette . 2009 . Strengthening the link between performance measurement and decision making . Public Administration 87 : 853 – 71 . Google Scholar CrossRef Search ADS Vakkuri , Jarmo . 2010 . Struggling with ambiguity: Public managers as users of NPM-oriented management instruments . Public Administration 88 : 999 – 1024 . Google Scholar CrossRef Search ADS van Helden , Jan G. , Åge Johnsen , and Jarmo Vakkuri . 2012 . The life cycle approach to performance management: Implications for public management and evaluation . Evaluation 18 : 159 – 75 . Google Scholar CrossRef Search ADS Wooldridge , Jeffrey M . 2002 . Econometric analysis of cross section and panel data . 4th edn. Mason; Cambridge, MA : MIT Press . ———. 2009 . Introductory econometrics—A modern approach . South Western College Pub . Yang , Kaifeng , and Jun Yi Hsieh . 2007 . Managerial effectiveness of government performance measurement: Testing a middle-range model . Public Administration Review 67 : 861 – 79 . Google Scholar CrossRef Search ADS © The Author(s) 2018. Published by Oxford University Press on behalf of the Public Management Research Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Journal of Public Administration Research and TheoryOxford University Press

Published: May 7, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off