by the Ministry of Local Government and Regional Development.

Transcription

1 EFFICIENCY IN PUBLIC SERVICE PRODUCTION: THE IMPACT OF POLITICAL AND BUDGETARY INSTITUTIONS * Lars-Erik Borge **, Torberg Falch and Per Tovmo Department of Economics Norwegian University of Science and Technology Dragvoll University Campus N-7491 Trondheim, Norway Abstract The purpose of this paper is to investigate whether political and budgetary institutions have any impact on the efficiency in the production of public sector services. In order to address this issue we take advantage of a new measure of total output in Norwegian local governments based on indicators of production for each service sector. Efficiency is measured as the ratio between total output and available resources. We find that political fragmentation, socialist influence and a high level of revenue contributes to low efficiency. Budgetary institutions do not seem to have any significant impact on efficiency. * The paper is part of a project on efficiency in the Norwegian local public sector financed by the Ministry of Local Government and Regional Development. ** Corresponding author: larseb@svt.ntnu.no, phone , fax Date: March 2004

2 1. Introduction Efficiency in public service provision has been a heavily debated issue in the Western countries for many years. The emphasis on efficiency is related to the fact that many countries face a demographic development that increases the demand for public services, and at the same time experience substantial fiscal deficits and public debt. Improved efficiency in public service production facilitates more services without increasing government expenditures. In this paper we pay attention to the impact of political and budgetary institutions on efficiency in the production of public services. The role of political and budgetary institutions have received large attention in the economics literature since the seminal papers by Roubini and Sachs (1989) and von Hagen (1992). Roubini and Sachs showed that weak governments lead to high budget deficits in OECD countries, whereas von Hagen found that hierarchical budget procedures were associated with low spending and budget deficits in EU-countries. The following empirical literature has to a large extent focused on deficits, debt, taxes and public spending (see e.g Kontopoulus and Perotti 1999, Volkerink and De Haan 2001). A parallel theoretical literature has focused on the common pool problem in the financing of the public sector, which leads to excessive spending and taxation in a static context and to excessive spending and deficits in a dynamic context (Persson and Svensson 1989, Alesina and Tabellini 1990, Velasco 1999). In the present paper we do not ask whether provision of public services is excessive, but whether output is maximized given available resources. Several studies have calculated the relative efficiency of public sector units based on the Data Envelope Analysis (DEA), but this is merely descriptive analyses. A few studies (Vanden Eeckhaut et al., 1993, De Borger et al., 1994, Kalseth and Rattsø, 1998, Athanassopoulos and Triantis, 1998 and Hayes and Chang, 1990) have studied the impact of political institutions on productive efficiency. These studies share the common feature that they undertake a two step analysis. In the first step, efficiency is calculated based on DEA (De Borger et al., 1994, 1

3 use the less restrictive Free Disposal Hull method), and in the second step the revealed efficiency is used as a dependent variable. One problem with the DEA method, which in general allows for several output and input measures, is that typically only a restrictive number of input and output measures can be included to avoid that almost all units are classified as more effective that comparative units. In two-step analysis, one must pay particular attention to this problem when interpreting the results from the second step. Like the existing studies of the impact of political institutions on productive efficiency we use data from the local government level, but we utilize an alternative measure of efficiency. We take advantage of a new measure of total output in Norwegian local governments. The output measure we apply is very comprehensive as it covers most service sectors and include 19 disaggregated output measures. Since 2001 this indicator of service production has annually been reported publicly. Due to the restrictions on the number of outputs that can be analyzed simultaneously in the traditional methods, the literature is dominated by studies on the efficiency in providing specific local public services. However, there are a few exceptions and De Borger and Kerstens (2000) present an overview of studies on overall efficiency. These studies follow different methodological approaches but they all involve output along several dimensions. Our focus is also on overall efficiency. With our approach this is possible without being forced to leave out detailed information on the production in each sector. While sectoral analyses have their merit in the understanding of specific service provisions, the most relevant decision-making to influence efficiency in most countries are taken at the local government level. To evaluate the performance of different political structures, one should be concerned with efficiency aggregated across several service sectors. The focus on the aggregated level implies that the input side can be summarized by available resources in the local government, and that the ratio between total output and revenues (deflated by an index capturing cost conditions) can be applied as a first rough indicator of efficiency. In the econometric analysis we control for some weaknesses of the output and revenue measures in a first step analysis. Using the revealed efficiency measure, we find that political fragmentation, socialist influence and a high level of 2

4 revenue contributes to low efficiency, while budgetary institutions do not seem to have any significant impact on efficiency. The rest of the paper is organized as follows: Section 2 presents the measure of total output and productive efficiency. In Section 3 we develop hypotheses regarding the impact of political and budgetary institutions, and discuss the variables used in the econometric analysis. Section 4 presents the basic estimation results, while Section 5 provides some robustness checks and includes some additional explanatory variables in the model. Finally, Section 5 offers some concluding remarks. 2. Measuring output and efficiency Ideally, variation in a measure of production across local governments should reflect differences in either economic conditions or production efficiency. Measuring production within the public sector is not an easy task, however, particularly because output is multidimensional. Both quantity of production and the quality of the service are important aspects of output, and quantity and quality may be multidimensional themselves. In this paper we take advantage of earlier research conducted for the Advisory Commission on Local Public Finance (TBU). In Borge et al. (2001) we developed an aggregate indicator of local government service production. The idea was to establish an aggregate indicator of service production based on indicators of production for each service sector, and utilizing a new database for local government service production and expenditures (KOSTRA). Borge et al. (2001) discuss features of an ideal production index that only captures differences in economic conditions and production efficiency. Since 2001 this indicator of service provision has been reported annually in reports from the Advisory Commission on Local Public Finance. 1 1 In this Section we present the data for Almost identical indicators are used for 2001, and in the econometric analysis below we also use data for

5 The aggregate output measure is based on disaggregated output measures for the following six service sectors; day care, compulsory education (1st to 10th grade), child welfare, primary health care, care for the elderly and social assistance. Care for the elderly and compulsory education are the largest sectors within Norwegian local governments. Total expenditures in the sectors included in the production index account for more than 80 percent of the total expenditures on average. 2 The main idea behind the choice of indicators is that they should measure output, not input. The indicators for all services in 2002 are reported in Appendix 1. As can be seen there are some exceptions from this rule. The clearest case is primary health care where man years for different professional groups are used as indicators. In compulsory education teaching hours can be regarded as a border case. As a robustness check we include the exam results for the 10th graders in some of the analysis below. Care for the elderly is the sector that is most broadly covered by several equally weighted indicators. For this sector, there exist some earlier analysis of efficiency variation (Kalseth, 2003, Førsun??), and the indicators used here are similar to the ones used earlier. Most of the indicators used in establishing an index for total output are related to the quantity of a particular service. This reflects that quantity is easier and more frequently measured than quality measures, and similar biases are common in all studies of production efficiency in the public sector. Out of the 19 production indicators used to calculate total output, only four of them (such as the share of single rooms in institutions for care for the elderly and play and outdoor area per child in day care) are quality measures. The weighting of the indicators is based on their expenditure shares within the different service sectors. In cases where it is impossible to divide expenditures among two or more indicators, the indicators under consideration are given equal weight. The weight for each 2 The sectors not included due to lack of output measures are culture, housing, communication and infrastructure, in addition to central administration. 4

6 indicator in the sector indexes and the weight for the sector indexes in the aggregate production index are reported in Appendix 1. Table 1 reports production indexes where the local governments are grouped by economic conditions. Economic conditions are represented by local government revenue, where revenue includes local taxes and block grants per capita, and is deflated by an index that captures varying cost conditions across local governments. The point of departure for the deflating is the cost index used in the needs equalization system, which is based on variables such as population size, settlement pattern, the age composition of the population, and social factors. This revenue measure is widely accepted as the most reliable indicator of differences in economic conditions across local governments. Here we also take account of the regional variation in the payroll tax. In Table 1 both the production indexes and revenue are normalized such that the weighted averages (with population as weight) are equal to 100. Table 1: Production indexes, local governments grouped by revenue level Revenue level Day Educati Child Primary Elderly Social Aggr. care on welfare health care assist. Index Below Above No. of local governm. Standard deviation Correlation with revenue Weight in aggregated production index Notes: The figures are based on data for 379 out of 435 local governments. 5

7 It appears that the level of revenue correlates positively with total output, as well as with output in each service sector. In other words, local governments with good economic conditions tend to have good services in all sectors. The correlation is strongest for day care, primary health care and compulsory education. It is an interesting observation that the correlation with total output is higher than any correlation for the individual service sectors. This finding probably reflects that differences in priorities across sectors to a great extent are aggregated out in the indicator of total output. We also observe that the standard deviation of total output is lower than the standard deviation for each sector, with one exception. In our context production efficiency can be interpreted as getting maximum production for a given level of revenue. The ratio between total output and revenue is therefore a natural indicator of efficiency. The main difference between our approach and methods like DEA is that we compare efficiency across local governments without specifying an efficiency benchmark. This indicator of efficiency shows large variation across local governments. When the weighted average (with population as weight) is normalized to 100, it varies from 53.8 to 134.3, and has a standard deviation of This indicator (denoted E) is the starting point for the econometric analysis in Section 4 where we analyze the relationship between efficiency and political and budgetary institutions. We also apply two alternative efficiency indicators that are adjusted for possible shortcomings by our output and revenue measures in order to increase the robustness of the econometric results. The alternative indicators are constructed on the basis of the following regression E = α + α Z + u, (1) i 0 1 i i where Zi is a vector of control variables in local government i that capture possible shortcomings in our output and revenue measures. Including a variable in Z means assuming that this variable affect revenues or production but not efficiency. The elements 6

8 of the Z vector are discussed below. The residuals following from the estimation of (4.2) A can be interpreted as adjusted efficiency, denoted E = u. i i In the first model we take account of the following three shortcomings in the output and revenue measures. First, a few minor services (culture, communication, infrastructure and housing) are not included in total output due to lack of indicators of production. As a consequence, there is a tendency to underestimate output, and thereby efficiency, in local governments that have given high priority to these services. To account for this the expenditure shares for the four services that are left out are included in Z. Second, since user charges are not included in the revenue measure, we will tend to overestimate efficiency in local governments with high user charges. This is accounted for by including user charges for day care and care for the elderly as elements in Z. Third, net operating surplus and expenditures on interests and installment vary across local governments and affect service provision. All other equal, there will be a tendency to overestimate efficiency in local governments with deficits and low expenditures on interests and installment. This is accounted for by including net operating surplus and interests and installments (as fractions of revenue) in Z. The first alternative indicator of efficiency takes account of the factors discussed above, and is denoted A1 E. In the A2 second alternative, which is denoted E, we also take account of some structural characteristics by including the variables in the cost index in the need equalization system. Among other variables, these include the population size and settlement pattern. This means that we disregard scale efficiency in the current study. The three alternative indicators of efficiency are clearly positively correlated Theoretical considerations Conflict of interest between different actors will in general give inefficient outcomes. This is a common feature within public sector production, and utilization of the principal- 3 The correlations are 0.84, 0.74 and 0.88 between E and E A1, E and E A2, and E A1 and E A2, respectively. 7

9 agent approach is heavily used to describe optimal incentive schemes for public sector agencies (see for example Dixit, 2002). The central theorem in public choice is that individuals always pursue self-interest. The modeling of the behavior of public sector agencies taking bureaucrats self-interest into account was initiated by Niskanen (1971) who assumed that bureaucrats prefer high budgets, implicitly assuming that things from which they derive satisfaction all vary directly with the budget size. Later contribution (Niskanen 1975, Moene 1986 and Chan and Mestelman 1988) has used the concept of discretionary profit, and a more flexible objective function that includes both the level of production and discretionary profit. Discretionary profit is interpreted as all kind of nonproductive activities, including low effort, over-employment and extra salary. Thus, discretionary profit is conceptionally equal to inefficiency. To what extent are civil servants preferences for low efficiency fulfilled? This is a question of to what extent voters and political leaderships are capable to hinder agencies in extracting discretionary profit. In basic principal-agent models, the principal has all the bargaining power such that efficiency can be achieved in simple settings by correct incentives. In reality the public sector settings are complex with multiple principals, multiple tasks and lack of competition. Dixit (2002) shows that such complexities weaken the overall incentives to the agencies in optimum. In addition, the assumption that agencies have no bargaining power, underlying all principal-agent models, is likely to be violated in a public sector setting. At the extreme, it is a question on whether the principal or the agency is the Stackelberg leader of the game. One popular specification of the influence of bureaucrats is public sector trade union power. Powerful public sector unions will be able to raise the costs by increasing wages, which at least within a given budget will reduce production and efficiency, see for example a theoretical discussion by Falch (2001), an empirical approach in Falch and Rattsø (1997), and the overview of Gregory and Borland (1999). Kalseth and Rattsø (1998) formulate the interaction between politicians and the bureaucracy as a bargaining game. The political leadership enters the bargain with a community preference function while the bureaucrats prefer overspending. 8

10 For empirical analyses the main question is how to measure the ability of the bureaucrats to extract discretionary profit. In motivating the variables included in the empirical model below, we will rely on a setting where the public sector agencies (bureaus) have some power in the interaction with the principal (political leadership). We follow the large literature that relates the bargaining power of the political entity to the political structure. Political strength is likely to be important to hold down the agency demand for discretionary profit, making strong political leaderships more able to pursue its optimal policy than weak political leaderships. 4 Regarding budgetary procedures, a centralized budgetary procedure may hinder the different sectors to be Stackelberg leaders in a game with the politicians, or more generally reduce the power of the agencies in the interaction with the political leadership, working in the direction of more efficient production. In addition, in this setting the service producing agency will be able to take advantage of a rich sponsor by increasing discretionary profit. We therefore expect efficiency to be a decreasing function of local government revenue. Fragmentation is used as a measure of political strength in a large number of analyses, see for example Roubini and Sachs (1989) and Kontopoulos and Perotti (1999). Several analyses on Norwegian local governments have found strong and consistent effect of a Herfindahl index of party fragmentation (see for example Borge, 1995, Falch and Rattsø, 1997, Kalseth and Rattsø, 1998, Tovmo and Falch, 2002). The index is calculated as HERF = 100* S, P p= 1 2 p where S p is the share of the seats in the local council held by party p and P is the number of political parties. The index is inversely related to party fragmentation as it is equal to 4 A related literature, initiated by Roubini and Sachs (1989), links cost control measured by the size of budget deficits to political structure. Within this literature, deficits are mainly viewed as a result of a dynamic common pool problem, and political strength decreases deficits because strength helps to overcome the common pool problem. 9

11 the minimum value 100/P when the seats are equally divided among the parties and equal to 100 if only one party is represented in the local council. In the Norwegian context party fragmentation is negatively correlated with the share of socialist seats in the local council. The reason is that the labor party dominates the socialist camp, while the non-socialist camp is more fragmented. Moreover, a few recent studies indicate that socialist influence may affect costs and efficiency in the local public sector. Kalseth and Rattsø (1998) find that socialist influence increases the administrative share of the budget, and the results in Strøm (1995) indicate that socialist influence increases the wage level. These results can be understood as results of orientation towards planning and administration, and close interaction with trade unions. Both these mechanisms work in the direction of a negative effect on our efficiency measure of the percentage of socialists in the local council (SOC). We will distinguish between centralized and decentralized budgetary procedures. In the Norwegian case, a decentralized budgetary procedure starts with each sector of the local government working out a budget proposal for their sector. Afterwards the head of the administration or the executive board coordinate the different proposals in order to prepare a proposal to be approved by the local council. In contrast, under a centralized budgetary procedure the head of the administration or the executive board presents an overall budget proposal to the different sectors, including the budget size for each sector, and the sectors only work out the details within their specific sector. The evidence clearly suggests that centralized budgetary rules enhance lower deficits, see for example von Hagen and Harden (1995), Poterba (1996) and Tovmo (2003). Centralization may be seen as one way to reduce the bargaining power of the agencies, and in this way helping overcome some common pool problems. Regarding efficiency, there may also be other effects. One hypothesis is that a centralized budgetary procedure harms efficient production. Poterba (1996) speculate that the cost control may be achieved at the cost of reducing the amount of information that can be brought to bear on particular budget decisions (p. 48/49). The tradeoff between cost control and 10

12 efficiency has a clear parallel in fiscal federalism. Fiscal federalism is usually thought of as a tradeoff between taking externalities into account on the one hand and local discretion or efficiency by information advantages on the other hand, see for example Seabright (1996). All together, the effect of on efficiency of a centralized budgetary procedure (Cent) in contrast to a decentralized procedure is ambiguous. 4. Estimation results The hypotheses outlined above are tested empirically using data from Norwegian local governments for 2001 and Out of a total number of 435 local governments, data on total output and efficiency are available for 359 in 2001 and 379 in We estimate simple linear models with either the benchmark efficiency indicator E as the dependent variable or the adjusted efficiency measures E A1 and E A2 as dependent variable. E = β + β X + u, (2) i 0 1 i i where Xi is a vector of variables including local government revenue per capita, the Herfindahl-index of party fragmentation, the share of socialist seats in the local council, and a dummy variable for centralized budgetary process. The model is estimated with OLS and the results are presented in Tables 2-4. Instead of pooling the data we estimate separate models for each year. Moreover, we estimate equations with and without the budgetary variable since it is not available for all local governments. Estimation results using the benchmark indicator of efficiency are shown in Table 2. The first column reports the results using data for 2002 and with local government revenue and the two political variables as regressors. All three variables come out highly significant, and it appears that political fragmentation, socialist influence and a high level of revenue contributes to low efficiency. The results implies that an increase in the Herfindahl-index by 10 percentage points (the sample range of the Herfindahl-index is 11

13 slightly above 45 percentage points) increases efficiency by 4.1 percentage points. This negative impact of political fragmentation is consistent with the findings in several analyses of overall efficiency in Belgian municipalities summarized by De Borger and Kerstens (2000), which indicate that the number of coalition partners has a negative effect on efficiency. The finding that party fragmentation contributes to low efficiency is also consistent with Kalseth (2003), who analyzes the efficiency in long-term care sector in Norwegian local governments using DEA in the first step of the analysis. Among several variables capturing political strength, she find a negative effect of the effective number of parties, measured as the inverse of the Herfindahl-index used in the present analysis. Table 2: The determinants of efficiency, the benchmark indicator (E) 2002 data 2001 data Constant (33.84) (27.44) (42.85) (36.71) Revenues (7.80) (6.79) (10.87) (10.12) SOC (3.45) (3.45) (1.60) (1.73) Herf (4.97) (4.22) (3.75) (3.32) Cent (1.69) (1.18) # observations R adj Note: Absolute t-values based on robust standard errors in parentheses. The coefficient of the share of socialist seats in the local council indicates that an increase in socialist representation by 10 percentage points reduces efficiency by around 1.5 percentage points. This is in line with Kalseth (2003) who also found that socialist influence has a systematic negative impact on efficiency. The evidence on socialist influence on efficiency is not unambiguous as de Borger and Kerstens (2000?) found that socialist influence increases overall efficiency in Belgian municipalities. 12

14 Regarding our measure of economic conditions, there is a clear negative effect. De Borger and Kerstens (2000) have summarized the results from studies based on data from different countries and a negative connection between revenues and efficiency seems robust across countries. When interpreting our results, there is, however, at least one important caveat. As the economic conditions improves, the local governments may change focus from the quantity of production to quality aspects. As quality is poorer measured than quantity in our production index, this may explain the negative effect of economic conditions. In the second column of Table 2 we also include the dummy variable for centralized budgetary procedure. The variable comes out with a negative sign and is borderline significant. A switch from a decentralized to a centralized budgetary procedure is predicted to reduce efficiency by 2.4 %-points. The magnitude and significance of the other variables are not affected by the inclusion of the budgetary dummy, even though the sample is reduced. The two last columns of table 2 report corresponding results using data for It appears that the impact of all variables except revenues are smaller than in 2002, and at the 10 percent level, the share of socialists in the local government is only borderline significant while the dummy variable for budgetary institutions is insignificant. In Table 3 we report estimation results where the two alternative measures of efficiency described in section 2 are used as dependent variables. The underlying first step regressions are reported in Table A2 in Appendix 2. The two first columns shows the results for alternative A1 E, where the structural characteristics are not included in the first step, whereas the third and fourth column shows the results for alternative where structural characteristics are included in the first step. A2 E, It appears that the impacts of revenue and ideology are not much affected by the two-step procedure taking into account possible shortcoming by the output and revenue measures. Revenues come out as highly significant, and it is still the case that a high level of 13

15 revenue contributes to low efficiency. The impact of fragmentation is still significant although the quantitative impact is somewhat reduced compared to the corresponding results in Table 2, especially when the dependent variable is A2 E. Table 3: The determinants of efficiency, the alternative indicators (E A1 and E A2 ) Indicator E A Constant (5.28) (5.74) Revenues (6.00) (7.94) SOC (2.56) (1.35) Herf (4.33) (4.55) Indicator E A (7.60) (6.82) (7.13) (8.60) (2.07) (0.98) (2.10) (3.12) # observations R adj Note: Absolute t-values based on robust standard errors in parentheses. Table 4 is similar to Table 3, with the only exception that the budgetary dummy and the variable capturing internal competition are included as additional determinants of efficiency in the second step. Again, we reach similar results as with the benchmark indicator of efficiency for the budgetary procedure. Although adding more control variables when adjusting our measure of efficiency increases the quantitative impact, but the dummy for a centralized procedure is still insignificant. A1 Table 4: The determinants of efficiency, the alternative indicators ( E and Indicator E A1 Indicator E A Constant (5.72) (5.59) (7.80) (7.74) Revenues (6.17) (7.76) (6.86) (8.64) SOC (3.04) (1.66) (2.68) (1.30) Herf (3.73) (4.26) (1.61) (2.97) Cent A2 E ) 14

16 (1.30) (1.00) (1.71) (1.53) # observations R adj Note: Absolute t-values based on robust standard errors in parentheses. 5. Some extensions In this section we firstly look at some extensions of the efficiency measure used by including a new control variable in the first step analysis and by extending the indicator set of compulsory education. Second, we introduce some new explanatory variables that may affect the efficiency monitoring of the local councils and the voters. The evidence in Kalseth and Rattsø (1998) indicate that political strength (measured by for example party fragmentation) has a negative effect on the size of administration in Norwegian local governments, while the share of socialist representatives in the local government has a positive effect. This is in line with the results in the present paper because the lower efficiency may go through a larger administration. To investigate whether the effect via administration explains the present results, we included the budget share of administration in the first step analysis. In this way, the adjusted efficiency E A must be interpreted as efficiency for given size of the administration, variation in administration does not influence the efficiency measure. Using the new efficiency measures (E A1 and E A2 ) give almost identical results to those reported in Table 3. If anything, the effects of the political and budgetary variables become even stronger in this case. This clearly indicates that efficiency variation in the different sectors may be related to political factors. The output measure used in the present study is solely based on the new database for local government service production and expenditures (KOSTRA). In this database few output measures in compulsory education is available, which may be a serious limitation because compulsory education has a weight of 0.35 in the total index. Recently, exam 15

17 results from the last year of compulsory education (10 th grade) have been released, a variable which is clearly separated from the input in the sector. We change the production index (and thereby the benchmark efficiency index E) in two ways. First we give the average exam results and teaching hours per student equal weights (denote E E1 ), and secondly we simply replace teaching hours per student by the average exam results (denoted E E2 ). 5 Table 5 reports the results using the alternative efficiency measures. The effect of revenues becomes stronger. As the revenues of a local government increases, purchased inputs in schooling increases (the demand elasticity of school expenditures is positive). This increases teaching hours per student, but have a small and perhaps no effect on exam results. On the other hand, the impact of socialist influence and party fragmentation becomes smaller. This indicates that political factors have a small effect on student achievement. Table 5: The determinants of efficiency, alternative efficiency measures E E1 Constant (41.94) (35.16) Revenues (12.31) (10.63) SOC (2.57) (2.83) Herf (4.31) (3.88) Cent (1.89) E E (51.92) (43.23) (17.89) (15.06) (1.12) (1.63) (2.94) (2.86) (2.05) # observations R adj Note: Absolute t-values based on robust standard errors in parentheses. 5 The correlation coefficient between teaching education hours and average exam results are -0.02, and the correlation coefficients between E and E E1, E and E 2, and E E1 and E E2 are 0.095, 0.79, and 0.94, respectively. 16

18 The analysis so far has only included political and budgetary variables in addition to economic conditions. Understanding the decision-making of local governments within a principal-agent framework, politicians may have additional channels to influence the outcome. For example, contracting out some of the service production may signal to the service providers (agents) that inefficiency may change organization of the service production. In table 6, we extend the model by including a variable that captures internal competition. The variable (Priv) measure to what extent the municipalities have contracted out garbage collection. 6 Evidence on the influence of competition is mixed, see Duncombe et al. (1997), but we find a positive and significant impact, while the effects of the other variables in the model are not altered. The result may reflect a threat -effect, but it may also capture that preferences for efficiency varies across local governments. There may also be an endogeneity bias because local governments with low efficiency may be more likely to use contracting out. Such an negative relationship between efficiency and contracting yields a negative bias on the effect reported. Table 6: The determinants of efficiency, the benchmark indicator (E) 2002 data 2001 data Constant (35.71) (29.06) (45.78) (38.45) Revenues (8.21) (7.06) (12.15) (11.15) SOC (3.22) (3.23) (1.44) (1.57) Herf (4.78) (4.19) (3.84) (3.35) Cent (1.63) (0.95) Priv (2.24) (2.13) (2.46) (2.17) # observations R adj Note: Absolute t-values based on robust standard errors in parentheses. 6 The variable Priv is the amount of purchased services (in 100-NOK) in garbage collection per household. 17

19 Several studies find that how local public services are financed affect efficiency, see for instance de Borger et al. (1994) and Athanassopoulos and Triantis (1998). The findings reflect that effort in monitoring production increases with the tax price. To investigate whether the same mechanisms are present in Norwegian local governments we included a dummy variable that equals one if a local government has property taxation. We found no such effects in the analysis. We also tried a variable measuring fiscal competition between local governments influence efficiency, utilizing commuting data on workers to proxy the moving costs of the citizens. However, we did not find any systematic effect of this variable either. 5. Concluding remarks The purpose of this paper was to investigate whether political and budgetary institutions have any impact on production efficiency in Norwegian local governments. In order to address this issue we took advantage of a new indicator of total output based on indicators of production for each service sector, and efficiency was measured as the ratio between total output and revenues (available resources). We are able to document that political fragmentation, socialist influence and a high level of revenue contribute to low efficiency. Budgetary institutions do not seem to have any significant impact on efficiency. 18

20 References Athanassopoulos, A., and K. Triantis (1998): Assessing aggregated cost efficiency and the related policy implications for Greek local municipalities. Infor 36, Alesina, A. and G. Tabellini (1990): A positive theory of fiscal deficits and government debt. Review of Economic Studies 57, Borge, L.-E., T. Falch and P. Tovmo (2001): Produksjonsindeks for kommunale tjenester (Index of production for local government services), Research Report, ALLFORSK, Trondheim. Borge, L.-E. (1995): Economic and political determinants of fee income in Norwegian local governments, Public Choice 83, Chan, K. S. and S. Mestelman (1988). Institutions, efficiency and the strategic behavior of sponsors and bureaus. Journal of Public Economics 37, De Borger, B., and K. Kerstens (2000): What is known about municipal efficiency? The Belgian case and beyond. In J. L. T. Blank (ed.), public provision and performance, North-Holland. De Borger, B., K. Kerstens, W. Moesen and J. Vanneste (1994): Explaining Differences in Productive Efficiency: An Application to Belgian Municipalities. Public Choice 80, Dixit, A. (2002): Incentives and organizations in the public sector. An interpretative review. Journal of Human Resources 37, Duncombe, W. D., J. Miner and J. Ruggiero (1997): Empirical evaluation of bureaucratic models of inefficiency. Public Choice 93, Falch, T. (2001). Collective Bargaining in the Public Sector and the Role of Budget Determination. European Journal of Political Economy 17, Falch, T. and J. Rattsø (1997). Political economic determinants of school spending in federal states: Theory and time series evidence. European Journal of Political Economy 13, Gregory, R. G. and J. Borland (1999): Recent developments in public sector labor markets, in O. Ashenfelter and D. Card (eds.): Handbook of Labor Economics, Volume 3, Elsevier Science B. V., Amsterdam. Hayes, K., and S. Chang ()1990): The relative efficiency of city manager and mayorcouncil forms of government. Southern Economic Journal 57, Kalseth, J. (2003): Political Determinants of Efficiency Variation in Municipal Service Production: An Analysis of Long-term Care in Norway. Mimeo, PhD Dissertation, Department of Economics, NTNU. Kalseth, J. and J. Rattsø (1998): Political control of administrative spending: The case of local governments in Norway. Economics and Politics 10, Kontopoulos, Y. and R. Perotti Government Fragmentation and Fiscal Policy Outcomes: Evidence from OECD Countries. In J. M. Poterba and J. von Hagen (eds.), Fiscal Institutions and Fiscal Performance. The University of Chicago Press, Chicago. Moene, K. O. (1986): Types of bureaucratic interaction. Journal of Public Economics 29, Niskanen, W. A. (1971): Bureaucracy and representative government. Aldine Atherton, Chicago. 19

21 Niskanen, W. A. (1975): Bureaucrats and politicians. Journal of Law and Economics, Persson, T. and L. Svensson (1989): Why a stubborn conservative would run a budget deficit: Policy with time inconsistent preferences, Quarterly Journal of Economics 104, Poterba, J. M. (1996): Do budget rules work? NBER Working paper Roubini, N. and J. Sachs (1989): Government Spending and Budget Deficits in the Industrial Countries. Economic and Policy 8, Seabright, P. (1996): Accountability and decentralisation in government: An incomplete contracts model. European Economic Review 40, Strøm, B. (1995): Envy, Fairness and Political Influence in Local Government Wage Determination: Evidence from Norway. Economica 62, Tovmo, P. (2003). Budgetary procedures and deficits in Norwegian local governments, Mimeo, Department of Economics, Norwegian University of Science and Technology. Tovmo, P. and T. Falch (2002): The flypaper effect and political strength. Economics of Governance 3, Vanden Eeckhaut, P., H. Tulkens and M-A. Jamar (1993): Cost Efficiency in Belgian Municipalities. In : Fried, H. O., C. A. K. Lovell and S. S. Schmidt, eds. The Measurement of Productive Efficiency: Techniques and Applications, Oxford University press. Velasco, A. (1999): A model of endogenous fiscal deficits and delayed fiscal reforms. In: J.M Poterba and J. von Hagen, eds. Fiscal Constitutions and Fiscal Performance, Chicago University Press. Volkerink, B. and J. De Haan (2001): Fragmented government effects on fiscal policy: New Evidence. Public Choice 109, Von Hagen, J. (1992): Budgeting procedures and fiscal performance in the European Communities. Economic Paper 96, Commission of the European Communities. Von Hagen, J., and I. Harden (1995): Budget processes and commitment to fiscal discipline. European Economic Review 39,

22 Appendix 1. Documentation of the sector indexes and the measure of total output Table A1: Indicators and weights for the production indexes Sector/ Indicator Weight Day care Adjusted hours of attendance in local public day care inst. per child 0-5 years Play and outdoor area per child Share of children 0-5 years in private day care inst. receiving financial support from the local government The amount of support per child to private day care inst Primary education Teaching hours per student Share of children 6-9 years in day-care facilities for school children Share of users with attendance above 15 hours a week Primary health care Man-hours by doctors per inhabitants Man-hours physiotherapists per inhabitants Man-hours by nurses per10000 inhabitants 0-6 years Care for the elderly Share of inhabitants above 80 years of age receiving home-based services Share of inhabitants above 80 years of age living in institutions Share of single rooms in institutions Share of recipients of home-based services receiving additional services Share of inhabitants above 80 years of age in dwellings for the aged Child welfare Number of investigations as share of inhabitants aged Number of children receiving help as share of inhabitants aged Social assistance Share of inhabitants aged receiving economic assistance Average amount of support per month Index of total output Kindergartens Primary education Primary health care Care for the elderly Child welfare Social assistance

23 Appendix 2. The first step regressions Table A2: First step regression with the benchmark efficiency indicator as dependent variable 2002 data 2001 data Constant (22.81) (13.44) (23.92) (14.17) Budget share culture (4.53) (4.51) (4.94) (4.38) Budget share communication (0.19) (1.46) (2.54) (3.50) Budget share housing (1.26) (1.38) (3.00) (3.07) Budget share infrastructure (1.82) (0.06) (0.14) (1.43) Net operating surplus (3.63) (3.76) (2.87) (2.94) Interests and installment (2.96) (3.12) (1.74) User charges, day care (6.45) (5.44) (0.34) User charges, care for elderly (5.58) (3.99) User charges, institutions (1.24) User charges, home-based care (1.22) (1.47) (2.15) (0.95) (0.55) Structural characteristics No Yes No Yes included # observation Std. dev. of residuals R adj Notes: Absolute t-values based on robust standard errors in parentheses. 22