(BEING CONTINUED FROM 10/01/18)
In consequence, in both cases of adaptive behavior of the bureaucrat, budget cuts will lead to a decline of the quality of the produced public good. As regards the research question of this paper, the quality of public schooling might be assessed using objective measures of student achievement. For the U.S., during the 1990s various empirical multivariate analyses of the impact of newly introduced tax limits on student performance were carried out.12 Most of these studies are based on a simple educational production function augmented with a tax limit dummy that replaces traditional school resources and input variables such as e.g. class size or teacher quality (e.g. Figlio, 1997). For example, in a cross-state analysis, Figlio (1997) reports a substantially lowering influence of tax limits on student performance in science, social studies, and reading examinations. Moreover, Downes and Figlio (1997) find a tax limit associated with a sizeable and significant decline in statewide mean student performance in mathematics (cf. also Downes, Dye, & McGuire, 1998).13
However, to my knowledge, non of the ‘tax limit literature’ tests the direct relation between educational spending and student performance, although all of them conjecture the school budget to work as tax limit’s transmission channel.14 Other educational economists, however, report ambiguous results with respect to the influence of spending or school resources on students’ test scores (for reviews, see Hanushek 1986, 1996, 2002). Based on this evidence, I formulate the following hypotheses which I intend to test with Swiss data: Hypothesis 1: Direct Democracy exerts a spending restraint impact on educational spending. 10 Hypothesis 2: Direct democracy increases efficiency in the provision of schooling. Thus, through smaller school budgets direct democracy has no detrimental impact on student achievement. Hypothesis 3: Direct democracy does not lead to efficiency gains. In consequence, it leads to worse student performance caused by the induced school budget cuts. Hypothesis 4: Direct democracy provokes counteraction of Leviathan-like school administrators. Consequently, not only through smaller school budgets, but even more through activities of the bureaucrats going beyond the financial realm student performance is worsened. 3 Data and Model 3.1 The Data Sociodemographic, economic and fiscal determinants at the cantonal level and a cultural (language) factor are obtained from the Swiss Federal Statistical Office; some aggregate indicators such as ideology of cantonal government, tax competition and fiscal constraints are based on my own calculations or that of my colleagues Lars P. Feld, G. Kirchgässner and Ch. Schaltegger. Cantonal culture is measured by a dichotomous variable for French- and Italianspeaking cantons. The expenditure variables are for combined local and cantonal educational 11 per pupil spending in primary and secondary I education.15 All monetary variables are deflated to the base year 1980.
For estimating an educational production function composed of school characteristics, classroom-related characteristics, peer characteristics, and, student background information, the so-called national study accompanying the PISA 2000 survey is employed.16 This national studies used identical questionnaires for the same test subjects, statistical methods of sample selection and test score construction as the OECD PISA 2000 survey. Due to its focus, more students were assessed in reading than in the two other subjects. However, in contrast to the OECD-PISA study, which sampled the 15-year old irrespective of their educational stage, the population of the national study includes only ninth graders. For this reason, the matching of schools and students makes it possible to construct classroom-based peer variables. As in the OECD PISA data, test scores are obtained as weighted likelihood estimates. More specifically, they are constructed as a weighted average of correct responses, with the weights reflecting the level of difficulty of the question (Hambleton & Swaminathan, 1985; Warm, 1989). In consequence, a student who answers one more challenging question correctly might perform as equally well as someone who gave correct responses to a small number of more simple questions. Students attending classes of less than 20 are excluded from the analysis to prevent endogeneity in the peer group variables because in smaller classes the respondent’s performance is more likely to have an impact on the average achievement of her peers than in larger classes.17 The mean of the reading test score was originally normalized at 500, with a standard deviation of 90 for the whole national dataset, but because of the deletion process, the mean of the sample I use is about 530, with a standard deviation of approximately 80 based on a final sample of about 3,530 observations12. For descriptive statistics of the core subject test scores, see Table 1.
——————————————————— Insert Table 1 about here ——————————
Finally, as explanatory variable of interest a measure of direct democracy is employed which was constructed for the year 2000 based on the methodology described in Stutzer (1999). It is an unweighted average of four subindices that evaluate the power of the constitutional initiative, the statutory initiative, the fiscal referendum, and the statutory referendum. The strength of these popular rights are assessed based on the stipulations in the cantonal constitutions with respect to the number of signatures to be collected, the number of days available for their collection, and the financial threshold of the expenditure project, if applicable. The four subindices are constructed based on the awarded points for each single requirement with a higher number reflecting stronger popular rights. As most of these subindices of direct democracy are highly correlated with up to ρ= 0.8, using the overall index in place of the single subindices is highly recommended. The overall index of direct democracy takes on values between 1 and 6, with 6 indicating the highest degree of empowerment of the cantonal electorate. In our data, the lowest value (1.75) is observed for the canton ‘Geneva’ while the highest is achieved by ‘Glarus’ (5.75). For the year 2000, the values for all 26 cantons are displayed in Table A1 of the Appendix. Nevertheless, this index measures the presence of these institutions rather than their effective use. Feld & Kirchgässner (2001) demonstrate that the mere existence of such an institution is already sufficient to induce a change in policy outcomes because it serves as a credible threat by the citizenry in a game theoretical context. According to this model, popular rights are only actively exerted in case of strong deviations of the politicians’ decisions from the median voter’s preferred policy. In consequence, employing a measure of effective exertion of direct legislation would understate the true effect of this institution. In addition, it 13 should be noted that this approach follows most of the public finance and public choice literature. Definitions and descriptive statistics of all dependent and explanatory variables are provided in Tables A2 and A3 of the Appendix.
3.2 The Model To test
hypothesis 1, a typical model of public finance will be estimated. The following equation describes such an expenditure function: expenditure = f(democracy, economy, politics, sociodemographic factors, culture). In this model, government expenditure is regarded as a function of the degree of direct democracy. As controlling variables, also included are measures accounting for fiscal decentralization (defined as share of local expenses in total cantonal and local expenses), urbanization of the canton, cantonal wealth, size of canton, tax competition, a fiscally effective constitutional ‘break’, the share of young, old people ( < 20 years, > 60 years, respectively) and educated people, government ideology (with positive values indicating a conservative position), coalition size, and cantonal culture, proxied by the dominating language. A prediction of the impact of these controls and their theoretical foundation can be found in Feld, Fischer, & Kirchgässner (2006). Hypotheses 2 through 4 will be tested by estimating a reduced form and a structural form of an educational production function model. In contrast to the reduced form, the structural form also includes revenue-driven ‘endogenous’ input factors that serve as potential budgetary transmission channel of direct democracy. These endogenous input factors include teacher qualification, teacher shortages, total hours of schooling, student-teacher ratio, access to PCs, 14 availability and quality of instruction material, and state of school building or availability of space. Consequently, the structural form to be estimated looks as follows: performance = f(democracy, culture, individual, peers, school, canton, school inputs), where democracy denotes again direct democratic institutions, and culture the main regional culture of the school location. Individual denotes the student’s individual and family characteristics such as gender and parents’ education. Peers stands for peer group characteristics that aim at measuring the external effects of the peer group on an individual’s academic performance.18 School denotes non-revenue-driven school/class-related characteristics like the selectivity of the institution or problems with class discipline. Canton represents cantonal sociodemographics which serve as proxies for missing individual and peer group characteristics in class (e.g. religion, poverty). Finally, school inputs denote revenuedriven school inputs as described above. For predicting the impact of the sociodemographic and peer controls, see e.g. Winston & Zimmerman (2003) or Figlio (1997). In the reduced form of the model, the ‘endogenous’ variables are excluded, so that the following equation results: performance = f(democracy, culture, individual, peers, school, canton).
Estimation of both the reduced and the structural form of the model allows then to test hypotheses 2 through 4, in particular, (1) whether school inputs affect student performance, (2) whether they serve as transmission channel of direct democracy, and (3) whether a nonbudgetary influence of direct democracy prevails. Most of the US literature which identfies an institutional impact employs variations of the reduced form of the model, but not of the structural form. To my knowledge, only one estimation reported in Downes, Dye, & McGuire 15 (1998) resembles the structural model most.19 In other words, most of the US literature does not directly test the presence of a transmission channel of the budget constraining institution. 3.3 Methodology The expenditure regression is estimated with OLS using aggregate data that form a synthetic panel with 26 cantons as observational units per year between 1980 and 1998. Newey-West standard errors correct for heteroscedasticity and serial autocorrelation, and the Jarcque-Bera test assesses the presence of outliers. As robustness test, the identical model is estimated with those observations excluded with residuals above or below 1.5 standard deviation. As regards the educational production function, both the reduced and the structural model are estimated using ordinary least squares (OLS). Standard errors of the coefficients are corrected for heteroscedasticity, but also clustered at the school level (Moulton, 1990). This latter correction takes into account that students who attend identical schools share common factors both at the school and cantonal level – for example, condition of the school building and political institutions in the canton.20 Besides the peer effects discussed above, some of the remaining determinants of student achievement, however, might be subject to potential simultaneity. For example, frequency of individual homework feedback or a higher age could be proxies for bad grades at school. Additionally, the selection of pupils into different school types (and classes) is not fully taken into account with this estimation method. Simultaneity might induce a bias in the estimated coefficients. An instrumentation of endogenous variables or a correction for sample selection, however, cannot be carried out because the PISA data do not provide the necessary exogenous instruments (for a discussion, see also Rangvid 2004, Graddy & Stevens 2005).
For space constraints, the discussion of the estimation results for educational spending and student performance focuses on the influence of the variables of interest, in particular the extent of direct democratic rights and the ‘endogenous’ budget-driven input factors.21
(TO BE CONTINUED)
Justina A.V. Fischer*
Stockholm School of Economics
SSE/EFI Working Paper Series in Economics and Finance
No 688-December 2007
12. Earlier contributions to this topic from the 1970s until the very early 1990s, some empirical but most informal, are described in Downes & Figlio (1999). These contributions, however, suffer from methodological shortcomings.
13. In this study, although mitigated by competition among schools, a small negative impact of tax limits on student performance in mathematics is detected.
14. Some of them, however, employ educational spending at the district level as a controlling factor in addition to the policy measure and institutional dummies (e.g. Downes & Figlio 1997).
15. Combined cantonal and communal spending must be employed because in every single canton the financing of schooling is shared differently between these two tiers.
16. The PISA data are used in analyses of e.g. Fuchs & Woessmann (2004), Fertig & Wright (2005).
17. The peer group variables should be included to the model because an indirect institutional influence might be mediated through them. In this paper, however, we are interested in the direct institutional influence on a single student.
18. In small classes, there might even exist a feedback relation and continuing interaction between the one and the other(s) (for empirical literature on peer effects, see Zimmer & Toma, 2000; Summers & Wolfe, 1977; Epple, Romano, & Sieg, 2003; Rangvid, 2004).
19. The endogenous input factors in this study are district-level student-teacher ratios, student-administrator ratios, mean teaching experience of teachers, and share of teachers with B.A. or B.S. degree.
20. This estimation method is also applied by educational economists to the analysis of PISA results using an international sample containing several countries.
21. The full estimation results for the educational production function can be found in Table A5 of the Appendix.