Social democrats and the new partisan politics of ... - SSRN papers

Social democrats and the new partisan politics of public investment in education

Dr. Marius R. Busemeyer

Abstract: This paper studies the impact on public education spending of social democratic participation in government. By means of a pooled time-series analysis of spending in OECD democracies, it is shown that social democrats have increased public spending primarily on higher education. This finding is at odds with simple class-based models of partisan preferences (Boix) that predict a preference for non-tertiary education. As an alternative, the notion of a “new politics of public investment in education” (Iversen) is presented. From this perspective, political parties are not merely transmission belts for the economic interests of social classes, but use policies and spending strategically to attract and consolidate voter groups. By increasing public investment in tertiary education, social democrats cater to their core electoral constituencies (for example, by expanding enrollment) and, at the same time, new middle class constituencies to escape electoral dilemmas and re-forge the cross-class alliance with the middle class.

Key words: education, higher education, OECD countries, pooled time-series analysis, public spending, social democratic parties

1 Electronic copy available at: http://ssrn.com/abstract=1259855

1. Introduction What is the impact of social democratic government parties on public investment in education? Why should social democrats be interested in expanding public education in the first place? Are social democrats driven and constrained by external forces such as globalization or is there room for independent action? Can we observe differences in the dynamics of spending across educational sectors? In seeking to answer these questions, the present article makes both an empirical and a theoretical contribution. The empirical contribution comprises a comprehensive study of the impact of partisan factors on different types of education spending (total as well as sectoral). Despite the recent scholarly interest in the political economy of skill formation (Iversen 2005; Thelen 2004), study of the dynamics and determinants of education spending has been relatively neglected (notable exceptions are Castles 1982, 1989, 1998; Schmidt 2002, 2007; Busemeyer 2007). In the political economy literature, the most significant contribution is Boix (1997, 1998). His insight (building on Garrett / Lange 1991 and others) was that, in an era of economic internationalization, public investments in human capital formation have become a popular supply-side-oriented policy instrument for social democratic parties. Due to data limitations at the time, Boix did not test his argument with data on sectoral education spending (higher education vs. primary and secondary education). The theoretical contribution of this paper is an alternative explanatory approach to the Boix model. Empirically, this paper shows that social democratic parties have primarily pursued a strategy of increased investment in public higher education, in association with an expansion of university enrollment. From the purely class-based perspective of the Boix model, this empirical finding is hard to explain because it is not apparent why social democrats should increase spending in higher education – a policy area with limited redistributive capacity – instead of other social policies with a stronger redistributive component. In contrast, the notion of a “new partisan politics of public investment in education” (inspired by Iversen 2 Electronic copy available at: http://ssrn.com/abstract=1259855

2006a: 5) is put forward. This thesis holds that social democratic parties are not simple transmission belts for the economic interests of lower income classes, but try to appeal to their core constituency (by expanding enrollment) while at the same time reaching out to new middle class constituencies. The paper proceeds as follows: section 2 anchors the theoretical approach in the literature (partisan theory, power resources theory) and discusses the state of research on education spending. At the end of section 2, the notion of a “new partisan politics of public investment in education” is presented, along with testable hypotheses. Section 3 contains an empirical analysis of the impact of partisan forces on total and sectoral education spending (that is, higher education, primary and secondary education). Section 4 summarizes and concludes with remarks about the findings’ implications for the further development of partisan theory.

2. Theory and hypotheses 2.1 Partisan theory and welfare state development Partisan theory emphasizes the importance of partisan factors in explaining welfare state development. It comes in two “flavors”: First, for power resources theory (for example, Stephens 1979; Hicks / Swank 1984; Esping-Andersen / Korpi 1984; Esping-Andersen 1985; more recently: Hicks 1999; Huber / Stephens 2001; Bradley et al. 2003), economic classes build the fabric of society. The function of political parties is to mobilize these latent class interests and act as a “transmission belt” for political demands. Second, partisan theory (or party difference theory, see Hibbs 1977, 1987; Castles 1982; Schmidt 1982, 1996; Alt 1985; Chappell / Keech 1986) adopts a slightly different perspective. The theory is framed from the perspective of government parties and entails two claims: 1. Government parties matter: public policies are not shaped completely by exogenous forces (institutions, socio-economic factors), but the parties in power have an impact on policy output as well. 2. Government parties of the left and of the right behave differently in government and pursue different 3

policy programs. This claim is posited against the “partisan null hypothesis” of median voter models. The logic behind party differences is more “political” than sociological, as in the case of power resources theory. Parties differ because they pursue policies “in accordance with the objective economic interests and subjective preferences of their class-defined core political constituencies” (Hibbs 1987: 291) in order to get re-elected (Schmidt 1996: 156). Thus, political parties are not only “transmission belts” for socio-political demands, but utilize policies to cater to and consolidate their respective bases. The common element between power resources theory and party difference theory is the notion that partisan preferences can be derived from an inspection of the objective economic interests of classes or societal strata. The political struggle over public welfare is portrayed as a struggle over distribution (Huber / Stephens 2001: 17). In its simplest version, partisan theory assumes that leftist parties represent the interests of low income classes and thus favor more redistribution, for example, in the form of public spending, while rightist parties represent upper income classes who oppose public spending because they foot the tax bill.

2.2 Partisan theory and education spending Studies on the determinants of education spending (Cameron / Hofferbert 1974; Verner 1979, Castles 1982, 1989, 1998; Schmidt 2002, 2007; Busemeyer 2006a, 2006b, 2007; Wolf 2006; Nikolai 2007) are mostly y-centered, that is, they have a generic, explorative approach to the study of education expenditure, with partisan factors being one important determinant among other socio-economic or institutional variables. Still, they often find a significant impact of government parties on education spending, with leftist parties spending more on education than rightist parties.

Class-based approaches

4

Political economy models of partisan preferences for education policies (Ansell 2006; Boix 1997, 1998; Garrett / Lange 1991; Garrett 1998) are rooted in power resources and partisan theories, that is, party strategies and preferences are derived from the objective economic interests of their assumed core electoral constituencies. The model developed by Boix (1997, 1998) provides clear expectations concerning why social democratic parties pursue strategies of increased investment in education. Processes of economic liberalization and internationalization are exogenous background factors that impact on the availability of strategies of macroeconomic policy such that Keynesian-style demand management becomes increasingly unfeasible (Boix 1998: 2–3). Thus, investments in human capital – that is, increasing the productivity of low-skilled workers – are a viable alternative approach “to modify market outcomes and redistribute wealth to favor the least advantaged sectors and advance equality in general” (Boix 1998: 4). In contrast, conservative parties believe in the power of markets and want to maximize individual liberties (Boix 1998: 4). Consequently, the Boix model expects the government participation of social democrats to be associated with higher public spending on education, while conservatives in government are expected to reduce spending. Boix (1997, 1998) tested his model on total education spending only, but he hinted at some ideas about the relative importance of sectoral education spending. In line with the notion of a class-based model of redistributive politics, Boix (1998: 37) makes the point that “generic assets” imparted in the general school system are more important than various forms of postsecondary education. If the economic interests of the working class are taken as point of departure, one could argue that vocational training (that is, non-tertiary post-secondary education) should be equally important for the core social democratic constituencies. Public investment in higher education, however, should not be a priority for leftist government parties. The expansion of publicly financed tertiary education primarily benefits the (upper) middle classes, not the working class. And even if the children of the working class manage to 5

enter higher education, they will eventually end up in the higher income strata and so become members of the rightists’ constituency. Thus, with regard to sectoral education spending, the Boix model predicts a positive association of social democratic government parties with spending on primary, secondary, and non-tertiary post-secondary education, but a negative one with spending on tertiary education (see also Ansell 2006).

The new partisan politics of public investment in education The notion of what is here called “the new partisan politics of public investment in education” (Iversen 2006a, b) differs from purely class-based approaches on two points: First, following Cusack (1997) and Kitschelt (1994, 1999), I surmise that political parties are not simple transmission belts for the objectively definable economic interests of their constituencies. Instead, parties take into account public sentiments about various policy areas (Wlezien 1995, 2004) in order to attract new voter groups (primarily from the middle classes). Second, I posit that the process of educational expansion has not only impacted on the educational preferences of the median voter, but has also altered the preferences of left parties’ core constituencies. Before I elaborate on these ideas, it is important to point to the crucial importance of educational expansion as a background condition and driving force behind changing partisan and individual preferences for education policies. Educational expansion is a process of universalizing access to ever higher levels of education, spanning a large number of countries. In most industrialized countries, primary and secondary education became universalized in the post-war decades; educational expansion has continued in the area of higher (tertiary) education, but remaining differences between countries in terms of enrolment and spending are far more significant in the case of higher education than in that of primary and secondary education (UNESCO 2006; Busemeyer 2007). This leads to the expectation that partisan factors will contribute more to explaining the former. 6

More importantly, educational expansion impacts on the preferences of the median voter as well as the core constituents of left parties. A full-blown analysis of individual preferences for different types of education is beyond the scope of this article, but it can be expected that a side-product of educational expansion is the inflation of educational expectations both for the median voter and the core constituencies of left parties. How are political parties, and social democratic parties in particular, coping with this popular will to expand higher education? A simple and purely class-based model predicts that social democrats will want to install a partially private higher education system (tuition fees combined with scholarship programs for working class children) instead of subsidizing the “poor rich” (the middle class) through public higher education (Ansell 2006: 11). In contrast, I argue that political parties not only act as “transmission belts” for economic class interests, but try to blend ideological preferences with a sensitivity to the preferences of the electorate (Cusack 1997; see also Wlezien 1995, 2004). The challenge for social democratic parties is thus to cater for middle class voter groups without alienating their core constituencies. The expansion of public tertiary education can become a means of re-forging this cross-class alliance between the lower and the middle classes. High-quality public higher education is attractive to middle class voters because, in comparison to private education, it reduces the price they have to pay for university education. For lower income classes, the improvement of access to higher education institutions is most important. Thus, a crucial precondition for the long-term success of the cross-class coalition is that the expansion of public education institutions takes place alongside efforts to open up enrollment and to reduce the class bias in university access (Iversen 2006a: 5). Over time, the opening up of access to higher education in previous years results in a self-reinforcing circle: working class parents increasingly favor expanding public higher education institutions in order to send their children to universities, who in turn become supporters of the continuing expansion of public higher education. 7

In sum, while traditional class-based approaches expect social democrats to focus on “generic assets” (Boix 1998: 37) and to favor a partially private higher education system (Ansell 2006), according to the hypothesis presented here social democrat government participation is associated with increases in public spending on higher education. This can be explained only by taking into account the fact that political parties do not act as simple transmission belts for economic class interests, but use public policies to cater to new voting groups and to re-forge cross-class alliances. Despite this, it is important to keep in mind that the core claim of partisan theory (“parties matter”) remains relevant. In our case, the difference between leftist and rightist parties is that the former favor the expansion of public authority in the education system, while the latter prefer private alternatives. The reason social democrats prefer public higher education institutions is that in this case the decision on the expansion of access to higher education is not delegated to private institutions, but remains within the reach of public authority. It is also important to note that the strategy of expanding public education was used by social democratic government parties well before the 1990s (Schmidt 2007). But with the increasing internationalization of the economy, structural changes towards the service and knowledge economy, and the continuing process of educational expansion, the focus has shifted from secondary education and vocational training to higher education.

2.3 Hypotheses Summing up the theoretical discussion, I shall test the following hypotheses empirically: 1. Total education spending: Social democratic government participation is associated with increases in total public education spending, in particular when leftist government participation takes place alongside increasing trade openness (the Boix model). 2. Sectoral education spending: 8

2.a) Social democratic government participation is associated with increases in public spending on primary, secondary, and non-tertiary post-secondary (PSNTPS) education, but decreases in higher education spending (extension of Boix model). 2.b) Social democratic government participation is only weakly associated with spending on PSNTPS education, but positively associated with increases in public spending on tertiary education. Moreover, the expansion of spending is fuelled by partisan strategies in the domestic arena and thus is not related to changes in trade openness (extended and applied “new politics of public investment in education” argument (Iversen)).

3. Empirical analysis 3.1 Variables Dependent variables The central dependent variables analyzed here are total public education spending, public spending on primary, secondary, and non-tertiary post-secondary (PSNTPS) education, as well as public spending on tertiary education (taken mainly from the OECD’s Education at a Glance database).1 All spending variables are given as a percentage of Gross Domestic Product (GDP) and cover public spending on educational institutions.2 The time period covered ranges from 1980 to 2002 in the case of total public education spending, and from 1991 to 2002 for sectoral education spending because reliable data on the latter (namely public investment in higher education) have been available only since the 1990s (cf. Schmidt 1

Details on sources as well as descriptive statistics can be found in the online Appendix at:

www.mpifg.de/people/bus. 2

The OECD distinguishes public spending on educational institutions (narrow definition of education spending)

from “total public expenditure on education” (OECD 2006: 222), which includes education-related public subsidies to households. I decided to stick with the narrow definition because in the case of total public expenditure on education, cross-national ambiguities concerning the true classification of spending programs remain.

9

2007). The analysis covers 21 OECD democracies.3 In addition to spending, we will take a brief look at tertiary enrollment as dependent variable to check whether the partisan effect of social democrats is mainly quantitative (expansion of enrollment and spending) or both qualitative and quantitative (expansion of spending on top of expansion of enrollment).

Independent variables and controls The most important independent variable is, of course, the government participation of social democrats. The share of cabinet seats held by social democrats and various other party families in a given country-year is taken from Schmidt (2003). Another important independent variable is exposure to the forces of economic globalization. In the Boix model, the internationalization of the economy is a background condition forcing social democrats to adapt new partisan strategies. But if there really is a connection between internationalization and social democrats’ willingness to invest in education, we should expect a stronger reorientation towards human capital investment in those countries that are very open and governed by social democrats (interaction effect). In principle, if the Boix model is taken seriously, this interaction effect should hold regardless of the type of education spending. Alternatively, trade openness might impact on education spending independently of the partisan composition of governments (Ansell 2008; Crouch et al. 1999). As a measure for trade openness, the average of exports and imports as a percentage of GDP is used.4 The analyses also include a number of control variables. Demographic demand for education services is captured by an “age” variable (i.e., the ratio between the population share of those aged 65 and above, and that of those aged 5 to 29). Female labor force participation is included because it can be expected to increase the demand for educational services, both in 3

Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan,

Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, UK, USA. 4

See the Appendix for a justification of the use of this measure.

10

the PSNTPS and in the higher education sector. The level and change of economic well-being (gross national income (GNI) per capita) is included to capture the impact of fluctuations in the business cycle on changes in spending. A dummy for federal countries is included to control for the fact that federalism is generally ascribed a retarding effect on public spending (Obinger et al. 2005). Also, policy-making responsibilities for education are often delegated to lower levels of government in federal countries. Finally, I include the total level of public spending (as a percentage of GDP) and the level of gross public debt to model the budget constraint of spending decisions, which takes into account the fact that decisions about individual policy areas have to be considered within the context of changing budget conditions.

3.2 Statistical methods Despite the frequent use and popularity of the framework of time-series cross-section analyses, a common and widely accepted methodological standard has yet to emerge, as evinced by current discussions in the relevant literature (Kittel / Winner 2005; Plümper et al. 2005; Plümper / Troeger 2007). The core hypotheses of this article were tested using a variety of model specifications. Due to space constraints, I focus on the one specification that seems most advisable. Interested readers can find a full description of the statistical methods used as well as further empirical evidence and multiple robustness checks in the online Appendix. From a theoretical standpoint, we are interested in changes in spending as a result of changes in the partisan composition of governments. The preferred statistical method for testing this kind of hypothesis is an error-correction model (ECM) (Beck 1991; Iversen / Cusack 2000; Kaufman / Segura-Ubiergo 2001). Here, the dependent variable is given in changes (first differences) and the independent variables are included both as lagged levels (t-1) and changes. The coefficients on the lagged levels of independent variables give an estimate of the permanent (long-term) effect of a change in this variable on the dependent variable, while the 11

coefficients on the changes in these variables are an estimate of the transitory (short-term) effects of a change in the respective variable. As is customary, I use panel-corrected standard errors as advised by Beck and Katz (1995, 1996; Beck 2001).

3.3 Findings

[Table 1 about here]

Government participation of social democrats The impact of the main independent variable of interest, the government participation of social democrats, varies considerably from one type of spending to another. From table 1 it can be seen that the simple version of the Boix model (hypothesis 1) finds some, but not overwhelming support. Government participation of social democrats has a positive and statistically significant impact on total public education spending in the long run, but a negative transitory impact (models 1 and 2 of table 2), independent of whether one looks at the longer time period (model 1) or only the 1990s (model 2).5 The long-term permanent impact of a change from a government with no social democrats to one consisting only of social democrats (for example, in Australia (1983), Greece (1981 and 1993), New Zealand (1984 and 1999), Norway (1986), Spain (1982), Sweden (1982 and 1994), the UK (1997)) is an increase in public education spending of 0.5 percentage points.6 This is about 10 percent of the OECD average for public education spending for 2002.

5

Other studies (Busemeyer 2006a, b; Schmidt 2007) have shown that partisan effects had some impact in the

1980s, but that this effect disappears for the larger observation period and the 1990s. 6

This effect is calculated by dividing the coefficient of the lagged level variable by the coefficient of the lagged

dependent variable (see Iversen / Cusack 2000: 330).

12

When sectoral education spending is chosen as the dependent variable instead of total education spending, it becomes obvious that the relationship between partisanship and education spending is more complex than is assumed by the simple version of the Boix model. When we look at the impact of leftist parties on primary, secondary, and post-secondary education spending (model 3), we find the expected positive association, although it is not statistically significant in all specifications (see Appendix). In addition, the positive association is mainly a long-term effect. According to the coefficient estimates of model 3, the long-term permanent effect of a transition from a government without social democratic participation to one with only social democrats is about 0.6 percentage points (with an OECD average of roughly 3.6 percent of GDP). Thus, in comparison to the effect on total public education spending, the magnitude of the effect on PSNTPS education spending is larger (16.6 vs. 10 percent of the OECD average). Analysis of the determinants of public spending on higher education (model 5) reveals a more coherent finding, however. Here, we find a clearly positive association between the government participation of social democrats and spending. The positive association holds regardless of the model specification (see Appendix). More importantly, the positive effects of social democratic government participation show up both in the long term (permanent effect) and in the short term (transitory effect). Both coefficients have a positive sign and are statistically significant. The magnitude of the partisan effect is considerable: Based on the coefficient estimates of model 5, the predicted long-term permanent impact of a change in government from one with no social democratic participation to one consisting only of social democrats is an increase in spending of 0.25 percentage points (with an OECD average of 1.1 percent of GDP in 2002). The magnitude of the effect on higher education spending is larger (23 percent) than in the case of PSNTPS spending (17 percent) or total education spending (10 percent). 13

In sum, a preliminary conclusion of the analysis is that the dynamics of education spending are complex. First, the relationship between leftist government parties and total education spending is somewhat inconclusive. Second, with regard to sectoral education spending, social democrats have preferred to increase spending on tertiary education to a greater extent than spending on primary and secondary education.

Trade openness and its interaction with social democratic government participation Above, it was suggested that trade openness is more than a diffuse background condition that can be held constant. The findings in table 1 suggest that differences in trade exposure have important consequences for public investment in human capital formation, depending on the type of spending. First, the direct association between trade openness and total education spending (see models 1 and 2 in table 2) is not robust. The association between spending on PSNTPS education and trade openness is equally weak (models 3 and 4) and consistently fails to attain statistical significance. The strongest impact of trade openness on education spending, however, can be seen in the case of tertiary education. Here we find a positive and significant short-term impact, indicating that advances in economic internationalization are accompanied by increases in public investment in tertiary education. In sum, if there is evidence that economies open to international economic competition have increased public investment in education, it can be found primarily in the dynamics of higher education.

[Figure 1 about here]

What about the interaction between trade openness and the government participation of social democrats? Figure 1 contains plots of this interaction term and its effects on total public education spending (chart A), PSNTPS education spending (chart B), and tertiary education 14

spending (chart C).7 The first chart in Figure 1 suggests that, for low levels of trade openness, the marginal effect of social democratic government participation on total education spending is actually negative. In line with the Boix argument, the effect becomes positive with higher levels of trade openness. But the 95% confidence band almost always includes zero. Consequently, the interaction between trade openness and social democratic government participation fails to meet common standards of statistical significance in the case of total education spending. A look at the dynamics of sectoral education spending might explain why there is no significant relationship on the aggregate level. Chart B shows that there is a strong interaction effect for the case of PSNTPS education spending. This finding is in line with the extended version of the Boix argument (hypothesis 2a): social democrats in government in open economies increase spending in non-tertiary education to cushion the impact of economic internationalization for their core electoral constituencies. However, the findings for the case of higher education (chart C) show that, in addition to the Boix logic, another mechanism must be at work here. Chart C shows that the positive impact of social democratic government participation on public spending in tertiary education is completely independent of trade openness. The slope of the estimated marginal effect of social democrats stays virtually constant across the range of values for trade openness and the confidence interval includes zero almost all the time. Hence, the positive impact of social democratic government participation on higher education spending is independent of trade openness. The Boix logic does not hold in the case of tertiary education.

Tertiary enrollment Finally, I shall briefly analyze the impact of government participation of social democrats on tertiary enrollment. This is necessary to determine the extent to which the expansion of 7

Graphs were plotted using the STATA code provided by Brambor et al. (2006).

15

spending has gone along with an expansion of access to higher education. According to the logic of the “new politics of public investment in education,” social democratic efforts in increasing public spending on tertiary education should go along with an expansion in enrollment. And, according to the results presented in table 1, this is exactly what has happened. The cabinet share of social democrats (moving average) is positively and significantly associated with changes in gross tertiary enrollment during the 1990s (model 8). Because of better data availability, it was possible to extend the period of analysis back to 1980 (model 7). For the time period 1980 to 1991, the effect of social democratic government participation dilutes, loses statistical significance, and even changes sign (see also table 5 in the Appendix). This finding could indicate that social democrats in government have intensified their commitment to the expansion of access to public higher education only since the 1990s. Still, one must conclude that the relationship between social democratic government participation and tertiary enrollment is not as strong as in the case of public spending. Furthermore, it is stronger in the analysis of levels of enrollment than in the analysis of changes: differences in access to higher education seem to be more resilient to change, and country differences are probably shaped more by partisan heritage than the current partisan composition of the government. In other words: leftist parties in government have expanded public spending on higher education and have also tried to increase access, but the new emphasis on tertiary education goes beyond a simple quantitative expansion; it is also part of an electoral strategy to attract new constituency groups in the middle classes.

Control variables Without going into too much detail, a brief comment on the performance of the control variables is in order. Demographic demand, captured by the age ratio variable, is positively associated with total education spending as expected, that is, the negative coefficient 16

estimates indicate that a higher share of young people is associated with more spending. The business cycle (levels and change of Gross National Income per capita) have either no discernible or a negative effect on public education spending, which evidences a relative insensitivity of education to short-term fluctuations in the business cycle. Female labor force participation is associated positively with all types of education spending, as expected. The federalism dummy has a relatively weak explanatory power. The variables modeling the budget constraint (total public spending, gross public debt) were included primarily to get an unbiased estimate of the other independent variables of education spending. Nevertheless, their performance shows that education spending moves in line with general budget cycles. Increases in public spending are associated with increases in education spending. Rising levels of public debt, however, result in lower education spending.

4. Summary and conclusion This article has studied the impact of social democratic government participation on public education spending. It has been shown that partisan effects play out very differently in different educational sectors. Hence, it is necessary to analyze sectoral education spending in addition to total education spending. Second, economic internationalization is more than a diffuse background condition. The results suggest that trade openness and social democratic government participation have a joint impact on spending in the case of PSNTPS education, but not in the case of higher education. Furthermore, trade openness has an additional, separate impact on education spending, predominantly in the case of tertiary education. Third, social democrats in government have expanded access opportunities, but efforts to increase public spending have been more successful and consistent. Partly this can be explained by the resilience of class biases in university access, partly by the limited influence government parties can exert on outcome indicators such as enrollment. In addition, increasing public investment in higher education serves as an instrument by means of which social democratic 17

parties can re-forge the cross-class alliance with the middle class. Since the expansion of the welfare state has come to a halt in most industrialized countries, leftist parties have to look for new ways of solving the electoral dilemma of catering to their former core constituencies and appealing to new voter groups in the middle class (Kitschelt 1994, 1999). Public investment in higher education can help to solve this dilemma. What are the implications of the findings for partisan theory in general? First, it has clearly been shown that partisan differences continue to matter. Even controlling for a number of socio-economic and institutional variables, the partisan composition of the government continues to be an important determinant of policy outputs in terms of spending. This finding stands in contrast with the generally prevailing opinion that, in an era of retrenchment and progressing internationalization, partisan effects have lost their explanatory power with regard to short-term changes and that they remain important mostly with regard to explaining historical, cross-national differences (Castles 1998; Kittel / Obinger 2003). This article has shown that partisan differences matter, both in the short term and in the long term. Instead of arguing about the general relevance or irrelevance of partisan differences, we should engage in more concrete research into where partisan differences matter, and why they matter in some areas and not in others. Second, partisan theory needs to be less static in building hypotheses about partisan preferences. For the sake of parsimonious models, it might be defensible to rely on “naive” models of class politics (Boix 1998: 40). But these models have merit only as long as they produce hypotheses that can explain empirical reality. The simple class perspective, in which social democrats represent the interests of the lower classes and rightist parties are the spokespersons for upper income classes, neglects changes in the composition of the most important electoral constituencies of political parties and thus the impact of these changes on party and government policies. It also neglects the fact that political parties are more than just “transmission belts” of socio-political demands in that they also cater to electoral 18

constituencies out of strategic interests. In the case of higher education, social democrats are trying to cater to the interests of their core constituency by expanding access to higher education, but they are also trying to appeal to new middle class voting groups by emphasizing education instead of other social policies. The consequence for partisan theory is that it should be more aware of the micro-level foundations of partisan preferences and the logic of party competition, without giving up its central claim: parties continue to matter.

19

Biographical note: Marius R. Busemeyer is a researcher at the Max Planck Institute for the Study of Societies. His research focuses on education and social policies in advanced industrial countries. He has published, amongst others, in West European Politics, Journal of European Public Policy and Politische Vierteljahresschrift.

Address for correspondence:

Dr. Marius R. Busemeyer Max Planck Institute for the Study of Societies Paulstr. 3 D-50676 Köln [email protected] www.mpifg.de/people/bus Acknowledgements: I would like to thank Achim Goerres, Alexander Petring, Armin Schäfer, Martin Schröder, Frieder Wolf, and Guido Tiemann for their helpful comments and suggestions. Part of this research was conducted during a research project financed by the German Research Group (DFG) under the direction of Manfred G. Schmidt. I would like to thank the Max Planck Institute for the Study of Societies (Cologne) for providing a congenial environment in which to finish this project. Word Count: 6,750 plus one table and one figure Date of submission: September 28, 2007 Revised version: December 22, 2007 Final version: February 28, 2008.

20

References: Alt, J. E. (1985) ‘Political Parties, World Demand, and Unemployment: Domestic and International Sources of Economic Activity’, American Political Science Review 79 (4): 1016–40. Ansell, B. (2006) ‘University Challenges: Explaining Institutional Change in Higher Education’, paper presented at the MPSA Conference, Chicago, April 2006. Ansell, B. (2008) ‘Traders, Teachers, and Tyrants: Democracy, Globalization, and Public Investment in Education’, International Organization (2008): 289-322. Beck, N. (1991) ‘Comparing Dynamic Specifications: The Case of Presidential Approval’, Political Analysis III: 51–87. ——— (2001) ‘Time-Series-Cross-Section Data: What Have We Learned in the Past Few Years?’, Annual Review of Political Science 2001 (4): 271–93. Beck, N. and Katz, J. N. (1995) ‘What to Do (and Not to Do) with Time-Series Cross-Section Data’, American Political Science Review 89 (3): 634–47. ——— (1996) ‘Nuisance vs. Substance: Specifying and Estimating Time-Series-CrossSection Models’, Political Analysis 6: 1–36. Boix, C. (1997) ‘Political Parties and the Supply Side of the Economy: The Provision of Physical and Human Capital in Advanced Economies, 1960–1990’, American Journal of Political Science 41 (3): 814–45. ——— (1998) Political Parties, Growth and Equality: Conservative and Social Democratic Economic Strategies in the World Economy, Cambridge, New York: Cambridge University Press. Bradley, D., Huber, E., Moller, S., Nielsen, F. and Stephens, J. D. (2003) ‘Distribution and Redistribution in Postindustrial Democracies’, World Politics 55 (January 2003): 193– 228. Brambor, T., Clark, W. R. and Golder, M. (2006) ‘Understanding Interaction Models: 21

Improving Empirical Analyses’, Political Analysis 14: 63–82. Busemeyer, M. R (2006a) ‘Der Kampf um knappe Mittel: Die Bestimmungsfaktoren der öffentlichen, privaten und sektoralen Bildungsausgaben im OECD-Länder-Vergleich’, Politische Vierteljahresschrift 47 (3): 393–418. ——— (2006b) Die Bildungsausgaben der USA im internationalen Vergleich: Politische Geschichte, Debatten und Erklärungsansätze, Wiesbaden: Deutscher UniversitätsVerlag. ——— (2007) ‘The Determinants of Public Education Spending in 21 OECD Democracies, 1980–2001’, Journal of European Public Policy 14 (4): 582–610. Cameron, D. R. and Hofferbert, R. I. (1974) ‘The Impact of Federalism on Education Finance: A Comparative Analysis’, European Journal of Political Research 2: 225– 58. Castles, F. G. (1982) ‘The Impact of Parties on Public Expenditure’, in The Impact of Parties: Politics and Policies in Democratic Capitalist States, ed. F. G. Castles. London: Sage. ——— (1989) ‘Explaining Public Education Expenditure in OECD Nations’, European Journal of Political Research 17: 431–48. ——— (1998) Comparative Public Policy: Patterns of Post-War Transformation, Cheltenham: Edward Elgar. Chappell, H. W. and Keech, W. R. (1986) ‘Party Differences in Macroeconomic Policies and Outcomes’, American Economic Review 76 (2): 71–74. Crouch, C., Finegold, D. and Sako M. (1999) Are Skills the Answer? The Political Economy of Skill Creation in Advanced Industrial Countries, Oxford, New York: Oxford University Press. Cusack, T. (1997) ‘Partisan Politics and Public Finance: Changes in Public Spending in the Industrialized Democracies, 1955–1989’, Public Choice 91: 375–95. Esping-Andersen, G. (1985) ‘Power and Distributional Regimes’, Politics & Society 14 (2): 22

223–56. Esping-Andersen, G. and Korpi W. (1984) ‘Social Policy as Class Politics in Post-War Capitalism: Scandinavia, Austria, and Germany’, in J. H. Goldthorpe (ed.), Order and Conflict in Contemporary Politics, Oxford: Oxford University Press, pp. 179-208. Garrett, G. (1998) Partisan Politics in the Global Economy, Cambridge, New York: Cambridge University Press. Garrett, G. and Lange, P.(1991) ‘Political Responses to Interdependence: What’s ‘Left’ for the Left?’, International Organization 45 (4): 539–64. Hibbs, D. A. (1977) ‘Political Parties and Macroeconomic Policy’, American Political Science Review 71: 1467–87. ——— (1987) The Political Economy of Industrial Democracies, Cambridge, London: Harvard University Press. Huber, E. and Stephens J. D. (2001) Development and Crisis of the Welfare State: Parties and Policies in Global Markets, Chicago: University of Chicago Press. Iversen, T. (2005) Capitalism, Democracy, and Welfare, Cambridge, New York: Cambridge University Press. ——— (2006a) ‘Class Politics is Dead! Long Live Class Politics! A Political Economy Perspective on the New Partisan Politics’, APSA-CP Newsletter 17 (2): 1–6. ——— (2006b) ‘Responses and Some Agenda Items for the Future Study of Democratic Capitalism’, Labor History 47 (3): 439–49. Iversen, T. and Cusack, T. R. (2000) ‘The Causes of Welfare State Expansion: Deindustrialization or Globalization?’, World Politics 52: 313–49. Iversen, T. and Soskice, D. (2001) ‘An Asset Theory of Social Policy Preferences’, American Political Science Review 95 (4): 875–93. Kaufman, R. R. and Segura-Ubiergo, A. (2001) ‘Globalization, Domestic Politics, and Social Spending in Latin America: A Time-Series Cross-Section Analysis, 1973–97’, World 23

Politics 53 (July 2001): 553–87. Kitschelt, H. (1994) The Transformation of European Social Democracy, Cambridge, New York: Cambridge University Press. ——— (1999) ‘European Social Democracy between Political Economy and Electoral Competition’, in H. Kitschelt, P. Lange, G. Marks and J. D. Stephens (ed.) Continuity and Change in Contemporary Capitalism, Cambridge, New York: Cambridge University Press, pp. 317-345. Kittel, B. and Obinger, H. (2003) ‘Political Parties, Institutions, and the Dynamics of Social Expenditure in Times of Austerity’, Journal of European Public Policy 10 (1): 20–45. Kittel, B. and Winner, H. (2005) ‘How Reliable Is Pooled Analysis in Political Economy? The Globalization-Welfare State Nexus Revisited’, European Journal of Political Research 44: 269–93. Nikolai, R. (2007) Die Bildungsausgaben der Schweiz im intranationalen und internationalen Vergleich, Berlin: dissertation.de. Obinger, H., Leibfried, S. and Castles, F. G. (ed.) (2005) Federalism and the Welfare State: New World and European Experiences, Cambridge, New York: Cambridge University Press. ——— (2006) Education at a Glance: OECD Indicators 2006, Paris: OECD. Plümper, T. and Troeger, V. (2007) ‘Efficient Estimation of Time Invariant and Rarely Changing Variables in Panel Data Analysis with Unit Effects’, Political Analysis 15 (2): 124–139. Plümper, T., Troeger, V. and Manow, P. (2005) ‘Panel Data Analysis in Comparative Politics: Linking Method to Theory’, European Journal of Political Research 44 (2): 327–54. Schmidt, M. G. (1982) Wohlfahrtsstaatliche Politik unter bürgerlichen und sozialdemokratischen Regierungen: Ein internationaler Vergleich, Frankfurt a.M., New York: Campus. 24

——— (1996) ‘When Parties Matter: A Review of the Possibilities and Limits of Partisan Influence on Public Policy’, European Journal of Political Research 30: 155–83. ——— (2002) ‘Warum Mittelmaß? Deutschlands Bildungsausgaben im internationalen Vergleich’, Politische Vierteljahresschrift 43: 3–19. ——— (2003) ‘The Partisan Composition of Governments in OECD Democracies Dataset’, Heidelberg: University of Heidelberg, Institute for Political Science. ——— (2007) ‘Testing the Retrenchment Hypothesis: Education Spending, 1960–2002’, in F. G. Castles (ed.), The Disappearing State? Retrenchment Realities in an Age of Globalisation, Cheltenham: Edward Elgar, pp. 159-183. Stephens, J. D. (1979) The Transformation from Capitalism to Socialism, Urbana, Chicago: University of Illinois Press. Thelen, K. (2004) How Institutions Evolve: The Political Economy of Skills in Germany, Britain, the United States and Japan, Cambridge, New York, Melbourne: Cambridge University Press. UNESCO (2006) Global Education Digest 2006: Comparing Education Statistics across the World, Montreal: UNESCO Institute for Statistics. Verner, J. G. (1979) ‘Socioeconomic Environment, Political System, and Educational Policy Outcomes: A Comparative Analysis of 102 Countries’, Comparative Politics 11 (2): 165–87. Wlezien, C. (1995) ‘The Public as Thermostat: Dynamics of Preferences for Spending.’ American Journal of Political Science 39 (4): 981–1000. ——— (2004) ‘Patterns of Representation: Dynamics of Public Preferences and Policy.’ The Journal of Politics 66 (1): 1–24. Wolf, F. (2006) Die Bildungsausgaben der Bundesländer im Vergleich: Welche Faktoren erklären ihre beträchtliche Variation? Münster: LIT.

25

Figure 1: Interaction effects between trade openness and various types of education spending Chart A: Openness and total education spending Marginal Effect of SD on Spending As Period Changes

.005 0 -.005 -.01

Marginal Effect of SD

.01

Dependent Variable: Public Spending on Education

0

10

20

30

40

50

60

70

80

Trade Openness

Marginal Effect of SD 95% Confidence Interval

Chart B: Openness and public spending on PSNTPS education Marginal Effect of SD on Spending As Openness Changes

.01 .005 0 -.015 -.01 -.005

Marginal Effect of SD

.015

.02

Dependent Variable: Public Spending on PSNTPS Education

0

10

20

30

40

50

60

70

80

Trade Openness

Marginal Effect of SD 95% Confidence Interval

Chart C: Openness and public spending on tertiary education Marginal Effect of SD on Spending As Openness Changes

.005 0 -.005

Marginal Effect of SD

.01

Dependent Variable: Public Spending on Tertiary Education

0

10

20

30

40

50

60

70

80

Trade Openness

Marginal Effect of SD 95% Confidence Interval

Note: SD=Social democrats. 26

Table 1: The impact of government participation of social democrats on various types of education spending. Dependent variable

(1) (2) Public education spending (% of GDP), first differences (changes)

(3) (4) (5) (6) Public spending on primary, Public spending on tertiary secondary and non-tertiary posteducation (% of GDP), first secondary education, first differences (changes) differences (changes) 1991–2002 1991–2002 1991–2002 1991–2002 Lagged Level Dependent Variables

Time period

1980–2002

1993–2002

Public education spending, lagged level (t–1) Public spending on PSNTPS education, lagged level (t–1) Public spending on tertiary education, lagged level (t–1) Gross tertiary enrollment, lagged level (t–1)

–0.1423 (6.73)***

–0.1431 (3.26)***

Cabinet share of social democrats, lagged level (t–1) Cabinet share of social democrats, change (∆) Interaction: social democrats and openness, lagged level (t–1) Cabinet share of social democrats (moving average), lagged level (t– 1)

0.0007 (2.19)**

0.0010 (2.19)**

Social Democratic Government Participation and its interactions 0.0013 –0.0010 0.0011 0.0024 (3.43)*** (0.66) (2.81)*** (1.85)*

–0.0013 (1.93)*

–0.0006 (0.51)

0.0012 (1.12)

–0.2202 (3.00)***

(7) (8) Gross tertiary enrollment, first differences (changes) 1980–1991

1992–2002

–0.8259 (2.49)**

–0.0970 (2.64)***

–0.0068 (0.09)

–0.0037 (0.29)

–0.0223 (0.36)

0.0273 (2.10)**

–0.2268 (3.08)*** –0.4321 (3.41)***

0.0013 (1.27) 0.7863 (1.65)*

0.0018 (1.92)*

–0.4422 (3.38)***

0.0017 (1.79)* –0.4348 (1.14)

Control Variables

27

Age ratio, lagged level (t–1) Age ratio, change (∆) National income per capita, lagged level (t–1) National income per capita, change (∆) Female labor force participation, lagged level (t–1) Female labor force participation, change (∆) Federalism

–0.5526 (3.30)*** 2.0471 (1.47) 0.0056 (1.91)* –0.0605 (2.60)*** 0.0050 (2.55)**

–0.2081 (1.28) 3.1475 (1.59) 0.0068 (1.23) –0.0799 (2.62)*** 0.0073 (1.82)*

–0.1605 (0.93) –1.4596 (0.59) 0.0011 (0.19) –0.0625 (1.91)* 0.0052 (1.70)*

–0.2475 (1.25) –1.3235 (0.55) 0.0011 (0.19) –0.0631 (1.94)* 0.0040 (1.36)

0.0010 0.0088 –0.0053 –0.0088 (0.11) (0.37) (0.23) (0.37) –0.0220 –0.0467 –0.0175 –0.0240 (0.69) (0.95) (0.64) (0.88) Public spending, lagged 0.0113 0.0094 0.0037 0.0035 level (t–1) (5.66)*** (2.95)*** (1.38) (1.33) Public spending, change 0.0436 0.0375 0.0054 0.0054 (∆) (5.16)*** (2.49)** (0.30) (0.31) Gross public debt, lagged –0.0015 –0.0014 –0.0015 –0.0020 level (t–1) (4.92)*** (4.16)*** (3.45)*** (3.01)*** Gross public debt, 0.0018 –0.0027 –0.0008 –0.0002 change (∆) (0.79) (0.68) (0.21) (0.06) Trade openness, lagged 0.0015 0.0011 –0.0003 –0.0024 level (t–1) (2.37)** (1.07) (0.40) (1.73)* Trade openness, change 0.0020 0.0029 0.0184 0.0204 (∆) (0.38) (0.26) (1.85)* (2.03)** Constant 0.0903 –0.1392 0.4801 0.7185 (1.36) (1.20) (2.12)** (2.74)*** R² 0.23 0.28 0.15 0.16 Observations 390 188 201 201 Number of countries 21 21 21 21 z statistics in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%.

–0.5169 (1.76)* –2.2090 (1.57) 0.0107 (1.36) –0.0381 (0.91) 0.0072 (1.89)*

–0.4882 (1.72)* –2.2971 (1.60) 0.0109 (1.40) –0.0378 (0.90) 0.0082 (1.80)*

–33.9769 (0.54) 29.1654 (0.05) 1.0559 (1.11) –0.7417 (0.17) 0.1986 (1.11)

–5.0268 (0.90) –2.9889 (0.13) –0.1150 (1.52) 0.5353 (1.43) 0.0481 (1.43)

–0.0050 (0.19) –0.0020 (0.06) 0.0076 (2.17)** –0.0020 (0.10) –0.0004 (0.83) –0.0050 (0.87) 0.0011 (1.04) 0.0258 (2.26)** –0.3266 (1.72)* 0.31 202 21

–0.0031 (0.12) 0.0015 (0.04) 0.0081 (2.13)** –0.0018 (0.09) –0.0002 (0.27) –0.0056 (0.96) 0.0022 (1.82)* 0.0248 (2.14)** –0.4597 (1.82)* 0.31 202 21

–1.4194 (1.59) 7.1761 (0.93)

–0.6983 (2.23)** –0.3952 (0.51)

–0.0406 (0.44) 0.4918 (1.23) 16.4739 (2.41)** 0.41 223 21

–0.0066 (0.46) –0.2640 (2.37)** 8.1004 (2.84)*** 0.24 210 21

28

Appendix Data sources Public education spending as a percentage of GDP Public spending on primary, secondary, and non-tertiary postsecondary education as a percentage of GDP Public spending on higher education as a percentage of GDP Tertiary enrollment Output gap GNI per capita Female labor force participation Population share of those aged between 5 and 29 Population share of those aged 65 and above Trade openness Cabinet share of social democrats Cabinet share of rightist parties Veto index

Real public educational expenditure as a percentage of real GDP. Source: OECD: Public Educational Expenditure, Costs and Financing: An Analysis of Trends 1970–1988, Paris, 1992, p. 84; OECD: Education at a Glance, Paris, several years. Public expenditure on education as a percentage of GDP for primary, secondary, and non-tertiary post-secondary education. Source: OECD: Education at a Glance, Paris, several years. Public expenditure on education as a percentage of GDP for tertiary education. Source: OECD: Education at a Glance, Paris, several years. Gross enrollment ratios, tertiary level. Source: UNESCO Statistical Yearbook, several years, and World Development Indicators. Output gap between real and potential economic output. Source: OECD Economic Outlook Database. Gross national income per capita in US dollars, current prices, PPP. Source: OECD Factbook 2006 (www.sourceoecd.com). Female total labor force as % of population from 15–64 years. OECD: Labour Force Statistics, 2003, Paris (1982–2002); OECD: Labour Force Statistics, 1984, Paris (1963–1983). Share of those aged 5 to 29 relative to total population. Source: OECD: Education at a Glance, Paris, various years. After 1996 own calculations, based on UNESCO Statistical Yearbook. Share of those aged 65 and above relative to total population. Source: OECD (2004): Health Data, Paris. Average of imports and exports (as a % of GDP). Source: OECD Factbook 2006 (www.sourceoecd.com). Cabinet share of social democratic parties. Source: Schmidt (2003). Sum of cabinet share of conservative, Christian democratic, liberal, and radical right parties. Source: Schmidt (2003b). Schmidt’s index of veto players. Source: Schmidt, Manfred G. (2000) Demokratietheorien, Opladen: Leske + Budrich, p. 352.

Statistical note: Linear interpolation was used to fill in missing values (one or two years max).

29

Note on statistical methods and more empirical evidence Despite the frequent use and popularity of the framework of time-series cross-section analyses, a common and widely accepted methodological standard has yet to emerge, as evinced by current discussions in the relevant literature (Kittel / Winner 2005; Plümper et al. 2005; Plümper / Troeger 2007). The “de facto Beck–Katz standard” (Plümper et al. 2005: 327), that is, using panel-corrected standard errors (PCSE) instead of simple OLS and including a lagged dependent variable (LDV) on the right hand side to control for serial correlation (Beck / Katz 1995, 1996; Beck 2001), has been used in a large number of studies. While PCSE are a widely accepted means of overcoming the problem of panel heteroskedasticity, the inclusion of an LDV has been criticized because it depresses the impact of the other explanatory variables (Plümper et al. 2005: 338; Kittel / Winner 2005: 276), inflates the R² (Iversen 2001: 64, fn. 10), and transforms the perspective of the analysis from a pooled model to a “within effects” (over time) model (Huber / Stephens 2001: 59–60). Another issue is the question of including unit (country) fixed effects (FE). The exclusion of FE may contribute to omitted variable bias when differences across units dominate and obfuscate associations within a given unit (country). In contrast, the inclusion of FE drains the pool of cross-sectional variation (“within-transformation”) and diminishes the explanatory impact of variables that are largely time-invariant (Plümper et al. 2005: 331; Plümper / Troeger 2007). A third problem currently discussed is stationarity of time series. Most time series of public spending items are non-stationary. In the case of (total and sectoral) education spending, Fisher tests suggest that the null hypothesis of non-stationarity can be rejected in most cases, although a close inspection of individual countries shows that the problem of nonstationarity cannot be completely ignored. Therefore, from a statistical viewpoint, first differencing the data is advisable. This, however, “crucially changes the interpretation of our estimation results” (Kittel / Winner 2005: 279) because, again, first differencing largely eliminates cross-sectional variance (ibid.: 281). 30

In sum, the current literature on methods of time-series cross-section analyses offers no easy way out, and difficult decisions about the appropriate specification of a model have to be made. Most of the problems discussed above point in the direction of transforming the models from “pooled” or “between effects” models to “within effects” models. A crucial precondition for such a transformation is that it fits the theory to be tested (Kittel / Winner 2005: 279–280; Plümper et al. 2005). In our case, this is possible because it is stipulated that changes in the partisan composition of governments will lead to changes in government expenditures. In order to demonstrate the robustness of the findings presented here and to address the concerns that have been raised in the literature, the following analyses and robustness tests will rely on three different specifications: 1. The “de facto Beck-Katz” standard will be used, because it is just that – a “de facto standard.” Its use is recommendable to make the results obtained here comparable to other studies (see above) and to assuage the concerns of Beck/Katz disciples. 2. Following suggestions by Plümper et al. (2005), an AR1 error correction process will be used instead of an LDV, thus freeing the estimation from its oppressive impact. In substantive terms, the difference between the LDV and AR1 specifications is smaller than one would assume. As Beck and Katz (2004) themselves have pointed out, the AR1 specification makes implicit use of a lagged dependent variable as well. 3. The third specification to be used is an error-correction model (ECM) (Beck 1991; Iversen / Wren 1998; Iversen / Cusack 2000; Iversen 2001; Kaufman / Segura-Ubiergo 2001). This specification has many advantages: The dependent variable is given in changes (first differences), thus eliminating eventual problems of non-stationarity. The independent variables are included as lagged levels and changes. The inclusion of lagged levels avoids the pitfalls of simple first difference models8 and allows

8

A simple first difference model specification includes dependent and independent variables only in changes. Neglecting the impact of levels of independent variables on changes in the dependent variables is questionable, however, because it is assumed that every change in the independent variable has the same impact regardless of

31

separating out long-term from short-term effects. The coefficients on the lagged levels of independent variables give an estimate of the permanent (long-term) effect of a change in this variable on the dependent variable, while the coefficients on the changes in these variables are an estimate of the transitory (short-term) effects of a change in the respective variable. This specification allows for a more nuanced treatment of the impact of government parties on spending. Its downside is that the dependent variable is given in first differences, so that the “within” perspective dominates. Therefore, the ECM specification should not be applied without due consideration of the theoretical underpinnings.

In addition, two further concerns by Plümper et al. (2005) are, at least partly, addressed. First, in order to evaluate the presence of period-specific effects, period dummies for the 1990s and interaction terms are used in the models on total public education spending. Second, given the long-term and often convoluted nature of policy-making processes, it has been questioned whether it makes sense to try to discern the impact of political parties on spending by looking at the current composition of the government. Instead, it has been advised to use a more differentiated lag structure (Plümper et al. 2005: 344) for partisan variables or a cumulative measure of cabinet shares to capture the “effect of long-term incumbency of a given political tendency” (Huber / Stephens 2001: 61). The measure used in the following analyses, in addition to more conventional measures, is a moving average of the cabinet share of social democrats, covering the last five years and attaching weights such that the effect of partisan heritage diminishes over time. Furthermore, I include an index measuring the strength of the constitutional veto structure (veto index) and its interaction with partisan effects to take into account the fact that the speed of progress in decision-making varies between countries.

the level. Plümper et al. (2005: 333) provide an example: Within a simple first difference model, it does not make a difference in terms of spending whether government participation of social democrats increases by five percentage points in the US or in Sweden. See also Beck (1991).

32

In this respect, Plümper et al. (2005: 344–345) propose using an “optimized lag structure” instead of a uniform lag structure. This approach was considered as well, but ultimately rejected. There are three reasons for this: First, “[t]here is no generally accepted indicator for the determination of the length of the lag” (Plümper et al. 2005: 344). Second, the lengths of the time series (especially in the case of sectoral education spending) are not sufficient to model complex lag structures And third, it cannot be assumed that the lag structure for partisan effects is the same in the case of education spending as in the case of total public or social spending (because of things such as federalism, and so on). Therefore, I refrained from simply taking Plümper et al.’s (2005) lag structure. In sum, because of the short length of the time series and the lack of generally accepted performance indicators, attempts to optimize the lag structure would involve too much arbitrariness, in my view. Instead of exposing oneself to accusations of tweaking the data, I decided to stick with more conventional, less perfect, but also less arbitrary measures that address the problems raised by Plümper et al. (2005) at least partially.

Additional references: Beck, Nathaniel / Katz, Jonathan, 2004: Time-Series-Cross-Section Issues: Dynamics, working paper, http://as.nyu.edu/docs/IO/2576/beckkatz.pdf.

33

Descriptive Statistics A brief look at the dynamics of education spending since the 1980s (Table 1) shows why the focus of the analysis should be on public investment in higher education. The general trend in public education spending since the 1980s is one of constancy. The OECD average of total public education spending hovers slightly above 5 percent of GDP, with primary and secondary education accounting for about 3.6 percent and higher education for about 1.2 percent. In addition, former low-spenders (Portugal, Greece) have caught up, while former high-spending countries (Canada, Sweden, the Netherlands) have decreased their spending levels. The variation of public higher education spending across countries is significantly larger than the variation of spending on primary and secondary education: the variation coefficient for spending on primary and secondary education is only half as large (0.17) as the one for tertiary education (0.35).9 Therefore, it can be expected that political and institutional variables will play a greater role in explaining differences in spending on higher education.

9

Based on actual spending data from OECD Education at a Glance (2005: 185).

34

Why trade openness as indicator for economic globalization? Studies of the impact of globalization on spending have identified two basic dimensions of globalization: exposure to trade (trade openness) and unrestricted capital mobility (openness of domestic financial markets). To measure capital mobility, studies have relied on the index of financial openness provided by Quinn / Inclán (AJPS, 1997). This index in turn is based on publications of the International Monetary Fund (IMF). In the present analysis, I adopt a slightly different approach. Instead of looking at the content of regulations on capital mobility, I look at realized capital mobility, defined as net transfers from the rest of the world as a percentage of GDP (based on data from the OECD National Accounts Database). The variable “net capital transfer” (financial openness) has been tested in different model specifications. It has not proved to be a significant determinant of either type of spending. In fact, this finding is not surprising because these two variables measure very different things. As argued above, since the realization of the Single European Market and the ongoing process of liberalization of capital flows, regulative financial openness has become a common background condition for the 21 OECD countries under analysis. Differences might still occur with regard to real financial openness (measured by net capital transfers), but the empirical evidence suggests that this effect has no immediate impact on government spending.

35

Additional Robustness Checks In addition to the model specifications presented here, further robustness tests were conducted (results can be provided on request). One obvious test is to determine whether the observed effects are mainly attributable to the Scandinavian countries. I included a simple dummy variable for Scandinavian countries and an interactive term (cabinet share of social democrats interacted with Scandinavian country dummy). In the specifications without country fixed effects, both variables usually are significantly positive, as expected. Because of problems of multicollinearity, the coefficient on cabinet shares of social democrats loses statistical significance once the regression includes several interactive terms (with openness and the Scandinavian country dummy). However, in the model specifications with country fixed effects, the Scandinavian country dummy and the interactive terms lose their explanatory power completely, indicating that the fixed effects specification is able to absorb the specifics of social democratic policies in Scandinavian countries. Another robustness test I performed is the panel Jackknife test. Here, too, the coefficient estimates for social democratic government participation are stable. Furthermore, I used real public education spending per capita instead of spending as a percentage of GDP as dependent variable (see tables 2, 3, and 4 in this Appendix). Again, government participation of social democrats has a positive and significant impact on spending.

36

Additional tables: Table 1: Public and private spending on education, 21 OECD democracies Country

Public education spending, % of GDP

Year Australia Austria Belgium Canada Denmark Finland France Germany Greece Ireland Italy Japan Netherlands New Zealand Norway Portugal Spain Sweden Switzerland UK USA

1980 5.6 5.6 5.7 7.7 5.8 5.1 4.6 3.2 6.4 4.5 5.9 7.1 6.7 5.8 3.7 8.5 5.2 5.7 4.9

1991 4.7 5.4 5.4 6.7 6.1 6.1 5.4 4.0 3.4 5.5 5.0 3.7 5.6 5.5

2002 4.4 5.4 6.1 4.9 6.8 5.9 5.7 4.4 3.9 4.1 4.6 3.5 4.6 5.6

6.8 5.5 4.4 6.5 5.4 5.3 5.5

6.7 5.7 4.3 6.7 5.7 5.0 5.3

Public education spending on primary. secondary and non-tertiary post-secondary education, % of GDP 1991 2002 3.0 3.6 3.6 3.7 3.3 4.1 4.3 3.1 4.4 4.1 4.8 3.8 3.7 4.0 2.3 3.0 2.6 2.5 3.7 3.0 3.4 3.4 2.8 2.7 3.5 3.3 3.2 4.4 4.4 4.2 3.3 5.1 3.8 3.7 3.8

4.2 4.2 2.9 4.6 4.0 3.7 3.8

Public spending on tertiary education, % of GDP

Private Private spending spending on on tertiary education, education, % of % of GDP GDP

1991 1.6 1.1 0.6 2.2 1.3 1.9 0.8 1.3 0.8 1.2 0.8 0.3 1.7 1.4

2002 0.8 1.1 1.2 1.9 1.9 1.7 1.0 1.0 1.2 1.1 0.8 0.4 1.0 0.9

2002 1.5 0.3 0.2 1.3 0.3 0.1 0.5 0.9 0.2 0.3 0.4 1.2 0.4 1.2

2002 0.8 0.1 0.1 1 0.1 0.1 0.2 0.1 0 0.1 0.2 0.8 0.3 0.6

1.3 0.9 0.8 1.2 1.2 1.0 1.4

1.4 0.9 1.0 1.6 1.4 0.8 1.2

0.1 0.1 0.5 0.2 0.4 1 2.1

0.1 0.1 0.3 0.2 0.3 1.6

Sources: OECD, Education at a Glance, various years; OECD: Public Educational Expenditure, Costs and Financing: An Analysis of Trends 1970–1988, Paris, 1992, p. 84. Note: In some cases, data were available only for adjacent years.

37

Table 2: Determinants of public education spending, 21 OECD democracies, 1980–2002 Dependent variable

Specification Fixed effects Public education spending, lagged level Real public education spending per capita, lagged level Age ratio GNI per capita Female labor force participation Federalism dummy

(1) (2) (3) Public education spending (% of GDP), level

(4)

(5) Public education spending, first differences

(6) Real public education spending per capita, level

PCSE (PSAR1) No

PCSE (LDV) Yes 0.6960 (21.72)***

PCSE-ECM No –0.1545 (6.75)***

PCSE (LDV) Yes

(7) Real public education spending per capita, first differences PCSE-ECM No

0.7742 (21.57)***

–0.0690 (2.57)**

–2.9122 (4.44)*** –0.0104 (1.00) 0.0302 (5.44)*** 0.2199 (2.01)**

Veto-index (Schmidt) Trade openness Public spending (% of GDP) Gross public debt (% of GDP) Cabinet share of social democrats Cabinet share of social democrats (moving average) Interaction: SD*Veto

0.0061 (2.68)*** 0.0813 (16.88)*** –0.0054 (3.26)*** –0.0016 (2.40)**

PCSE (PSAR1) No

–3.8720 (7.30)*** 0.0100 (1.03) 0.0221 (3.19)***

PCSE (PSAR1) No

–4.1654 (7.66)*** –0.0039 (0.39) 0.0302 (5.41)*** 0.2108 (2.35)**

–0.0268 (1.15) 0.0010 (0.43) 0.0826 (16.32)*** –0.0035 (2.47)**

0.0024 (1.13) 0.0854 (18.58)*** –0.0034 (2.25)**

0.0032 (3.16)***

0.0032 (2.83)***

0.7371 (1.41) 0.0051 (0.71) –0.0083 (1.16) 0.1723 (2.17)**

–0.1590 (1.58) 0.0170 (7.16)*** –0.0031 (2.47)** 0.0079 (0.46) –0.0064 (1.68)*

–0.0050 (1.60) 0.0353 (5.18)*** –0.0075 (4.65)*** –0.0001 (0.26)

–0.0019 (2.65)*** –0.0008 (1.51) 0.0031 (2.62)*** –0.0008 (2.85)*** 0.0001 (0.71)

–0.0002 (1.86)*

38

Dummy for 90s Interaction: Dummy90s* Cabinet Share of SD Age ratio, lagged level Age ratio, change GNI per capita, lagged level GNI per capita, change Female labor force participation, lagged level Female labor force participation, change Trade openness, lagged level Trade openness, change Public spending, lagged level Public spending, change Gross public debt, lagged level Gross public debt, change Cabinet share of social democrats, lagged level Cabinet share of social democrats, change Cabinet share of social democrats (moving average), lagged level Cabinet share of social democrats, change

0.0721 (1.21) –0.0002 (0.17) –0.6123 (3.91)*** 2.7198 (2.23)** 0.0085 (2.73)*** –0.0665 (3.41)*** 0.0044 (1.80)* –0.0014 (0.16) 0.0017 (2.72)*** 0.0018 (0.33) 0.0129 (6.89)*** 0.0424 (4.83)*** –0.0018 (5.73)*** 0.0031 (1.43)

–0.0552 (1.71)* 0.2149 (0.84) 0.0051 (3.99)*** 0.0335 (7.35)*** 0.0004 (0.76) 0.0009 (0.49) 0.0002 (1.81)* –0.0001 (0.10) 0.0016 (4.25)*** 0.0060 (3.64)*** –0.0003 (4.63)*** 0.0004 (0.91) 0.0001 (2.17)** –0.0003 (1.91)*

0.0005 (1.86)* 0.0041 (4.26)***

39

Constant

1.2428 1.8213 1.2352 0.3926 0.1281 0.0439 –0.0721 (3.42)*** (3.56)*** (3.08)*** (0.82) (1.75)* (0.54) (2.73)*** R² 0.93 0.95 0.94 0.93 0.23 0.98 0.18 Observations 411 395 411 393 374 393 390 Number of countries 21 21 21 21 21 21 21 Notes: z statistics in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%; SD=social democrats; PCSE=panel-corrected standard errors (Beck / Katz 1995, 1996); PSAR1=panel-specific autoregressive process to control for serial correlation; FE=(country) fixed effects; LDV=lagged dependent variable; ECM=error-correction model; SD=Social Democrats; for variable definitions and sources, see Appendix. When comparing the models using levels of spending with the ECM models, the reader will notice a significant drop in the R². This is due to the inflation of the R² in the levels specification as a consequence of the inclusion of the lagged dependent variable. R² values of about 0.2 to 0.3 are quite common in models using first differences as dependent variable.

40

Table 3: Determinants of public spending on primary, secondary, and non-tertiary post-secondary education, 21 OECD countries, 1991–2002 (1)

Specification Country fixed effects

(4)

(5)

(6)

Public spending on primary, secondary, and non-tertiary post-secondary education, level

Public spending on primary, secondary, and non-tertiary postsecondary education, first differences

Public spending on PSNTPS education per capita, level

Public spending on PSNTPS education per capita, first differences

PCSE (PSAR1) No

PCSE-ECM No –0.2111 (2.81)***

PCSE (LDV) Yes

PCSE-ECM No

0.3723 (3.78)*** –0.1267 (1.05) 0.0272 (6.71)*** –0.0022 (0.65) 0.0534 (1.02)

–0.2189 (3.02)***

Public spending on PSNTPS education (% of GDP), lagged level

(2)

PCSE (LDV) Yes 0.3576 (2.88)***

(3)

PCSE (LDV) Yes 0.3616 (2.94)***

Public spending on PSNTPS education per capita, lagged level Age ratio GNI per capita Female labor force participation Federalism dummy

–0.8879 (1.60) 0.0041 (0.28) 0.0375 (4.03)*** –0.0250 (0.18)

0.6807 (1.17) 0.0029 (0.22) 0.0006 (0.04) 0.2397 (1.04)

Veto-index (Schmidt) Trade openness Public spending % of GDP) Gross public debt (% of GDP= Cabinet share of social democrats Interaction: SD*Veto Age ratio, lagged level Age ratio, change GNI per capita, lagged level

0.0060 (1.64) 0.0329 (3.37)*** –0.0037 (2.85)*** 0.0011 (1.44)

–0.0063 (1.07) 0.0148 (1.46) –0.0036 (1.75)* 0.0010 (2.16)**

0.7657 (1.29) 0.0038 (0.28) 0.0045 (0.28) 0.0542 (0.52) –0.0074 (1.28) 0.0203 (1.91)* –0.0044 (2.24)** 0.0029 (2.14)** –0.0004 (1.87)*

0.0031 (0.32)

0.0001 (0.07) –0.0013 (0.98) 0.0013 (0.67) –0.0003 (0.60) 0.0003 (3.39)***

–0.0670 (0.35) –1.8765 (0.75) –0.0014

–0.0218 (0.52) –0.7520 (1.36) 0.0085

41

GNI per capita, change Female labor force participation, lagged level Female labor force participation, change Trade openness, lagged level Trade openness, change Public spending, lagged level Public spending, change Gross public debt, lagged level Gross public debt, change Cabinet share of social democrats, lagged level Cabinet share of social democrats, change Cabinet share of social democrats (moving average), lagged level Cabinet share of social democrats (mov. av.), change Constant

(0.20) –0.0445 (1.39) 0.0049 (1.17) –0.0018 (0.08) –0.0004 (0.41) 0.0172 (1.76)* 0.0043 (1.83)* 0.0042 (0.25) –0.0016 (2.56)** –0.0005 (0.12)

(2.83)*** 0.0240 (3.01)*** 0.0010 (1.10) –0.0022 (0.40) –0.0002 (1.11) 0.0045 (2.11)** 0.0010 (1.82)* 0.0022 (0.54) –0.0003 (2.90)*** –0.0005 (0.54) 0.0004 (4.81)*** 0.0004 (1.90)*

0.0005 (1.08)

0.0039 (2.39)** 0.0229 1.3287 0.7974 0.4585 0.0031 –0.0787 (0.03) (1.66)* (0.70) (1.74)* (0.02) (1.31) R² 0.97 0.88 0.88 0.15 0.95 0.25 Observations 222 202 202 201 202 201 Number of countries 21 21 21 21 21 21 Notes: z statistics in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%; SD=social democrats; PCSE=panel-corrected standard errors (Beck / Katz 1995, 1996); PSAR1=panel-specific autoregressive process to control for serial correlation; FE=(country) fixed effects; LDV=lagged dependent variable; ECM=error-correction model; for variable definitions and sources, see Appendix.

42

When comparing the models using levels of spending with the ECM models, the reader will notice a significant drop in the R². This is due to the inflation of the R² in the levels specification as a consequence of the inclusion of the lagged dependent variable. R² values of about 0.2 to 0.3 are quite common in models using first differences as dependent variable.

43

Table 4: Determinants of public spending on tertiary education, 21 OECD countries, 1991–2002

Specification Country fixed effects Public spending on tertiary education (% of GDP), lagged level Public spending on tertiary education per capita, lagged level Age ratio GNI per capita Female labor force participation Federalism dummy Trade openness Public spending % of GDP) Gross public debt (% of GDP= Cabinet share of social democrats Cabinet share of social democrats (moving average) Interaction: SD*Veto Interaction: SD*Trade Openness Age ratio, lagged level

(1)

(2)

(3)

(4)

(5)

(6)

(7)

PCSE (PSAR1) No

PCSE (LDV)

PCSE (LDV)

PCSE (LDV)

PCSE-ECM

PCSE-ECM

Yes 0.1682 (1.04)

Yes 0.1432 (0.87)

PCSE (PSAR1) No

No 0.5647 (5.02)***

No –0.4611 (3.86)***

No –0.4724 (3.96)***

–1.8842 (5.73)*** 0.0185 (2.42)** 0.0184 (4.21)*** 0.0970 (1.54) 0.0043 (3.09)*** 0.0221 (3.68)*** –0.0017 (1.77)* 0.0016 (3.18)***

0.3859 (0.62) 0.0147 (1.01) –0.0262 (1.70)* 0.8387 (2.41)** 0.0159 (3.12)*** 0.0033 (0.36) –0.0018 (0.80) 0.0009 (2.25)**

0.4155 (0.74) 0.0152 (1.08) –0.0296 (2.09)** 0.9081 (2.77)*** 0.0162 (3.21)*** 0.0019 (0.21) –0.0014 (0.65)

–2.0919 (5.53)*** 0.0183 (1.97)** 0.0172 (3.97)*** 0.1815 (2.58)*** 0.0046 (3.03)*** 0.0210 (3.56)*** –0.0010 (0.78) 0.0037 (3.15)***

–0.6070 (2.19)** 0.0081 (1.44) 0.0083 (2.39)** 0.0136 (0.44) 0.0026 (2.79)*** 0.0081 (2.24)** –0.0001 (0.16) 0.0017 (1.69)*

(8) pcbild3 PCSE (LDV)

(9) dpcbild3 PCSE-ECM

Yes

No

0.2299 (1.40)

–0.4071 (3.05)***

–0.0861 (0.58) 0.0144 (3.46)*** –0.0082 (2.10)** 0.2713 (2.97)*** 0.0036 (3.39)*** –0.0008 (0.39) –0.0003 (0.53) 0.0002 (2.14)**

0.0013 (2.35)** –0.0005 (2.73)*** –0.1335 (0.41) –0.4476

–0.3848

–0.1038

44

Age ratio, change GNI per capita, lagged level GNI per capita, change Female labor force participation, lagged level Female labor force participation, change Veto-index (Schmidt) Trade openness, lagged level Trade openness, change Public spending, lagged level Public spending, change Gross public debt, lagged level Gross public debt, change Cabinet share of social democrats, lagged level Cabinet share of social democrats, change Cabinet share of social democrats (moving average), lagged level Cabinet share of social democrats, change Cabinet share of rightist parties, lagged level

(1.49) –1.3409 (0.93) 0.0128 (1.42) –0.0343 (0.83) 0.0061 (1.55)

(1.51) –1.2002 (0.89) 0.0136 (1.51) –0.0476 (1.16) 0.0051 (1.24)

(1.44) –0.4403 (1.37) 0.0083 (3.14)*** 0.0052 (0.47) 0.0009 (0.91)

–0.0033 (0.14) –0.0149 (1.51) 0.0012 (1.17) 0.0247 (2.30)** 0.0084 (2.30)** –0.0073 (0.36) –0.0007 (1.12) –0.0053 (0.94)

–0.0085 (0.35) –0.0135 (1.29) 0.0024 (2.31)** 0.0277 (2.53)** 0.0080 (2.34)** –0.0038 (0.19) –0.0015 (2.13)** –0.0034 (0.61)

–0.0016 (0.26) –0.0041 (1.56) 0.0002 (0.94) 0.0067 (2.67)*** 0.0018 (1.99)** 0.0007 (0.14) –0.0002 (1.57) –0.0013 (0.95) 0.0003 (2.66)*** 0.0004 (2.02)**

0.0008 (2.08)** 0.0010 (0.97) –0.0015 (3.54)***

45

Cabinet share of rightist parties, change Constant

–0.0018 (2.19)** –0.7532 0.8745 1.0201 –0.6112 –0.4477 –0.2474 –0.0819 0.1019 –0.1468 (2.09)** (1.44) (1.85)* (2.06)** (2.26)** (1.66)* (0.54) (0.62) (2.44)** R² 0.88 0.81 0.81 0.86 0.72 0.30 0.33 0.88 0.31 Observations 223 203 203 223 203 202 202 203 202 Number of countries 21 21 21 21 21 21 21 21 21 Notes: z statistics in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%; SD=social democrats; PCSE=panel-corrected standard errors (Beck / Katz 1995, 1996); PSAR1=panel-specific autoregressive process to control for serial correlation; FE=(country) fixed effects; LDV=lagged dependent variable; ECM=error-correction model; “rightist parties” include conservatives, Christian democrats, liberals and radical right parties; for variable definitions and sources, see Appendix. When comparing the models using levels of spending with the ECM models, the reader will notice a significant drop in the R². This is due to the inflation of the R² in the levels specification as a consequence of the inclusion of the lagged dependent variable. R² values of about 0.2 to 0.3 are quite common in models using first differences as dependent variable.

46

Table 5: Determinants of tertiary enrollment, 21 OECD countries

Time period Specification Country fixed effects Gross tertiary enrolment, lagged level Age ratio GNI per capita Female labor force participation Federalism dummy Trade openness Cabinet share of social democrats Cabinet share of social democrats (moving average) Constant R² Observations Number of countries

(1) (2) Gross tertiary enrollment, level 1980–2002 1980–2002 PCSE (PSAR1) PCSE (LDV) No Yes 0.1813 (0.96) –5.4953 (0.07) 0.8638 (0.95) –0.3036 (1.00) 12.4090 (1.49) 0.2282 (0.82) –0.0232 (0.57)

38.0477 (0.92) 0.43 464 21

32.6618 (1.21) 1.4438 (3.64)*** –0.2241 (1.73)* 4.6511 (2.18)** 0.0172 (0.11) –0.0183 (0.70)

13.5972 (1.64) 0.54 443 21

(3)

(4)

1980–1992 PCSE (LDV) Yes –0.0243 (0.09)

1993–2002 PCSE (LDV) Yes 0.7425 (5.20)***

96.4899 (0.55) 0.8431 (0.51) –0.1361 (0.50) 39.3561 (1.47) 0.5706 (1.87)*

2.2006 (0.19) 0.1793 (0.78) –0.0948 (0.69) 4.0767 (1.10) 0.0360 (0.49)

0.0427 (0.78)

0.0334 (2.58)**

–49.8548 (1.15) 0.37 252 21

15.3083 (1.61) 0.94 191 21

Notes: z statistics in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%; SD=social democrats; PCSE=panel-corrected standard errors (Beck / Katz 1995, 1996); PSAR1=panel-specific autoregressive process to control for serial correlation; FE=(country) fixed effects; LDV=lagged dependent variable; ECM=error-correction model; for variable definitions and sources, see Appendix.

47

Table 6: Between and within variation of main dependent and explanatory variables Variable | Mean Std. Dev. Min Max | Observations -----------------+--------------------------------------------+---------------Public education spending overall | 5.303777 .9916864 2.4 8.5 | N = 462 between | .8509263 3.493043 6.871304 | n = 21 within | .5541069 3.795595 7.047255 | T-bar = 22 | | Public Spending on PSNTPS education ipbao~12 overall | 3.671605 .6250259 1.7 5.3 | N = 243 between | .5835765 2.455 4.675 | n = 21 within | .262253 2.512514 4.417059 | T-bar = 11.5714 | | Public Spending on Higher Education ipbaoe~3 overall | 1.16209 .3979466 .3 2.7 | N = 244 between | .3449497 .4291667 1.809091 | n = 21 within | .2125813 .5620902 2.220423 | T-bar = 11.619 | | Cabinet Share of Social Democrats proz_sd overall | 35.45017 37.99819 0 100 | N = 483 between | 21.7648 0 74.87783 | n = 21 within | 31.4925 -39.42766 110.8045 | T = 23 | | Trade Openness openness overall | 63.64502 29.95448 15.91972 175.5566 | N = 441 between | 29.54859 20.74026 136.5217 | n = 21 within | 7.989933 38.19936 115.9443 | T = 21

48