Introducing the puzzle

At first sight, the importance of educational institutions in the distribution of skills, life chances, and ultimately income appears obvious. To many policy-makers, investing in skills and education promises to contribute to lowering socioeconomic inequality (e.g., Tony Blair's mantra of “education, education, education”). It seems very plausible to expect that giving children from low-income or otherwise disadvantaged backgrounds access to higher levels of education will contribute to a more equal distribution of skills and therefore lead to a strong compression of incomes. Indeed, investing in human capital has become a popular tool in the hands of leftist government to counter the rising trend of inequality (Boix Reference Boix1998; Busemeyer Reference Busemeyer2008, Reference Busemeyer2009b), especially in times of economic globalization, when governments cannot resort to demand-side-oriented redistributive policies as easily as they could in the postwar decades (Busemeyer Reference Busemeyer2009b).

The redistributive implications of educational institutions and investments are inherently more complex (Ansell Reference Ansell2008, Reference Ansell2010; Jensen Reference Jensen2011) than those of other social policies, however. There are several reasons for this. First, investing in human capital always creates both public and private benefits. Private benefits materialize in the form of individual wage increases as a result of educational investment. If a student obtains a degree in medicine, for example, her wages will be significantly higher than they would be without such a university degree. Public benefits are more vague and harder to pin down. A well-educated populace may be more active and more engaged in politics. Higher average levels of education might also make economies more competitive on world markets, promoting overall economic well-being.

The total effect of educational investment on social inequality depends very much on the relative distribution of public and private benefits. If private benefits outweigh public benefits, educational investment increases the income of some groups (the well-educated), but not necessarily overall economic well-being. Most importantly, if private benefits are mostly concentrated in the upper half of the income distribution, higher levels of educational investment may actually increase social inequality in terms of income and wealth. This redistributive logic lies behind the formal model developed by Ansell (Reference Ansell2008, Reference Ansell2010). When public educational investments are concentrated on an elitist higher education sector, these investments increase the private benefits of those in that sector to the detriment of those outside it. Conversely, if educational investments are more evenly distributed across educational sectors, the private benefits of educational investments are also more equally distributed. When governments invest heavily in VET instead of higher education, the private benefits of educational investments will be more concentrated in the lower half of the skills distribution, potentially contributing to a compression of incomes on the labor market.

Thus the relationship between education and socioeconomic inequality would seem to be quite straightforward. Governments that aim to limit social inequality should invest in VET or general secondary education, whereas governments that aim to maximize wage premiums for the upper half of the skills distribution should focus on higher education (which is effectively the argument of Boix Reference Boix1998). This perspective is very much rooted in political economy, however, and neglects more sociological arguments that highlight the role of education in contributing to social stratification by exacerbating class divisions and limiting intergenerational social mobility. From the latter perspective, maintaining or even expanding VET as an alternative to higher education contributes to a stratification of educational opportunities, which are very much influenced by individual social background. Large-scale provision of VET might increase the private payoffs for those in the lower half of the skills distribution, but it also diverts the same group from pursuing the more rewarding academic track. Because educational choices are very much conditioned by social background (Boudon Reference Boudon1974; Breen & Goldthorpe Reference Breen and Goldthorpe1997), children from households with a nonacademic background are then also more likely to choose a nonacademic career, which stifles intergenerational social mobility and consolidates existing patterns of social stratification. This is why in some countries, VET can be a “safety net” for those with weak academic skills, even as in other cases it can be a “diversion,” discouraging youths from nonacademic backgrounds from pursuing an academic career (Shavit & Müller Reference Shavit and Müller2000).

In sum, we end up with very different expectations/predictions about the role of VET in social inequality. On the one hand, it increases labor-market returns for the low-skilled; on the other, it helps to maintain existing patterns of occupational and social stratification. An important reason for this confusion is that the pertinent literatures in comparative political economy and welfare state research and in educational sociology have often talked past each other, because of their different conceptions of the concept of inequality itself. Scholars in educational sociology (e.g., Allmendinger Reference Allmendinger1989; Blossfeld & Shavit Reference Blossfeld and Shavit1993; Breen et al.Reference Breen, Luijkx, Müller and Pollak2009; Müller & Shavit Reference Müller, Shavit, Shavit and Müller1998; Pfeffer Reference Pfeffer2008) associate inequality with class-related biases in access to higher levels of education; that is, an unequal distribution of educational opportunities. A well-known finding from this literature is that inequality of educational opportunities is driven by institutional stratification: education systems with segmented secondary school systems and early tracking are associated with higher levels of inequality, often expressed in terms of the influence of parental background on educational attainment (Pfeffer Reference Pfeffer2008). What is more, this literature argues that inequalities in access to education are “persistent” (Blossfeld & Shavit Reference Blossfeld and Shavit1993) in the sense that the decades-long process of educational expansion has not yet eliminated class bias in access to education (Raftery & Hout Reference Raftery and Hout1993), although the situation might be different for the most recent period (Breen et al.Reference Breen, Luijkx, Müller and Pollak2009). An important insight from this literature is therefore that the continued survival of VET very much depends on, or is at least associated with, institutionally segregated and stratified secondary education systems, especially in the case of firm-based dual apprenticeships (Pfeffer Reference Pfeffer2008). What is not studied in this field of research, however, is the association between educational stratification and actual labor-market inequalities in terms of the distribution of income and wealth (an exception is Solga Reference Solga, Becker and Solga2012).

The opposite holds for literature in comparative political economy, which for the most part has neglected the influence of educational institutions on socioeconomic inequality. Instead, scholars have emphasized the importance of partisan politics and power resources (Bradley et al.Reference Bradley, Huber, Moller, Nielsen and Stephens2003; Pontusson et al.Reference Pontusson, Rueda and Way2002; Rueda Reference Rueda2008), electoral institutions (Iversen & Soskice Reference Iversen and Soskice2006, Reference Iversen and Soskice2009), and the institutional setup of the economy, in particular the centralization of collective wage bargaining (Rueda & Pontusson Reference Rueda and Pontusson2000; Wallerstein Reference Wallerstein1999). Recent contributions to the field have focused on the implications of inequality for political participation and party competition (Anderson & Beramendi Reference Anderson and Beramendi2012; Pontusson & Rueda Reference Pontusson and Rueda2010) and the structure of inequality itself (Lupu & Pontusson Reference Lupu and Pontusson2011). Educational institutions, and VET opportunities in particular, are included among the control variables in some studies (Bradley et al.Reference Bradley, Huber, Moller, Nielsen and Stephens2003; Lupu & Pontusson Reference Lupu and Pontusson2011), but usually fail to produce significant effects. The motivation for including VET as an independent variable is given in Estévez-Abe et al. (Reference Estevez-Abe, Iversen, Soskice, Hall and Soskice2001), who essentially put forward the aforementioned argument that a well-established VET regime should be associated with lower levels of socioeconomic inequality because VET opens up opportunities for the low-skilled to get access to high-quality training and subsequent employment that is well paid and secure. In countries with a focus on academic education, by contrast, educational institutions likely reinforce the polarization of skills and labor-market outcomes between the high-skilled and college-educated on the one hand and the low-skilled without any post-compulsory education on the other (Estévez-Abe et al.Reference Estevez-Abe, Iversen, Soskice, Hall and Soskice2001: 176–80). Estevez-Abe et al. (Reference Estevez-Abe, Iversen, Soskice, Hall and Soskice2001) do not look at the degree of educational stratification in terms of class-related biases in access to different kinds of education related to the provision of VET, but instead focus solely on labor-market outcomes.

In this chapter, I would like to bring the two different perspectives of political economy and educational sociology together in order to disentangle the complex relationship between educational institutions and socioeconomic inequality. The link to the overall argument of the book is to show that the politically motivated choices concerning the design of educational institutions made during the critical period of the postwar years have important consequences for the contemporary distribution of income and wealth in Western democracies. This supports the notion that education needs to be considered as an integral part of the welfare state because it does have redistributive implications. It is a different and much more challenging argument to say that policy-makers were fully aware of these distributive implications when they made their choices. To be sure, as the case studies in Chapter 2 showed, political actors had an intuitive understanding of the broad implications, or at least developed some plausible expectations in that regard. Left parties, for example, attempted to expand access and increase public involvement in the financing and provision of education with the aim of lowering inequality. But these actions also had unintended effects, such as the tradeoff between inequality and youth unemployment, which will be revealed in this chapter. Policy-makers may or may not have been aware of such tradeoffs when deciding about the institutional design of their education system. This chapter does not and cannot decide this issue. The far more modest goal here is to analyze how the choices in education policy studied in the previous chapters have had consequences for inequality, rather than to argue that policy-makers were fully aware of all potential effects at the time they made these choices.

The structure of this chapter is similar to the previous ones. I start with a description of the empirical association between educational and socioeconomic inequality and the distribution of country cases on these two dimensions. The subsequent section will develop a theoretical argument for why and how educational institutions matter for socioeconomic inequality. Because there is a dearth of high-quality comparative data on important aspects of education systems, I will present and discuss empirical evidence in the form of bivariate scatterplots and simple cross-sectional regressions. In the final section, I engage in multivariate regression analysis to further substantiate the core claims.

The relationship between educational and socioeconomic inequality

The following section contains a slightly paradoxical argument about the complex relationship between education and socioeconomic inequality: there is no direct association between educational and socioeconomic inequality, but the institutional setup of the education system does affect the distribution of income and wealth, as well as labor-market risks. The first part of this argument is substantiated in Figure 4.1, which displays the bivariate association between different measures of educational inequality and wage dispersion – that is, the distribution of market income – without taking into account the impact of taxes and transfers. This is commonly measured as the ratio of the person with an income at the 9th decile in the income distribution to the person with an income at the 1st decile (D9–D1 ratio, taken from the OECD Earnings Inequality Dataset and averaged over the period 1997–2008). We could expect educational institutions to bemostly relevant for the distribution of wages, since educational institutions shape the distribution of skills and therefore market income.¹ But as Figure 4.1 reveals, there is no strong association between different measures of educational inequality and wage dispersion. In Panel A of Figure 4.1, I use a measure of educational inequality calculated from the PISA 2009 data (OECD 2010: 34). This measure captures the impact of parental background on educational attainment in reading for 15-year-old students. Higher values indicate a stronger impact of socioeconomic background on educational performance – that is, higher levels of inequality – in the sense that educational disadvantages are transmitted from one generation to the next and the education system is less able to effectively counter existing inequalities. Panel B uses a different measure of educational stratification, also provided by the OECD (OECD 2007: 87). This measure (already used in Chapter 3) captures the difference in expectation of completing higher education (ISCED Levels 5A or 6) between a student with high socioeconomic status and one with low socioeconomic status. The first measure therefore directly captures the association between socioeconomic background and educational attainment, while in the case of the latter, student expectations might also reflect differences in the institutional setup of the education system. In countries where access to university education is more restricted, students are also less likely to expect to complete higher education, even after taking into account socioeconomic background factors. Interestingly, these two measures are only weakly correlated (0.15), indicating that educational inequality is hard to measure and pin down.

Figure 4.1 Wage dispersion and educational inequality

The most interesting finding for the purposes of this chapter is that in both cases, there is no apparent relationship with socioeconomic inequality. This means there is no evidence that educational inequality, as conventionally defined, immediately translates into socioeconomic inequality on the labor market. Countries are distributed across the whole range of the two dimensions. Looking at Panel B of Figure 4.1, however, it is possible to identify a number of by-now familiar country groupings and clusters that correspond to a certain extent to welfare state regimes. The Scandinavian countries plus the Netherlands and France are located in the lower-left quadrant, combining low levels of educational inequality with low levels of socioeconomic inequality. They can also be found in a similar cluster in Panel A, although Sweden exhibits a higher-than-expected level of educational inequality in this case.

A second cluster is formed by the continental European countries with an extensive VET system (Switzerland, Germany, Austria, and Belgium). Belgium is an outlier here because its VET is largely school-based, while the other countries have a workplace-based training system. These countries are also located quite close to each other in Panel A, joined by the outlier France. This country grouping is characterized by medium levels of socioeconomic inequality combined with above-average levels of educational inequality. Germany has been known for its extraordinarily high levels of educational inequality for a long time, a fact usually attributed to the segmented nature of its secondary school system (Pfeffer Reference Pfeffer2008). The twist in the argument that I develop below is that if Germany's segmented school system does indeed ensure the long-term sustainability of apprenticeship training, it might in the end contribute to mitigating labor-market inequalities.

The liberal Anglo-Saxon countries form a loose cluster in both figures. In these cases, above-average levels of socioeconomic inequality are combined with medium levels of educational inequality. Japan can to a certain extent be considered a member of this group (see Chapter 3, as well as Busemeyer & Nikolai Reference Busemeyer, Nikolai, Obinger, Pierson, Castles, Leibfried and Lewis2010), but it exhibits lower levels of socioeconomic inequality. The interesting characteristic of this group is that in terms of formal institutional stratification, the countries included are much more egalitarian than the continental European countries. In the Northern American countries, as well as in the United Kingdom, comprehensive schools dominate secondary education. What is more, access to higher levels of education is much less restricted in these cases. The data on socioeconomic inequality show, however, that comprehensive secondary schools and relatively open and inclusive systems of higher education are not sufficient conditions for low levels of socioeconomic inequality.

Finally, the southern European countries are clustered in the upper-left corner of Panel A, although they are more spread out in Panel B. These countries suffer from a generally low level of educational attainment (Allmendinger & Leibfried Reference Allmendinger and Leibfried2003), which might explain the low degree of educational inequality: when the majority of the population has few educational qualifications, parental background matters less (although it will probably matter a lot for the few who obtain higher education credentials). The levels of socioeconomic inequality are above average in these countries, associated with dualized labor-market structures and high levels of youth unemployment.

Explaining the complex association between skills and inequality

Should the evidence presented in the previous section be taken as proof that there is no link between the design of educational institutions and socioeconomic inequality? The answer is no, because as I will argue in this section, the association between skills and inequality is complex and the redistributive implications of educational investment depend very much upon the distribution of costs in the financing of education (our first dimension of (de-)commodification of education) and the distribution of resources across educational sectors (related to our second dimension of educational stratification). A second contribution of this chapter is its study of the implications of educational institutions for different dimensions of inequality. A broader conceptualization of labor-market stratification that looks at labor-market risk (in particular youth unemployment) in addition to wage inequality brings to light the existence of tradeoffs between these different dimensions. For instance, statist skill-formation regimes with an emphasis on school-based VET are associated with lower levels of wage inequality than are workplace-based systems, but exhibit much higher levels of youth unemployment.

As a starting point for our discussion, one can imagine a unimodal distribution of academic skills in a given age cohort. The institutions of the education system can exacerbate or mitigate the shape of the distribution as this age cohort moves through the stages of the education system. The existence of a well-established VET system lowers the relative cost of acquiring post-compulsory education for students in the lower half of the academic skills distribution (Breen & Goldthorpe Reference Breen and Goldthorpe1997; Hilmert & Jacob 2002; Stocké Reference Stocké2007). On the one hand, this might distract those in the middle segment of the skills distribution from pursuing academic higher education, which they would (have to) do in a different institutional context where VET was not a viable option; on the other, the availability of VET creates incentives for those in the lower third of the skills distribution to work hard in school and opens up pathways to well-paid and secure employment for them (Estévez-Abe et al.Reference Estevez-Abe, Iversen, Soskice, Hall and Soskice2001; Soskice Reference Soskice and Lynch1994).

In countries without a well-established VET system but with a strong focus on academic post-secondary education, those in the middle segment of the skills distribution opt for higher education (and are willing to pay a significant amount of private resources for it, as in the United States). Those at the lower tail of the skills distribution who fail to gain admission to a university or college, however, are likely to be relegated to low-paid employment. This does not by itself imply that levels of inequality are always higher in countries with a strong focus on higher education, since the total effect very much depends on the openness of access, the financing mechanisms, and the informal stratification of educational institutions within the system (Allmendinger Reference Allmendinger1989). When access to academic education is very open and/or the system is differentiated (in the sense that it provides access routes for students on the vocational track), expanding higher education might lower socioeconomic inequality. In contrast, when access is limited institutionally or de facto by strong mechanisms of stratification between educational institutions within the system in conjunction with high tuition fees, it is likely to exacerbate inequality.

This line of thought can be linked to our first dimension: the division of labor between public and private sources of funding. A core finding of the previous chapters is that the long-term balance between different political parties is a crucial explanatory factor of this variable. Furthermore, the analysis revealed a straightforward left–right cleavage in this case, with leftist parties being more in favor of public involvement in the financing and provision of education and parties of the right more opposed to state involvement. These partisan differences are reflected in the distributive consequences of the division of labor in financing education. Higher levels of state involvement are expected to be associated with a more equalized distribution of private payoffs to education, meaning less income inequality. A high share of private financing, by contrast, is likely to be associated with a more unequal distribution of payoffs. Chapter 5 will explore the microfoundations of this mechanism in terms of attitudes and preferences, showing that high levels of private financing also lower the support for redistribution, as individuals strive to recoup their significant private educational investment on the labor market.

The situation is more complex for the second dimension, educational stratification. Figure 4.1 showed that low levels of educational stratification are not a sufficient condition by themselves for low levels of social inequality. Formally comprehensive secondary school systems and high levels of tertiary enrollment can be associated with higher levels of socioeconomic inequality when the private share in financing is high (e.g., the Anglo-Saxon countries, especially North America). This is because private financing signifies a secondary layer of stratification within a formally unstratified education system. A stronger relative focus on VET holds the potential to reduce socioeconomic inequalities, because it promotes the labor-market integration of those in the lower half of the skills distribution.

This effect needs to be considered in relation to the degree of state involvement, however. In countries where extensive provision of VET is associated with a strong degree of state involvement in the financing and provision of education (the statist skill regimes), levels of socioeconomic inequality are likely to be lower. These are the countries where VET is fully integrated into the upper-secondary school system, as in Sweden. This has two consequences: first, it promotes educational mobility for students on the VET tracks, in the sense that they can easily continue their educational careers at the tertiary level, and second, it marginalizes the role of employers in the provision of initial VET. The opposite can be expected for countries where there is a strict differentiation between VET and general secondary education; that is, in countries with well-developed apprenticeship training schemes (collective skill regimes). Levels of educational stratification are high in such countries, and moving from the VET sector to higher education is still difficult in practice (Nikolai & Ebner Reference Nikolai, Ebner, Busemeyer and Trampusch2012; Powell & Solga Reference Powell and Solga2011). The strong separation between academic and vocational education nonetheless helps to maintain high levels of employer involvement in initial VET, because employers are more willing to invest in intermediate-level skills when they can be sure that employees/apprentices will stay with the firm that has trained them, instead of moving on to higher education. In sum, apprenticeship systems may be less effective than school-based VET in lowering income inequality and promoting educational mobility because they have a higher degree of private (employer) involvement.

Employer involvement is crucial, however, when it comes to the second dimension of labor-market stratification after income inequality: youth unemployment. Since unemployment rates for university graduates are generally low in most countries, the institutional setup of the VET system is an important factor in explaining differences in youth unemployment. A standard finding in the pertinent literature is that transitions from education to employment are much less frictional in countries with a well-developed apprenticeship training system (Allmendinger Reference Allmendinger1989; Breen Reference Breen2005; Gangl Reference Gangl, Müller and Gangl2003; Ebner Reference Ebner2013; Müller & Gangl Reference Gangl, Müller and Gangl2003; Wolbers Reference Wolbers2007), because apprentices there are already employed in a workplace-based setting and acquire a specific combination of general and occupational skills during the training period. School-based forms of VET are not as effective in lowering youth unemployment as workplace-based forms because their curricula do not necessarily reflect the skill demands of employers, although Scandinavian countries do try to include business representatives in the process of designing and reforming VET curricula. School-based VET systems struggle with establishing direct links between schools and employers. Liberal skill-formation regimes, which privilege academic education over VET, also lack systematic linkages between schools and employers for low-skilled youths. Because labor markets are more flexible and deregulated in these regimes, however, the barriers that youths must overcome to get access to employment are not as high as in other countries. We know this because high levels of employment protection (as well as certain characteristics of the VET system) are associated with above-average levels of youth unemployment (Breen Reference Breen2005; Wolbers Reference Wolbers2007).

Bringing together the various pieces of the puzzle, we can identify a number of interesting tradeoffs between different dimensions of inequality, rooted in the complex nature of the redistributive politics of education. Expanding state involvement in both VET and higher education might help to equalize the distribution of private payoffs on educational investment and increase the educational mobility between VET and higher education. But a high degree of state involvement also leads to the crowding out of employers in the provision of initial VET, which in turn is expected to be associated with higher levels of youth unemployment. Maintaining a high degree of employer involvement therefore entails higher levels of educational stratification, in the form of a more differentiated secondary education system that maintains the separation between vocational and academic education in order to signal to employers that investments in skill formation pay off in the long term.

The different partisan approaches and preferences studied in the previous chapters are related to the question of whether partisan actors will strive to maximize state involvement and educational mobility, or employer and other private involvement. Representatives of the political left who wish to maximize the opportunities for educational advancement by individuals in the lower half of the distribution of skills and income will support the expansion of access to higher levels of education. In the Scandinavian countries, the leftist representatives promoted the integration of VET into the general school system because they did not trust employers to provide sufficiently general skills in firm-based settings (see Chapter 2). This may have contributed to lowering overall levels of social inequality (as will be shown below), but it also led to above-average levels of youth unemployment, because school-based VET is less effective in ensuring smooth transitions from training to employment, particularly in conjunction with medium to high levels of employment protection, as in Sweden.

In the collective skill-formation regimes studied, the dominant position of the Christian democrats in cross-class coalitions ensured that the voice of business representatives remained influential, which was crucial to the survival of the dual-apprenticeship training system. Although these cross-class coalition were less universal than the Scandinavian type, the institutional characteristics of the firm-based training systems ensured smooth transitions and low levels of youth unemployment for those at the lower end of the skills distribution. Conservative welfare states such as Germany are less redistributive than their Scandinavian counterparts (Bradley et al.Reference Bradley, Huber, Moller, Nielsen and Stephens2003), but the overall level of socioeconomic inequality is still significantly below that of the Anglo-Saxon cluster. The German education system remained segmented and stratified, however, because the welfare state regime did not aim for redistribution per se, but rather for status maintenance. The ensuing high level of educational inequality did not directly spill over into socioeconomic inequality, but we can see that there is an important link between the long-term sustainability of the apprenticeship training system and the segmented nature of secondary schooling, as discussed above.

In the liberal skill-formation regimes, the tradeoff was that the expansion of access to higher education and the abolition of credible alternatives in VET led to the emergence of new mechanisms of stratification (or reinforced old ones). Ironically, comprehensive and universal education policies (“academic education for all”) are in fact associated with above-average levels of socioeconomic inequality, because the broadening of access to higher education happened alongside a process of differentiation within the higher education sector. Access to the higher strata of these differentiated systems is limited, because of either high tuition fees or other barriers of access (selectivity).

Cross-sectional evidence

In order to substantiate the claims made in the previous section, I now present empirical data in the form of bivariate scatterplots and simple cross-sectional regressions. As in the previous chapter, I rely on data from the OECD Education Statistics Database. Because I am interested in explaining broad cross-country differences, the values are all averages for the time period 1997–2008, unless otherwise stated. See the technical appendix at the end of the book for details on sources.

I start out by replicating the scatterplot provided in Estévez-Abe et al. (Reference Estevez-Abe, Iversen, Soskice, Hall and Soskice2001: 178). The share of upper-secondary students in VET programs versus general academic programs is given on the x axis of the panels in Figure 4.2. This measure does not distinguish between school-based and workplace-based (apprenticeship) types of VET. Because the data availability is much better in this case than in the case of the more detailed indicators, it is the one most commonly used in the VoC literature (Bradley et al.Reference Bradley, Huber, Moller, Nielsen and Stephens2003; Cusack et al.Reference Cusack, Iversen and Rehm2006; Estévez-Abe et al.Reference Estevez-Abe, Iversen, Soskice, Hall and Soskice2001; Iversen & Soskice Reference Iversen and Soskice2001; Lupu & Pontusson Reference Lupu and Pontusson2011). By and large, the figures confirm the expected negative association between VET and inequality. In Panel A, I display levels of inequality based on household data (the net Gini index) on the y axis, while I use a more narrow definition of wage inequality (the ratio between the person at the 9th decile and the person at the 1st decile) in Panel B. Again, we could expect the institutional setup of the education and training system to be more directly related to the distribution of wages, because educational institutions affect the distribution of skills and, in turn, incomes. The net Gini index of the distribution of household income takes into account the inequality-reducing impact of taxes and transfers, as well as the household structure. In the following section, this will be the more important measure of inequality: first, because it is more encompassing than the wage-dispersion measure, taking the household composition into account, and second, because it makes it possible to assess whether education has an inequality-reducing impact above and beyond the impact of taxes and transfers.

Figure 4.2 VET share and inequality

Figure 4.2 shows that the negative association between VET and inequality holds in both cases. Furthermore, we see a rather strong clustering of European countries in the lower-right quadrant (high levels of VET and low levels of inequality) and of Anglo-Saxon countries in the upper-left quadrant. The United Kingdom is actually much closer to the European cluster than to the Anglo-Saxon. This is an important difference from the figure in Estévez-Abe et al. (Reference Estevez-Abe, Iversen, Soskice, Hall and Soskice2001), according to which the share of VET in the United Kingdom is 10 percent, rather than the 60 percent given in the OECD data. The difficulty of classifying the complex British VET system (see Chapter 2) is obvious. The NVQ system is clearly an important component of the skill-formation regime in terms of sheer numbers, but it remains an open question whether the NVQ should be regarded as a form of VET at the same level as that in other European countries or whether it should be classified as labor-market policy. For example, Payne and Keep have stated that 75 percent of apprenticeships in the United Kingdom are at Level 2 of the NVQ framework, “which would not be recognised as an apprenticeship in countries such as Germany, Denmark and Norway” (Payne & Keep Reference Payne and Keep2011: 12). Classifying the United States is equally difficult. The OECD data do not provide any information on the share of upper-secondary students in VET in the United States. Given the institutional characteristics of the American comprehensive high school, one could easily imagine putting the United States at the same (low) level as Canada and Ireland (as was in fact done in some analyses in the previous chapter). The clustering of countries and the negative association between the incidence of VET and inequality would be even more pronounced if we were to adjust for these country-specific anomalies.

Considerable evidence exists that the provision of VET in general is negatively associated with different measures of socioeconomic inequality, but how much does the type of VET matter? In Figure 4.3, I present data on the share of students going through apprenticeship programs at the secondary level (as a percentage of all upper-secondary students). Unfortunately, this measure is only available for a limited number of countries (see discussion in Chapter 3). Even with these data limitations, however, Figure 4.3 shows that the association between apprenticeship training and inequality is much weaker than in the case of VET in general. This is not surprising, because in contrast to school-based VET, apprenticeship training is not primarily intended to maximize educational opportunities and mobility. In contrast, existing differences between occupations on the labor market are mirrored in training occupations with regard to apprenticeship pay and employment conditions during and after the completion of training.

Figure 4.3 Apprenticeship training and inequality

Turning from VET to higher education, I also find little support for a statistically significant association between enrollment in tertiary education and levels of socioeconomic inequality (Figure 4.4, Panel A). Expanding access to higher education is not a sufficient measure by which to mitigate social inequality, because investments in higher education also create significant private benefits in terms of higher wages for the high-skilled. The overall effect of expanding access to higher education therefore depends on the interaction between the higher education system and the prevailing division of labor between public and private sources of funding. Levels of socioeconomic inequality are significantly higher in countries where a significant share of spending on higher education is financed from private sources, mostly households (Figure 4.4, Panel B). The underlying causal mechanism might be related to self-interest or to differences in welfare state culture (see Chapter 5, as well as Busemeyer Reference Busemeyer2013). Individuals who have had to pay for a large share of their human capital stock out of their own pocket are less likely to support redistribution and an expansive welfare state, because they want to recoup a part of their private investment via higher wage premiums. In terms of welfare-state culture, the popular support for residual welfare states might go along with a higher willingness to pay for education. Conversely, individuals who have experienced the benefits of a universal, comprehensive, and publicly financed education system might be more likely to support redistribution and the expansion of the welfare state. Although it is hard to pin down the relative importance of these different causal mechanisms, the overall conclusion is straightforward: when it comes to levels of socioeconomic inequality, differences in the way higher education is financed matter more than enrollment levels as such.

Figure 4.4 Higher education and inequality

I have now presented evidence of a negative association between VET and socioeconomic inequality, but how is VET related to educational inequality? From the perspective of educational sociology, we could expect that well-established VET systems would go along with higher levels of educational inequality, since the segmentation of education systems into different academic and vocational tracks is an indication of institutional stratification. And indeed, looking at Panel A of Figure 4.5, which employs the same measure of educational inequality as above, we find a positive association between educational inequality and the share of students in VET. The association also holds when we use the measure of educational inequality based on the PISA study (not shown here). Furthermore, we could expect the association between apprenticeship training and educational inequality to be even more pronounced than in the case of school-based training; countries with extensive school-based VET, such as the Netherlands and some Scandinavian countries, ensure a certain degree of educational mobility by integrating vocational tracks into the general school system. In contrast, options to move from the apprenticeship track to higher education are very limited in the German-speaking countries, which have strong apprenticeship systems, although in recent years all these countries have been slowly moving towards improving the linkage between academic and vocational education (Nikolai & Ebner Reference Nikolai, Ebner, Busemeyer and Trampusch2012; Powell & Solga Reference Powell and Solga2011). The data displayed in Panel B of Figure 4.5 show that there is indeed a positive association between educational inequality and the share of apprenticeship training, although this finding needs to be interpreted cautiously because of the low number of cases. Denmark constitutes an interesting outlier: as in the German-speaking countries, apprenticeship training is the dominant form of VET at the level of secondary education. Unlike in Germany, however, youths who fail to get access to regular training in Denmark can continue their training in school-based or out-of-firm training courses, resulting in a more inclusive training system that is also better connected to the sectors of higher and continuing education (Busemeyer Reference Busemeyer2012c; Nelson Reference Nelson, Busemeyer and Trampusch2012). Excluding the outlier of Denmark (dashed line) increases the slope of the regression line.

Figure 4.5 VET and educational inequality

When we move from income inequality to youth unemployment as the dependent variable, the simple bivariate scatterplots (Figure 4.6) largely confirm previous findings on the role of VET institutions as determinants of youth unemployment. There is a negative association between the share of upper-secondary students in VET as a whole and levels of youth unemployment, but it is very weak (Panel A). The reason is that in this case, the kind of VET matters enormously. When VET is provided in the form of combined school- and workplace-based programs (apprenticeships), we find a strong negative association (Panel B). Conversely, when VET is provided mostly in schools, levels of youth unemployment are significantly higher (not shown here, since it is basically the mirror image of Panel B), because the transition from education to employment is less smooth than in the case of apprenticeship training.

Figure 4.6 VET and youth unemployment

Before I move on to multivariate cross-sectional time-series analyses in the following section, I would like to present some findings from simple cross-sectional regression analyses in order to back up the main claims. Unfortunately, the number of cases varies between fifteen and twenty-one in the regressions because of data limitations, particularly with regard to the share of apprenticeship training. What is more, the number of independent variables that can be included in this kind of analysis is necessarily limited because of the small number of cases. Despite these limitations, several findings stand out. First, educational inequality is not statistically associated with socioeconomic inequality. Even controlling for other variables, the coefficient estimate of this variable remains statistically insignificant, and its sign varies across model specifications. Second, the share of upper-secondary students in VET has a statistically significant and negative impact on socioeconomic inequality, even when controlling for educational inequality and social transfers. Based on Model 2, an increase in the enrollment share by 50 percentage points (roughly the difference between Canada and Germany) is predicted to be associated with a reduction in the Gini index of 4.5 points, which is slightly more than one standard deviation (SD = 4.3, mean = 29.5) in the sample. In contrast, there is no statistically significant association between the share of apprenticeship training and socioeconomic inequality, confirming the bivariate observation from above. Third, I also find that in the case of higher education, it matters greatly whether education is financed by public or private sources. Public spending on higher education is negatively and significantly associated with socioeconomic inequality, whereas the opposite can be observed in the case of private education spending (Models 5 and 6 versus Models 7 and 8 in Table 4.1). If we simulate a similar change in the independent variable “public higher education spending” from the level of Germany (0.97 percent of GDP for the time period under observation) to the level of Canada (1.46 percent of GDP), the predicted effect on inequality is a decrease of 4.1 points in the Gini index. Conversely, increasing private spending on higher education from the level of Denmark (0.08 percent of GDP) to that of Australia (0.74 percent) is associated with an increase in the Gini index of 3.3 points. These cases are chosen in order to make the coefficient estimates more tangible. They are not at the extreme points in the distribution of cases, but represent typical cases at the upper and lower ends of the distribution. Finally, the amount of social transfer spending does not have a particularly strong impact on socioeconomic inequality, most likely because the dependent variable is post-tax and post-transfer inequality. The effect is negative, as could be expected, but given that this variable captures the influence of traditional social insurance and assistance schemes, it could have been much stronger. In any case, these regressions show that the institutional setup of the education system matters a lot with regard to inequality, above and beyond the impact of traditional social policies.

Table 4.1 Educational institutions as determinants of socioeconomic inequality, cross-sectional regressions

Robust standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

Cross-sectional time-series analyses

I have already commented briefly in Chapter 3 on the advantages and disadvantages of cross-sectional time-series analysis in cross-country comparative research. Since data for a number of crucial independent variables are available only for the years since the mid-to-late 1990s, the time period of observation (as well as the size of the panel) is restricted. Including a lagged dependent variable, which is advised as a remedy against the problem of serial autocorrelation (Beck & Katz Reference Beck and Katz1995, Reference Beck and Katz1996), would further reduce the length of the time series. Furthermore, including a lagged dependent variable drains the explanatory power of other independent variables (Achen Reference Achen2000), especially when the time series is short and these other variables change little over time, as in our case. Because of the short time series and the limited variation of both the core independent variables and the dependent variable over time, it is also not possible to apply the more advanced error-correction model that has recently become quite popular. In short, data limitations impose certain restrictions on the methods that can be used. The panel analysis allows us to include more independent variables as controls, but in all honesty, the added value compared to simple cross-sectional regressions is limited in the present case. Nevertheless, robust findings can emerge when we combine different methods. For the panel analyses, the preferred method is to calculate panel-corrected standard errors in order to correct for panel heteroskedasticity and contemporaneous correlation across countries (Beck & Katz Reference Beck and Katz1995) and model serial correlation in the residuals with an autoregressive (AR1) process instead of including a lagged dependent variable. I refrain from using fixed effects (country dummies) because these would be correlated with institutional variables.

Table 4.2 shows the findings of an analysis of the determinants of inequality in terms of the (net) Gini index, which measures the inequality of income distribution at the household level. The control variables perform largely as expected. Economic growth is associated with lower levels of inequality (even though the coefficient estimate usually fails to reach conventional levels of statistical significance), whereas the opposite holds when unemployment increases. The centralization of collective wage bargaining is also associated with lower levels of inequality, confirming one of the central findings in the comparative political economy literature, but again, the effect is not statistically significant. Government partisanship does have an effect, but the direction is surprising. This variable is taken from the Comparative Political Data Set (Armingeon et al.Reference Armingeon, Engler, Potolidis, Gerber and Leimgruber2011) and ranges from 1 (dominance of rightist parties in government) to 5 (dominance of leftist parties). Higher values indicate more “leftwardness.” The negative sign of the coefficient estimates (in some cases statistically significant) indicates that inequality is higher under leftist governments, which runs counter to the expectations of power resources theory (Bradley et al.Reference Bradley, Huber, Moller, Nielsen and Stephens2003). This finding should not be overinterpreted, however: it could simply be picking up cross-national differences that coincide with changes in government, or it could be related to the reduced number of countries in the sample. More comprehensive tests of the power resources theory must adopt a long-term perspective (see Chapter 3).

Table 4.2 Socioeconomic inequality and VET share, 1997–2008

Standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

The most important finding here is that a higher share of upper-secondary students in VET is associated with a lower level of socioeconomic inequality. This coefficient estimate is statistically significant across all model specifications (Models 2, 4, and 6) and performs much better than the other control variables. As in the bivariate analysis, the share of apprenticeship training is not associated with levels of inequality (Models 1, 3, and 5), meaning that workplace-based types of VET are less effective at reducing inequality than VET as a whole. The negative and significant association holds even after controlling for public social spending and social transfer spending (Models 4 and 6, respectively), which themselves are expected to have a depressing effect on inequality. In contrast to the latter two variables, the magnitude (size of the effect) of the VET variable varies less across model specifications. Compared to the bivariate regressions, the magnitude of the effect of the VET variable is smaller in the case of multivariate regression due to the inclusion of additional control variables. An increase in VET enrollment of 50 percent (the same as above) is predicted to be associated with a decrease in the Gini index of 2.5 points.

Table 4.3 presents my findings on the association between the public/private division of labor in education financing and socioeconomic inequality. In Chapter 3, this variable was shown to be statistically correlated with long-term averages in the distribution of power across different party families. The findings in Table 4.3 corroborate the expectation of a significant and positive (i.e., inequality-enhancing) association between the private share of education financing and socioeconomic inequality, independent of whether I include the private share for all levels of education (Models 1 and 2) or for higher education only (Models 3 and 4). The magnitude of the effect is comparatively large (see Figure 4.7). A change of one standard deviation in the private spending share for all levels of education is predicted to be associated with an increase in inequality of 1.5 points on the Gini index (Model 1). For the private spending share in higher education only, the size of the predicted effect is an increase of 2.1 points (Model 3).² The magnitude of the effects diminishes somewhat when social transfer spending is included as a control variable (Models 2 and 4), but the coefficient estimate remains statistically significant and positive.

Figure 4.7 Predicted effect on inequality of an increase in the respective independent variable by one standard deviation

Table 4.3 Socioeconomic inequality and the public/private division of labor in education financing, 1997–2008

Standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

In Table 4.4, I analyze the association between different kinds of education spending and socioeconomic inequality. This provides a further robustness test of the association between education and socioeconomic inequality. Here I focus on education spending as a percentage of GDP, in contrast to the previous analysis, which looked at the private share of education spending and disregarded differences in spending levels. The results confirm that higher levels of public investment in any kind of education (higher education, VET, or all levels of education) are associated with lower levels of socioeconomic inequality. These effects remain highly significant even when controlling for social transfer spending (Models 2, 4, and 6 in Table 4.4), although the magnitude of the coefficient estimate is reduced, as could be expected.

Table 4.4 Education spending and socioeconomic inequality (Gini index), 1997–2008

Standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

What is the magnitude of the effects, comparatively speaking? Figure 4.7 presents the estimated effects resulting from an increase of one standard deviation in the respective variable. Clearly, social transfer spending is an important negative determinant of inequality. An increase in social transfer spending of one standard deviation is predicted to be associated with a reduction in inequality of 1.91 points. The effects of a similar increase in public investment (as percentage of GDP) in higher education and VET are quite similar: a predicted reduction in the Gini index of 1.07 and 1.04 points, respectively. The overall effect of a one-standard-deviation increase in public education spending (for all levels of education) is slightly larger (1.29 points), which could indicate that other sectors of the education system, such as primary and early childhood education, might further contribute to reducing inequality. Further research that lies beyond the scope of the present chapter would certainly be welcome here. Finally, the public/private division of labor emerges as a very strong positive predictor of levels of inequality, particularly in the case of higher education. As I said above, the predicted effect at the tertiary level of a one-standard-deviation increase in the private spending share is an increase of 2.14 points in the Gini index.

Table 4.5 looks at the interaction between the private spending share and patterns of enrollment. The theoretical expectation presented above is that higher levels of private involvement in the financing of education should negate the inequality-reducing effects of an expansion in enrollment, especially at the level of higher education. This expectation is largely confirmed by the findings. In Models 2 and 3, I look at the association between enrollment in higher education and private spending share. Model 2 employs a measure of gross tertiary enrollment supplied by UNESCO, which most likely overestimates the size of the higher education sector (see discussion in Chapter 3). Model 3 uses a more narrowly defined measure supplied by the OECD: the entry rate into tertiary type-A higher education (academic higher education). In both cases, I find a significant and negative association with levels of inequality, but this must be interpreted as the effect of enrollment on inequality when private involvement is zero, since the models include an interaction term. The positive coefficient estimate of the interaction term indicates that the negative coefficient becomes positive for high levels of private spending. The predicted positive effect of tertiary enrollment on inequality for the maximum value of the private spending share variable in the sample is 0.045 for the entry-rate measure and an even larger 0.12 for the gross enrollment measure.

Table 4.5 Interactions between private involvement and enrollment patterns

Standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

At first sight, the pattern appears to be similar in the case of VET (Model 1). This model confirms the negative association between the share of upper-secondary students in VET and inequality, but again this effect needs to be interpreted as the effect of the VET share when the private spending share is zero. In the case of Model 1, I use the private spending share for all levels of education instead of focusing on higher education. As in Models 2 and 3, the sign of the interaction term is positive and the term is statistically significant, indicating that higher levels of private involvement negate the inequality-reducing impact of VET enrollment. In contrast to higher education, however, the positive effect of the interaction is merely equal to the negative effect of enrollment. The joint impact of VET enrollment and private spending share, when the latter reaches its maximum value, is a meager 0.0028 point increase in the Gini index, which statistically speaking is indistinguishable from zero. This nicely corresponds to the finding of a nonassociation between the apprenticeship training share and socioeconomic inequality in Table 4.2, although the private spending share should not be interpreted as a direct indicator of employer involvement in initial VET. To put it simply, private involvement in the financing of education negates the inequality-reducing effect of enrollment expansion. This effect is much stronger in the case of higher education, in the sense that higher levels of enrollment are associated with lower/higher levels of inequality when the private spending share is low/high. In the case of VET, by contrast, high levels of private spending merely negate the inequality-reducing effect of VET enrollment; they do not contribute to higher inequality themselves.

In Table 4.6, I use a different dependent variable as a further check on the robustness of the findings. Until now I have looked at the Gini index as a measure of net household income inequality. The measure used in Table 4.6, by contrast, measures wage inequality, the inequality of market incomes before taxes and transfers. Besides allowing for more robustness checks, this new dependent variable makes it possible to look at the differentiated impact of educational institutions in the upper and lower half of the income distribution, respectively. Models 1 and 4 look at wage inequality across the whole distribution (ratio between incomes at the 9th and 1st deciles). Models 2 and 5 analyze the upper half of the income distribution (D9–D5 ratio), and Models 3 and 6 the lower half (D5–D1 ratio).

Table 4.6 Educational institutions and wage inequality, 1997–2008

Standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

In this setup, the negative effect of centralized wage bargaining is more robust and consistent than in the earlier analyses, which is probably related to the shift from the household to the individual level and from net to market income. Socioeconomic control variables do not perform well, probably because changes in inequality take longer to manifest and do not react to short-term changes in the business cycle. As before, there is no association between the share of apprenticeship training and wage inequality, no matter whether we look at the whole distribution of incomes or the upper and lower halves. This non-finding could also be a consequence of the low number of country cases in this specification, but it resonates well with previous findings. On the other hand, there is a statistically significant negative association between the share of upper-secondary students in VET as a whole and wage inequality. The effect is significant across all three dependent variables (Models 4, 5, and 6). Somewhat surprisingly, the magnitude of the effect is larger for the upper half of the income distribution (Model 5) than for the lower half (Model 6), although VET could well have been expected to be associated with a stronger compression of incomes in the lower half of the income distribution. But despite this counterintuitive result, the main finding still holds: a larger share of upper-secondary students in VET is negatively associated with levels of wage inequality.

I will now turn to the analysis of the determinants of youth unemployment (Table 4.7). In addition to the control variables used in previous analyses, I also include an index on the strictness of employment-protection legislation developed by the OECD (EPL index). Higher values indicate a higher level of protection against dismissals. Confirming a main finding in the pertinent literature (Breen Reference Breen2005; Wolbers Reference Wolbers2007), I find a positive association between the EPL index and levels of youth unemployment, although the effect is only statistically significant in Models 1 and 3. When employment protection is strong, employers seem to be more reluctant to hire young people, leading to more difficult transitions from education to employment and higher levels of youth unemployment. As could be expected, economic growth depresses youth unemployment, while the overall level of unemployment in a political economy is positively associated with youth unemployment. Wage-bargaining centralization also has a negative and significant effect, indicating that corporatist institutions might be better able to ensure the integration of young people into the labor market than pluralist and deregulated labor relations.

Table 4.7 Determinants of youth unemployment, 1997–2008

Standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

Coming to the main variables of interest, the findings in Model 1 confirm that there is no statistically significant association between the share of upper-secondary students in VET as a whole and youth unemployment. However, the share of apprenticeship training has a highly significant and negative effect on levels of youth unemployment (Model 2). An increase in the share of apprenticeship training by 45 percentage points (which is roughly the difference between Norway and Switzerland) is predicted to be associated with a decrease in youth unemployment of 7.9 percentage points. The average in the sample is 13.7 percent with a standard deviation of 6.8, so the magnitude of the effect is larger than one standard deviation. When apprenticeship training is included as a variable, the effect of EPL becomes insignificant. This confirms an important finding of Breen (Reference Breen2005), who has argued that a well-developed apprenticeship training system can compensate for the negative effects of strong employment regulation on youth unemployment. This also resonates well with core claims in the VoC literature on the institutional complementarity between employment protection and skill formation (Estévez-Abe et al.Reference Estevez-Abe, Iversen, Soskice, Hall and Soskice2001). Model 3 shows that public spending on education is not an effective cure against high levels of youth unemployment. In fact, the coefficient is positive and statistically significant. This finding should be treated with caution and needs to be verified by additional analyses, but it nevertheless hints at an important tradeoff: higher levels of public involvement in the financing and administration of VET might be effective in lowering levels of inequality (see above), but a strong degree of state involvement can have negative side effects in the form of higher rates of youth unemployment, because it leads to a marginalization of employers in the provision of initial VET.

Summary and conclusion

This chapter complements the previous ones by moving from politics to outcomes. Addressing a major research gap at the interstices between comparative political economy and educational sociology, I found significant evidence for the claim that educational institutions matter with regard to socioeconomic inequality. Scholarship in comparative political economy tends to underestimate the impact of educational institutions on income inequality, because it is more concerned with labor-market institutions and often fails to take into account the crucial importance of different types of education (vocational versus academic higher education), as well as differences in the financing of educational investments. Educational sociology, in turn, concentrates on assessing the impact of institutions on inequality within the education system, often neglecting the question of how educational inequalities are translated into labor-market and income inequalities.

This chapter has also revealed that the redistributive politics of education are complex and involve multiple tradeoffs. The core findings can be summarized in the form of several theses. First, public involvement in skill formation is associated with lower levels of socioeconomic inequality. A high share of private spending on education, particularly in the case of higher education, is associated with higher levels of socioeconomic inequality. In contrast, higher levels of public spending – independent of whether it is concentrated on VET or higher education – contribute to lowering income inequality. Second, inequality is also lower in countries with well-established VET systems. Workplace-based types of VET may be less effective in reducing inequality than school-based forms, however (this chapter presents correlations, not a more stringent proof of causality), for the simple reason that the link between VET and higher education is more institutionalized and better developed in countries with extensive school-based VET. Third, there is a tradeoff between lower income inequality and youth unemployment: public involvement in VET reduces income inequality, but has no effect (or maybe even a positive effect) on youth unemployment. Apprenticeship training, by contrast, does not reduce income inequality, but it has a strong negative effect on youth unemployment.

The variation of education policy preferences in OECD countries

This section offers some descriptive statistics on patterns of popular support for our two main dependent variables: public support for increased government spending on education and support for different kinds of education. Increasing public investment in education is a popular issue (Ansell (Reference Ansell2010: 136) calls it the “archetypical crowd-pleaser”). Figure 5.1 displays country averages of public support for increases in public educational spending. The data are taken from the 2006 ISSP Role of Government IV survey. Respondents were asked the following question:

Figure 5.1 Percentage share of respondents in favor of “more” or “much more” government spending on education, ISSP Role of Government IV, 2006

“Education” is listed as one of several areas where government spending could be increased. Other areas are “pensions,” “unemployment,” and “health” (care), which I will also include as dependent variables in the next section when I study the peculiarities of education compared to other social policies. The answers of respondents to these questions are coded on a scale from 1 (spend much more) to 5 (spend much less). To simplify the analysis and interpretation of results, I collapse these five categories into two: “more” and “much more” spending are recoded as 1, while the other categories are recoded as 0.

Figure 5.1 documents the high levels of public support for education spending. In almost all countries covered in the survey, the share of respondents in favor of increased spending is significantly above 50 percent. The one notable exception is Finland, which has enjoyed broad international recognition as a role model of education reforms since its top placement in the OECD PISA study (Dobbins & Martens Reference Dobbins and Martens2011). As we saw in Table 3.1, Finland is also a country with above-average levels of public spending on education: 5.9 percent of GDP on average for the period between 1997 and 2008. The level of public education spending in Japan, by contrast, is only 3.5 percent of GDP for the same period. As in Finland, Japanese popular support for increased levels of government spending on education is low. At the other end of the scale, Portugal and Spain display the highest levels of public support for increased education spending (more than 85 percent). Yet their current levels of public education spending are very different: 4.3 percent of GDP in Spain compared to 5.5 percent of GDP in Portugal (see Table 3.1). These examples show that the feedback mechanisms between the macro-level of policy-making and the micro-level of preferences are much more complex than was initially assumed. It is not automatically the case that high levels of public spending are directly associated with high levels of public support for education spending; if they were, it would indicate some kind of positive feedback mechanism. Neither can we find strong evidence for a dominating negative feedback effect; were this the case, support for more education spending would be much higher in low-spending countries such as Japan. What we can learn from this descriptive evidence, then, is that similar levels of spending (status quo) can be associated with different patterns of popular support.

Citizens vary not only in their attitudes about the level of public investment in education, but also in their preferences concerning the kind of education this investment should target. Here it is important to note that I am not concerned with studying actual educational choices, as is usually done in educational sociology (Breen & Goldthorpe Reference Breen and Goldthorpe1997; Breen & Jonsson Reference Breen2005; Hillmert & Jacob Reference Hillmert and Jacob2002; Jaeger Reference Jaeger2007, Reference Jaeger2009; Pfeffer Reference Pfeffer2008). Individuals (students and their parents) make educational choices under constraints, such as individual academic aptitude and credit limitations. Educational choices are also only relevant for a subset of the population (young persons and, to a certain extent, their parents). Instead, I am interested in what kind of education policy citizens support as a matter of policy-making. For instance, a university student might support the expansion of opportunities in VET because she just read Hall & Soskice (Reference Hall, Soskice, Hall and Soskice2001). Conversely, a construction worker close to retirement might support the expansion of university education because she believed this would be more likely to maximize her retirement income. These examples are fictitious, of course, but the general claim is that there is a difference between individual educational choices and individual education policy preferences, and that the latter have not been analyzed sufficiently in the literature.

Unfortunately, there is a shortage of survey data on policy preferences for different kinds of education. Busemeyer et al. (Reference Busemeyer and Trampusch2011) use an original dataset for the case of Switzerland, but to my knowledge the only available dataset that covers such issues for cross-national comparisons is the Eurobarometer 62.1 from the year 2004, which contains this question:

The wording of the question is far from perfect. For one, it does not distinguish between general/academic education at different levels of education. Because the question explicitly mentions that it is about post-secondary education, however, most respondents will think of higher (university) education when they hear “general or academic studies.” Another weakness is that it asks about “recommendations” for young people, not actual policy preferences. It could well be that individuals would recommend one kind of education while believing that public policy should be more concerned with the other. For the present purpose, I have to assume that there is a sufficiently close correlation between the given recommendations and actual policy preferences of respondents. I therefore recoded answers to the question by deleting spontaneous and indecisive answers from the sample, so that we are left with a dichotomous variable where 1 equals a preference for general/academic studies and 0 a preference for vocational training.

In Figure 5.2, I plot the share of respondents recommending academic education instead of vocational education to young school-leavers. As in the case of preferences for spending, the ranking of countries is somewhat counterintuitive. For example, Sweden and Germany exhibit the highest shares of citizens recommending academic education. If positive feedback effects were the only causal mechanisms at work, the support for VET should have been much higher in Germany than in Sweden (see, e.g., the case of Austria). Negative feedback effects could again be at work in the case of the United Kingdom, since in this country the support for VET over academic education is strongly above average. The opposite holds in the case of the Netherlands, which combines a well-developed VET system with strong support for VET.

Figure 5.2 Share of respondents recommending general and academic studies rather than VET to school-leavers, Eurobarometer 62.1, 2004

How and why education is different

Before I engage in a more detailed analysis of the impact of prevailing institutions on preferences, I will briefly document how and why education may be different from other social policies. Going back to the model developed by Meltzer & Richard (Reference Meltzer and Richard1981), a core claim in the political economy literature on individual preferences for redistribution is that support for redistribution declines with rising income, as rich individuals have to contribute to the welfare state via higher taxes, whereas the poor benefit from generous social benefits. The redistributive implications of investment in education are less clear-cut than those of other social policies, however (Ansell Reference Ansell2008, Reference Ansell2010; Busemeyer Reference Busemeyer2012b; Fernandez & Rogerson Reference Fernandez and Rogerson1995; Jensen Reference Jensen2011; Levy Reference Levy2005). For one, educational investment always entails both public and private benefits. Higher levels of education spending might contribute to maintaining the competitiveness of the economy and the general well-being of the population (as education is related to health outcomes), but at the individual level, human capital investment creates significant private benefits: it leads to higher wages. Thus, educational investment is associated with very different outcomes in terms of inequality, depending on which sector of the education system investments are concentrated in, the relative ease of access to higher levels of education, and the impact of labor-market institutions on education-related wage premiums (see Chapter 4).

The distinctiveness of education can also be observed at the micro-level of individual preferences, because in contrast to other social policies, an individual's income position is not a strong predictor of their attitudes towards education (Busemeyer Reference Busemeyer2012b). Table 5.1 presents the results of a regression analysis of individual-level support for increases in public education spending (the dependent variable introduced above). The most important independent variables for the present case are income (given in standardized income deciles) and educational background (years of education). Compared to the inconclusive effect of income, I expect a positive association between educational background and support for education spending. From a very narrow self-interest perspective, it could be argued that highly educated individuals have an incentive to oppose further investment in education in order to limit access to high-skilled labor markets. But highly educated individuals could also support education spending; given the ubiquitous class bias in access to education, they can expect their children to benefit from such investment. Having spent more time in educational institutions, highly educated individuals may also have been socialized into supporting this kind of social service.

Table 5.1 Individual-level determinants of preferences on education spending

Robust standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

I include a number of control variables at the micro-level besides income and education, such as age, gender, educational background, labor-market position, and partisan ideology (party ID).¹ Age (or being retired) is expected to have a negative effect, since older people will obviously not benefit from more spending on education (Busemeyer et al.Reference Busemeyer2009; Cattaneo & Wolter Reference Cattaneo and Wolter2007; Plutzer & Berkman Reference Plutzer and Berkman2005). Women are found to be more supportive of welfare state spending in general (Svallfors Reference Svallfors1997: 292), so the expected effect on spending support is positive. Being in education likely increases support for more spending on education. Labor-market outsiders could also be expected to support more spending on education, to improve their chances of getting decent employment. They might also prefer to concentrate spending on social benefits, however, especially unemployment spending. Partisanship is hypothesized to contribute above and beyond the effect of self-interest related variables (Hasenfeld & Rafferty Reference Hasenfeld and Rafferty1989). Supporters of the left should be more supportive of increased levels of public spending in general, including education (see Busemeyer et al.Reference Busemeyer and Trampusch2011). But partisanship is at least partly endogenous, since it also correlates with the other control variables in the regression. I therefore present various model specifications, including and excluding party ID.

Table 5.1 shows that on average, and without taking institutional feedback effects into account, there is no statistical association between individual income position and support for public spending on education. The effect of educational background, by contrast, is positive and remains statistically significant across all different model specifications. The control variables perform largely as expected: age or being retired has a negative impact on support for spending; being female has a positive effect. Supporters of the left are more favorable to increasing government spending on education than are supporters of the right. The latter is a nice micro-level confirmation of our central finding in Chapter 3: that parties of the left are more likely to support public involvement in education funding, whereas conservative parties favor private involvement.

In Table 5.2, I run a similar regression model on individual support for other social policies, in particular public spending on pensions (Models 1 and 2), health care (Models 3 and 4), and unemployment (Models 5 and 6). In Models 7 and 8, the dependent variable is an indicator of general support for increases in social spending, derived from a factor analysis of the other three dependent variables. As expected, the effect of labor-market position varies across social policy fields. Students are more likely to support spending on education (Table 5.1), but less likely to support spending increases on pensions (Models 1 and 2 in Table 5.2). Labor-market outsiders might not be particularly supportive of education spending, but they support more spending on pensions and unemployment benefits (Table 5.1 versus Models 1, 2, 5, and 6 in Table 5.2). The most important difference, however, is that in all these models, both the individual's income position and their educational background have a significant and negative impact on support for social spending. Compared to other social policies, increasing investment in education is a much less contested issue across class lines.

Table 5.2 The individual level determinants of preferences for social spending, ISSP Role of Government IV, 2006

Robust standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

Figure 5.3 is a graphical presentation of the findings in Tables 5.1 and 5.2 and plots predicted probabilities for individual-level support for different social policies in relation to income and educational background.² Most importantly, this figure shows that support for education spending increases among well-educated citizens (see the positive slope of the thick black line in Panel A of Figure 5.3) and is independent of differences in income (see the flat black line in Panel B). A second important insight related to the previous finding is that in general, the average public support for increasing spending on education is much higher than that for other social policies. Increasing spending on health and pensions receives more support than increasing spending on unemployment benefits, which could be related to common perceptions about the “deservingness” of benefit recipients (Van Oorschot Reference Van Oorschot2006) or could simply reflect the different sizes of welfare state constituencies. Increasing spending on education is almost as popular as increasing spending on health. But even though spending on health is at least as popular on average as increasing spending on education, because it affects essentially everyone, the figure reveals stronger class-related conflicts over health spending (see the negative effect of income and educational background on support for health-care spending in Figure 5.3). In sum, this brief analysis confirms that education is indeed different from other social policies with regard to patterns of popular support, because it is less contested across the class divide and more popular on average than other social policies.

Figure 5.3 Comparison of the impact of income and educational background on spending preferences

Institutions and education policy preferences

In this section, I will examine how educational and welfare state institutions affect popular support for increasing levels of public education spending and investment in various kinds of education. A series of multilevel regression models will be used, which combine the micro-level survey data introduced in the previous sections with the macro-level data used in preceding chapters. I first analyze individual support for increasing public spending on education, then look at the preferences for various kinds of education. As stated above, the central hypothesis is that existing institutions create feedback effects at the micro-level of attitudes. Here it is important to distinguish between the effects of macro-level variables on average levels of support for a specific policy (random-intercept models) and the mediating effects of institutions on the impact of micro-level variables (cross-level interactions or random-slope models). In addition to the coefficient estimates, the following tables contain information on the degree of cross-country variance and its statistical significance, as well as the intraclass correlation (correlation of units/individuals within groups/countries). Since partly missing data for the macro-level variables have caused the sample sizes to vary, a straightforward comparative interpretation of these indicators is somewhat futile. I therefore concentrate on the main substantive effects and present the main findings graphically.

First, existing levels of public education spending are likely to have an impact on public support for spending increases. According to the “public as thermostat” model (Wlezien Reference Wlezien1995; Soroka & Wlezien Reference Soroka and Wlezien2010), the public will express opposition to further increases in spending when a certain level is reached and demand more spending when it appears too low. Existing levels of public education spending should be negatively associated with the average level of support, so that high levels of spending lower support and vice versa. To operationalize this hypothesis, I include levels of public spending as percentage of GDP for the year 2005 (roughly one year before the field work for the ISSP survey was conducted).

Second, if educational investments have redistributive implications, prevailing levels of inequality should influence the support for spending. A statistical association between these two variables would support the claim that public attitudes on education spending are related to the general conflict over redistribution in the welfare state. As famously argued by Meltzer & Richard (Reference Meltzer and Richard1981), a mean-preserving increase in the level of inequality leads to an increase in the overall demand for redistribution, because the distance between the person with average income and that with median income will increase, as will the share of the population with below-average income. It follows that we should expect a positive association between levels of inequality (the Gini index for the year 2005) and average levels of support for education spending. I again use the net Gini index that measures inequality in household income after taxes and transfers, since I am interested in whether inequality is still associated with support for redistribution even after the impact of the transfer system has been taken into account.³

Concerning my main independent variables, the impact of educational stratification on average levels of support is less clear-cut. On the one hand, it could be argued that high levels of educational stratification will depress average levels of support for education spending, as citizens become wary of putting more public resources into an elitist system; on the other, public support for education spending might be particularly strong in stratified education systems, since the expansion of public involvement could be regarded as a means by which to overcome elitism and stratification. Rather than influencing average levels of public support, educational stratification may be more relevant as an institutional variable mediating the impact of other micro-level variables, in particular income and educational background (Ansell Reference Ansell2010; Busemeyer Reference Busemeyer2012b). When the education system is highly stratified, rich and/or well-educated individuals are more likely to support increasing spending on education. This is because they can be more certain that these investments will benefit them or their children directly: access to higher levels of education is more class-biased in stratified education systems. When institutional stratification is low, increasing spending on education will have a more redistributive impact, so that the preferences of classes are more in line with the classical redistributive struggle between the rich and the poor.

These mediating effects of institutions will be modeled empirically as a cross-level interaction effect between individual income/educational background and educational stratification as a macro-level variable. Educational stratification is operationalized in two different ways. First, I employ the OECD measure of educational stratification used and defined in Chapters 3 and 4. As a reminder, this measure captures the difference in expectations of completing academic higher education between a student with a strong socioeconomic background and one with a weak background. Second, I simply include the share of upper-secondary students in VET as an indicator of the institutional stratification of the education system at the secondary level.

The private share of education spending is also expected to have a negative impact on support for increased public spending. On the one hand, this seems obvious, since a high private spending share should be reflected in strong support for private spending if there is indeed a direct connection between preferences and output; on the other, it might be the case that individuals in countries with high levels of private spending would in fact support an expansion of public spending to complement or supplant private spending (negative feedback). Again, the mediating impact of private spending on micro-level associations might be more important than its effect on the average level. In particular, I expect a negative cross-level interaction effect between individual income and the private spending share. Rich individuals in countries with high levels of private spending should be more opposed to increases in public spending than individuals in countries with less private spending, because the former group benefits from being able to price out low-income individuals from access to high-quality private education.

Table 5.3 presents the findings from multilevel regressions using only micro-level and macro-level variables; Table 5.4 presents additional models that include cross-level interactions. The micro-level control variables perform similarly to the models in Table 5.1. In the present specification, I also include having children as a control variable,⁴ which has a strong and positive effect on individual support for education spending. Existing levels of public education spending have a negative effect on average levels of public support for education spending (Model 2), lending some support to Wlezien's (Reference Wlezien1995) “public as thermostat” argument. The effect is not robust across model specifications, however, and becomes insignificant once I include additional macro-level variables. The prevailing level of socioeconomic inequality, by contrast, has a robust, significant, and positive effect on average levels of support. This association is documented graphically in Figure 5.4. Public support for increasing public spending on education is particularly high in countries with high levels of socioeconomic inequality (the United States and the United Kingdom, but also Portugal). It is low in the Scandinavian countries, which exhibit low levels of inequality. This is evidence that political conflicts over the level of educational investment should be considered part and parcel of the more general conflict over redistribution in Western welfare states.

Figure 5.4 Levels of socioeconomic inequality and public support for more government spending on education

Table 5.3 Multilevel regression of individual support for government spending on education, ISSP 2006

Standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

Table 5.4 Multilevel regression (cross-level interactions) of individual support for government spending on education, ISSP 2006

Standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

In contrast to socioeconomic inequality, educational stratification is not associated with average levels of public support for education spending (Models 4 and 5 in Table 5.3), but I did find strong evidence for the fact that educational stratification mediates the impact of income and educational background on support for education spending. As hypothesized, the rich and well-educated are more likely to support education spending in stratified education systems (indicated by the significant coefficient estimates of the cross-level interactions in Models 1 through 4 in Table 5.4). Models 1 and 2 include a cross-level interaction between individual income and educational background on the one hand and the OECD indicator of educational stratification on the other. The same procedure is applied in Models 3 and 4, employing the broader measure of institutional stratification; that is, the share of upper-secondary students in VET. In both cases, the empirical evidence points to the fact that an increase in income at the micro-level is predicted to be associated with higher levels of support for education spending when the system is stratified. Figure 5.5 is a graphical representation of this interaction effect. It shows that the effect of income is negative in countries with a low VET share; increased individual income is associated with lower levels of support for public education spending. When the VET share increases, the income effect becomes positive (richer people are more likely to support spending increases). It is likely that the effect would be even more pronounced if we could distinguish support for different educational sectors in the dependent variable (higher education versus vocational education).

Figure 5.5 Interaction between income and VET share

Models 4 and 5 in Table 5.3 also indicate a strong and statistically robust association between the private share of education spending and the average levels of support for public spending on education. The coefficient estimate is negative, which means that average support for increasing public spending on education is lower in countries with a high private share of education spending. This evidence supports the feedback argument, in the sense that the existing institutions shape patterns of public support concurrent with the dominant development path. We also need to take into account the associations between macro-level variables themselves, however. Chapter 4 presented evidence that a high private share of spending is associated with higher levels of inequality. We now find that the private share has a significant negative impact on support for spending, whereas inequality has a positive effect. To a certain extent, these two effects cancel each other out. When we include each without the other (not shown here for reasons of space), the private spending share loses statistical significance, whereas inequality remains a highly significant macro-level predictor of support for education spending.

Finally, Model 6 in Table 5.4 reveals a negative interaction effect between individual income and the private share of education spending, but there is no significant interaction effect between educational background and private spending share (Model 5). This interaction is presented graphically in Figure 5.6. On the left-hand side, I plot the interaction between income and the private spending share for all levels of education. On the right-hand side, I use private spending for tertiary education only (this regression is not shown in Table 5.4). The income effect is positive for low levels of private spending, but it turns negative and significant for high levels of private spending. This finding could be interpreted as indicating stronger opposition among rich individuals to increasing public involvement in education systems where the public/private division of labor is already tilted in favor of private investment. Thus the private share in education spending is not only associated with higher levels of social inequality (see Chapter 4) but also increases support among the wealthy for the maintenance of such a private regime.

Figure 5.6 Interaction between income and private share of education spending

Table 5.5 presents the determinants of individual preferences for different kinds of education – our second dependent variable of interest. Remember that these are not actual educational choices, but “recommendations” by regular citizens to young school-leavers. The dependent variable is a dummy variable. A value of 1 signifies a recommendation for academic education, while 0 indicates a preference for vocational education. I include a number of common control variables at the micro-level, such as age, years of education, and gender, as well as various indicators of labor-market status. Unfortunately, the Eurobarometer dataset does not contain any information on individual income. The number of countries covered in this set is also limited, because it is restricted to European countries. Missing data on some of the macro-level variables for Greece and Luxembourg means that the number of countries at the macro-level is only thirteen for most models in Table 5.5. For all of these reasons, the findings should be interpreted with great caution.

Table 5.5 Multilevel regression of preferences for different kinds of education, Eurobarometer 62.1, 2004

Standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

The analysis reveals that highly educated individuals, as well as students and individuals in other types of education, are more likely to recommend academic than vocational education. Retired and unemployed persons, as well as those doing manual labor, are more likely to recommend vocational than academic education. Having a white-collar job has a positive but not robust effect on recommending academic education. Age (probably because of the correlation with the “being retired” dummy) and gender do not have a statistically significant effect. The joint impact of these micro-level variables is considerable. The predicted probability of a highly educated student recommending academic education is 65.9 percent, compared to 21.1 percent for a retired blue-collar worker with little education.

The impact of macro-level variables is of greater interest for the purpose of this chapter. I again include the most important macro-level variables from previous chapters, which turn out to be statistically significant determinants of individual preferences. A high share of private spending on education is associated with a lower probability of recommending academic education (Models 2, 3, and 4). This could indicate that high levels of private spending might act as a deterrent to higher levels of participation in tertiary education, although this effect seems to be more relevant at the level of perceptions and preferences than with regard to actual enrollment levels (see Chapter 3). What is more, high levels of educational stratification are associated with a higher probability of recommending academic education instead of VET. When access to higher levels of education is blocked by institutional and other hurdles, the relative value of academic education increases. It is therefore more likely for people to recommend academic over vocational education.

However, there is also evidence for a positive feedback effect: in countries with well-developed VET systems, average levels of support for academic education instead of VET are lower (Model 3). This means that people are more likely to recommend VET over academic education when it is a credible and viable alternative to academic education rather than a dead-end educational track for low achievers. Finally, there is no statistically significant association between levels of public social spending and recommendations for different kinds of education, independent of whether this is included as an individual macro-level variable (Model 5) or in combination with others (Model 4). This finding is at odds with a central thesis of the VoC literature (Estévez-Abe et al.Reference Estevez-Abe, Iversen, Soskice, Hall and Soskice2001), which is that generous welfare states serve an insurance purpose and encourage individuals to invest in vocational skills. Given the limits of the dataset, however, this finding could be a consequence of the exclusion of important non-European LMEs.

Educational institutions and redistributive preferences

The previous section looked at the effects of educational institutions on preferences for education spending and different kinds of education. In the following, I will instead focus on the implications of educational institutions for individual-level support for redistribution more generally. The purpose of this section is to demonstrate the interconnectedness of the general conflict over redistribution and the design of educational institutions. It is therefore connected to the overall topic of unearthing the linkages between education and the welfare state, at both the micro- and the macro-level.

Unlike in the previous sections, the dependent variable in the following analysis is individual support for redistribution, measured by responses to the following question in the ISSP Role of Government IV survey:

Respondents’ answers have been grouped into four categories (“definitely should be,” “probably should be,” “probably should not be,” “definitely should not be”). As before, indecisive answers were deleted from the sample and the four categories have been collapsed into two – support (1) or oppose (0) government-induced redistribution – in order to keep the analyses as simple and accessible as possible.

Table 5.6 presents the findings of multilevel regression analyses of the determinants of support for redistribution. The set of controls at the individual level is similar to the one used in Table 5.2, with one exception: here, I also include an indicator of skill specificity. As is argued in the literature (Cusack et al.Reference Cusack, Iversen and Rehm2006; Iversen Reference Iversen2005; Iversen & Soskice Reference Iversen and Soskice2001), having a set of specific vocational skills that is less easily transferable between different jobs is expected to increase individual demand for social protection in the form of redistribution. The positive and significant coefficient estimate of this variable in Table 5.6 confirms this hypothesis. The other micro-level control variables also perform as expected: high levels of income and education are associated with less support for redistribution, while being female, retired, or a labor-market outsider increases support.

Table 5.6 Multilevel regressions of preferences for redistribution, ISSP 2006

Standard errors in parentheses

^*** p < 0.01, ^** p < 0.05, ^* p < 0.1

How do macro-level variables affect support for redistribution? Starting with socioeconomic inequality, an important implication of the Meltzer–Richard model (Meltzer & Richard, Reference Meltzer and Richard1981) is that higher levels of inequality should be associated with stronger popular support for redistribution. This relationship does not seem to hold at the macro-level, since the amount of redistribution as a difference between pre- and post-tax inequality is lower in states with high levels of pre-tax inequality (what Lindert (Reference Lindert2004) and Iversen & Soskice (Reference Iversen and Soskice2009) call the “Robin Hood paradox”). Recent research by Finseraas (Reference Finseraas2009) based on multilevel regression analysis of survey data has largely confirmed the predictions of the Meltzer–Richard model, however: he found that on average, support for redistribution is higher in countries with higher levels of inequality.

The regressions in Table 5.6 confirm the existence of a positive association between existing levels of inequality and average levels of support for redistribution. The coefficient estimate of this variable is statistically significant and robust across all model specifications. However, its significance depends on the inclusion of the private spending share variable (the bivariate correlation between the two is 0.56), making the overall association between inequality and support for redistribution rather weak. This is an interesting contrast to the relatively strong association between inequality and support for education spending (see Figure 5.5). The variation around the mean is particularly large for high levels of inequality. In some countries (Portugal, Spain, and Ireland), high levels of inequality are associated with more support for redistribution, while in others (the United States, the United Kingdom, Australia, and New Zealand), the opposite is the case. Of course, one core distinction between these groups of countries is that in the former, education is largely publicly funded, whereas in the latter private education spending is more important. In other words, high levels of private education spending in most Anglo-Saxon countries may increase the general acceptance of higher levels of inequality in the population and thus lower the popular demand for redistribution.

A simple explanation for this mechanism is rooted in self-interest: as mentioned above, an individual's stock of human capital is an important determinant of redistributive preferences in terms of the total amount of human capital accumulated (income, education), the difference in kinds of skills (Iversen & Soskice Reference Iversen and Soskice2001), and the associated labor-market risks (Rehm Reference Rehm2009, 2012). To this list, one could add differences with regard to how much individuals had to pay out of pocket to acquire human capital. It could be expected that individuals who had to pay for a significant share of their human capital stock on their own would be less likely to support government-induced redistribution, since this would reduce the return they received on their educational investments. An alternative explanation might focus on the effects of welfare-state institutions more broadly defined in terms of culture (Jo Reference Jo2011; Pfau-Effinger Reference Pfau-Effinger2005; Van Oorschot et al.Reference Van Oorschot, Opielka and Pfau-Effinger2008). After health care, schools and education are probably the most visible part of the welfare state for children and middle-class parents. Before individuals become unemployed or retire, they come into contact with the education system. This contact is not ephemeral (as it may be in the case of health care), but lasts over a period of several years. Individual experiences in the education system have a long-lasting impact on the formation of personal identities, as well as expectations with regard to the role of the state: when education is largely provided and financed by the state, individuals socialized into a state-run education system are more likely to also support government involvement in the welfare state and redistribution. The concrete hypothesis that can be derived from these considerations is that a high share of private education spending should be associated with lower average levels of support for redistribution.

Indeed, the findings presented in Table 5.6 reveal a statistically significant and negative association between the private spending share in education and average levels of support for redistribution, independent of whether I include the private share for all levels of education (Models 2, 4, and 5) or tertiary education only (Model 3), where the association is expected to be more pronounced, since private spending is more important in that sector (Wolf & Zohlnhöfer Reference Wolf2009). Unlike the effect of inequality, the negative association between private spending share and support for redistribution holds when the other macro-level variables are dropped from the regression (p-value = 0.022, not shown in Table 5.6). Figure 5.7 is a graphic representation of this association. There are some outliers (Portugal, Spain, Denmark), but overall there is a negative relationship between the private spending share and average levels of support for redistribution.

Figure 5.7 Popular support for redistribution and private share of tertiary education spending

In contrast to the division of labor in education financing, educational stratification does not matter with regard to redistributive preferences (Models 4 and 5). From a theoretical perspective, the causal connection between matters of education financing and redistribution is more direct than that between institutional stratification and redistribution. This is commensurate with one of the core findings of Chapter 4: there is no direct association between educational stratification and levels of socioeconomic inequality.

Summary and conclusion

This chapter has identified a third important linkage between education and the welfare state, in addition to politics and outcomes: the effects of educational institutions on individual-level attitudes and preferences for education policies and redistribution more generally. The analysis of feedback effects complements the previous analysis by identifying popular attitudes as important micro-level foundations of path dependency. Crucial decisions about the institutional design of education and training systems in the postwar period manifest themselves as contemporary cross-national differences. These differences have implications for the continued popular support of various policy options.

The chapter started out by reaffirming the distinctiveness of education in comparison to other social policies at the micro-level. Whereas income and educational background are strong negative determinants of individual support for increasing spending on pensions, health care, and unemployment benefits, their effect on support for higher levels of government spending on education is neutral (in the case of income) or even positive (in the case of educational background). Employing multilevel regression analyses, I showed that macro-level institutions mediate the micro-level effects of income and educational background. When the education system is stratified, high-income and well-educated people are more likely to support spending increases on education, because these will benefit them or their children directly.

The analyses also revealed complex feedback mechanisms. In some cases, I found evidence for a positive feedback effect, in the sense that citizens express support for existing institutions. For example, individuals are more likely to recommend vocational education instead of academic education in countries with a well-developed VET system. Average levels of support for redistribution are lower in countries with a high share of private education spending. In other cases, however, negative feedback effects are more important. High levels of inequality are associated with increased support for redistribution. High levels of public education spending are related to lower levels of support for additional government spending. Further research beyond the scope of this chapter should try to entangle these complex feedback mechanisms by identifying conditions under which positive or negative feedback dominates.

Skill-biased technological change

In a seminal contribution, Goldin & Katz (Reference Goldin and Katz2008) identified “the race between education and technology” as the crucial factor influencing changes in inequality over time. Their core argument is that these changes in inequality (mostly in the United States, but also in other countries) can largely be explained by changes in the mismatch between the supply and demand for high-skilled employees. When the supply of high-skilled workers increases relative to demand, for example because of a larger number of university graduates, the wages of the high-skilled are reduced due to wage competition, resulting in lower levels of inequality. Conversely, when the supply of the high-skilled decreases relative to demand, their wages will increase, resulting in more inequality. Since Goldin & Katz's (Reference Goldin and Katz2008) work is based on the assumption of rational choice, individuals have an incentive to pursue higher education as long as it promises higher wages. In the long run, this should lead to a rough balance between the supply and demand of high-skilled workers. Judging from a recent increase in inequality in the United States and elsewhere, however, the balancing-out mechanism does not work as smoothly as it used to, for two reasons: first, skill-biased technological change causes the demand for high-skilled employees to increase, while demand for low-skilled work declines; second, the supply of high-skilled workers has reached an upper limit. At least in the case of the United States, according to Goldin & Katz (Reference Goldin and Katz2008), the university system cannot produce more graduates than it does already. Not everyone can be turned into a knowledge worker. As limited supply meets an increasing demand for high-skilled workers, there is a relative increase in the wage premium of the high-skilled and thus a higher level of inequality.

Elegant as it may be, there are several fundamental flaws in the story about skills and inequality told by Goldin & Katz (Reference Goldin and Katz2008), which are revealed by the findings of this book. First of all, Goldin and Katz underestimate the role of politics in shaping the institutional design of the education system; that is, the “supply side” of the labor market (Boix Reference Boix1998; Busemeyer & Iversen Reference Busemeyer2012). Even though there might be strong economic incentives to pursue a university education, the question of granting access to higher levels of education is a highly political one, as the country case studies have shown. Those currently enjoying the payoffs of higher education in the form of wage premiums have a strong incentive to maintain this privilege by limiting access to higher education, whereas those left out have an incentive to gain access. This in turn is related to the partisan conflict about opening up access to higher education, as Ansell (Reference Ansell2010) has shown. Most importantly, from the perspective of comparative political economy, cross-country differences in the balance of power between partisan and economic actors have had strong implications for the institutional design of education and training systems and for the resulting patterns of socioeconomic inequality. In other words, skill-biased technological change, or socioeconomic changes in the structure of the economy more generally, do not translate automatically into changes in inequality. Instead, socioeconomic changes interact with and are filtered by the existing institutions of the skill-formation regime (Busemeyer & Iversen Reference Busemeyer2012).

In particular, the quantitative analysis of determinants of inequality in Chapter 4 has demonstrated that the private share of education spending is an important factor in influencing the supply of high-skilled workers. Although the macro-level analysis in Chapter 4 surely needs to be complemented with micro-level studies (for an overview, see Stevens et al.Reference Stevens, Armstrong and Arum2008), there is ample evidence for large cross-national differences in levels of tertiary enrollment for countries at the same technological level, suggesting that the availability of educational opportunities is not entirely determined by some kind of natural limit defined by the innate distribution of academic skills in the population; instead, countries with a strong public commitment to expanding access to higher levels of education exhibit higher levels of participation in tertiary education than private systems. For example, in the immediate postwar years, the United States was far ahead of European countries in terms of participation and graduation rates in tertiary education (Busemeyer Reference Busemeyer2006). These days, the United States has been surpassed in this measure by a number of European countries, particularly Nordic countries such as Denmark and Finland (OECD 2012: 67). The limiting of the supply of high-skilled university graduates in the United States cannot be explained by economic factors alone, however, but has been conditioned by political and institutional factors. It is the effect of a particular institutional setup of the skill-formation regime that favors private spending relative to public commitment. This particular division of labor, moreover, is related to differences in the balance of power between partisan actors, as Chapter 3 has shown.

A second major shortcoming of the Goldin–Katz story is that it does not take into account the role of VET. The Goldin–Katz distinction between the high- and low-skilled reflects the institutional design of the US/American skill-formation regime, which draws a stark distinction between university/college graduates on the one hand and high school graduates or dropouts on the other. The category of intermediate-level skills is completely left out of the picture. A core insight of this book, following Estévez-Abe et al. (Reference Estevez-Abe, Iversen, Soskice, Hall and Soskice2001), is that well-established VET regimes at the level of upper and post-secondary education contribute to lowering inequality because they enhance the skill levels of those in the lower half of the skills distribution. As the VoC literature (Hall & Soskice Reference Hall, Soskice, Hall and Soskice2001) has argued, once a skill-formation regime is in place, firms develop comparative institutional advantages based on the availability of different kinds of human capital in a particular economy. This in turn increases the demand for employees with intermediate-level occupational skills, potentially reducing the wage gap between university and VET graduates.

Complementing the perspective of Goldin & Katz (Reference Goldin and Katz2008), Brown et al. (Reference Brown, Lauder and Ashton2011) have emphasized the contribution of economic globalization to increasing labor-market competition, not only in the low-skilled service sector but increasingly also in the high-skilled one. In the “Global Auction” depicted by Brown et al. (Reference Brown, Lauder and Ashton2011), university graduates in Western countries are now facing competition for white-collar jobs at the intermediate skill level (e.g., back-office operations, accounting, and software programming) from the newly educated in emerging economies such as India and China. Certain kinds of service-sector jobs can be outsourced to other countries more easily than other, low-skilled jobs in the service economy, especially personal services that are tied to a particular place (such as hairdressing or waiting tables). Again, institutional differences in the design of skill-formation regimes are neglected in the account of Brown et al. (Reference Brown, Lauder and Ashton2011). Countries with a strong focus on VET instead of higher education have often been regarded as being ill-equipped for the transition from the industrial to the service economy (Anderson & Hassel Reference Anderson, Hassel and Wren2013; Wren Reference Wren and Wren2013) because their skill formation regimes are rooted in the declining manufacturing sector. But if joint investments in vocational skills are indeed based on broad cross-class compromise, as is argued in this book and the VoC literature more generally (Iversen Reference Iversen2005), these joint skill investments might prevent, or at least slow down, the outsourcing of jobs at the intermediate skill level. Recent developments in the wake of the global economic and financial crisis suggest that policy-makers in LMEs are increasingly perceiving the strong dependence of the economy on services as a vulnerability instead of a comparative strength. This is exemplified by the recent (re)discovery of the value of apprenticeship training in the United Kingdom, as well as by President Obama's attempt to restore the competitiveness of the manufacturing sector in the United States.

Despite my general critique of them above, both Goldin & Katz (Reference Goldin and Katz2008) and Brown et al. (Reference Brown, Lauder and Ashton2011) make an important contribution by highlighting how structural changes in the economy that are common across all advanced democracies affect the association between skills and inequality. This book has emphasized the importance of cross-country differences, but there is a general trend across Western industrialized countries that is often captured in vaguely defined concepts such as skill-biased technological change, globalization, and liberalization. While in previous eras of closed economies and Fordist mass production the socioeconomic developments might have favored or even caused a certain compression of wage inequality, these forces seem to pull in an entirely different direction in the contemporary period. A comprehensive understanding of the complex relationship between skills and inequality should recognize both the importance of this general trend and how it interacts with country- or context-specific institutions (this is why Thelen (Reference Thelen2012) refers to different “trajectories of liberalization”).

Interpreted from a different angle, the fact that institutions matter with regard to how the ubiquitous forces of structural change are translated into policy output and outcomes suggests that there remains some leeway for political action. One European example of how policy-makers and experts try to make use of the remaining room for maneuver is the recent debate on the merits of the social investment state, which I will discuss next.

Education in the social investment state

The rise of the paradigm of the social investment state started in earnest in the late 1990s and early 2000s, when a number of seminal contributions were published by leading academics such as Giddens (Reference Giddens1998) and Esping-Andersen (Reference Esping-Andersen2002). The central promise of the social investment state, and what made it so appealing to policy-makers of the center-left such as Tony Blair in the United Kingdom and Gerhard Schröder in Germany, was to strike a new compromise between the twin goals of promoting economic growth and maintaining social solidarity in the face of increasing globalization. Unlike the radical left, centrist social democrats have always been concerned about bringing these two goals into balance; Keynesianism provided a postwar rationale for why it should be both economically sound and socially just to increase social benefits. Expanding social transfers boosted demand (and therefore growth) and at the same time contributed to redistribution. The tumultuous 1970s witnessed the decline of Keynesianism in the United Kingdom and the United States, and the rise of neoliberalism as the dominant paradigm of economic policy-making (Hall Reference Hall, Steinmo, Thelen and Longstreth1992). After the demise of Keynesianism, policy-makers from the center-left lacked a convincing and credible alternative to neoliberalism that could continue to promote the twin goals of economic growth and social solidarity and be viable in the postindustrial and globalized economy.

The social investment approach holds the promise to be this alternative paradigm. In contrast to Keynesianism, and like neoliberalism, it focuses on the supply side of the economy (Morel et al.Reference Morel, Palier, Palme, Morel, Palier and Palme2012: 5–10). The shift from the demand to the supply side is explained by the fact that efforts to stimulate economic demand with social transfers will not be effective in an open economy, since beneficiaries might easily spend the money on foreign goods and services (Boix Reference Boix1998). Globalization also reinforces skill-biased technological change, due to the comparative advantages in high-skilled labor of advanced economies and the transformation from an industrial to a postindustrial service and knowledge economy.

In this situation, human capital investment plays a critical role in minimizing potential tradeoffs; promoting educational opportunities from early childhood education, through VET and higher education, to lifelong learning – together with active labor-market policies – is at the core of the social investment paradigm (Hemerijck Reference Hemerijck, Morel, Palier and Palme2012; Morel et al.Reference Morel, Palier, Palme, Morel, Palier and Palme2012). Human capital investment might be regarded as being more effective than social transfers at stimulating growth, because investment boosts productivity levels. Expanding educational opportunities, particularly for the low-skilled, may also contribute to lowering inequality. This focus on human capital goes along with a reorientation of the function and purpose of the welfare state. Instead of compensating individuals for income loss due to unemployment or skill obsolescence ex post, the social investment paradigm envisions an activist state that prevents poverty and low-skilled employment by investing in skill formation and education as early as possible in the life course. This has been argued most forcefully by Esping-Andersen (Reference Esping-Andersen2002), who promotes a “child-centred social investment strategy,” claiming that educational and other social inequalities can be prevented most effectively by concentrating on early childhood education. Nevertheless, the investment approach is also feasible for later stages in the life cycle. The second most obvious example of a recalibration of welfare state policies in line with the social investment paradigm is the turn from passive to activist labor-market policies across the OECD world (Bonoli Reference Bonoli, Morel, Palier and Palme2012, Reference Bonoli2013). Again, the idea here is to move from a passive and ex post compensation of social risk to an activist approach, preventing the emergence of social problems by improving the skills of individuals.

The political driving forces behind the rise of the social investment paradigm are varied. As hinted at above, the initial impetus came from policy-makers of the center-left, who regarded the social investment paradigm as a promising instrument by which to appeal to new electoral constituencies in the middle classes (e.g., Schröder's Neue Mitte). Some policy instruments in the investment catalogue have always been popular with the right, however; particularly activist labor-market policies with a strong workfare component (King Reference King1995). Expanding childcare, although initially pioneered by the social democratic Nordic welfare states, has become a popular project of both the left and the right in continental European countries that were once lagging behind the Nordic countries because of the latter's increasing female labor-market participation (Flecksenstein et al. 2011; Morgan Reference Morgan, Morel, Palier and Palme2012). The social investment paradigm also enjoys considerable support from international organizations such as the OECD, as well as the European Union (Hemerijck Reference Hemerijck, Morel, Palier and Palme2012: 46).

The empirical analysis of Chapter 4 showed that increasing public investment in education can indeed contribute to lowering inequality, in line with the social investment paradigm. However, I also found that the magnitude of this effect very much depends on the kind of education that is promoted. This book has concentrated on education policy at the upper and post-secondary levels; future research needs to focus more on the contribution of early childhood education to the lowering of socioeconomic inequalities. In contrast to some of the literature (Wren Reference Wren and Wren2013) and regular OECD recommendations to expand access to higher education (OECD 2012: 13–15), I found that promoting opportunities in VET may actually be more effective in mitigating inequalities. School-based forms of VET can help limit wage inequality and promote the educational mobility of low-skilled youths, because VET integrates this group into the general secondary schooling system. Workplace-based forms of VET such as apprenticeship training, on the other hand, are linked to a stronger segmentation and stratification of secondary education, which is associated with higher levels of educational inequality. But apprenticeship training is also much more effective at creating a smooth transition from training to employment, resulting in lower levels of youth unemployment than in school-based VET systems (Gangl Reference Gangl, Müller and Gangl2003; Wolbers Reference Wolbers2007). Thus, hybrid cases such as Denmark, which combines elements of the continental European and the Scandinavian models, might be better able to resolve the tradeoffs between different dimensions of inequality than the more frequently discussed models. In the case of higher education, expanding enrollment is not a sufficient condition for lowering inequality. When a large share of spending on higher education stems from private sources, and when labor markets are flexible and deregulated, enabling the high-skilled to reap large wage premiums, expanding higher education most likely does not contribute much to lowering inequality.

In sum, this book has shown that promoting educational opportunities at various stages in the life cycle does have implications for the distribution of income and life chances, but it is certainly not a panacea. I therefore side with Allmendinger (Reference Allmendinger2009), Allmendinger & Nikolai (Reference Allmendinger and Nikolai2010), and Vandenbroucke & Vleminckx (Reference Vandenbroucke and Vleminckx2011: 451), who argue that welfare state policies should be balanced on two pillars: the new social investment pillar and a social protection pillar rooted in more traditional social insurance policies. This balancing-out of activation on the one hand with insurance and redistribution on the other increases the chances that existing welfare states will be sustainable in the long run, from both a political and an economic perspective. If welfare state reforms focus too narrowly on the investment aspect and use it to justify and sugar-coat retrenchment in other parts of the welfare state (Taylor-Gooby Reference Taylor-Gooby2008), the public could end up opposing the social investment strategy. Furthermore, there is most likely a hard core of low-skilled individuals who are extremely hard to “activate” through labor-market or education policies and who will probably continue to depend on traditional social welfare policies (Allmendinger & Nikolai Reference Allmendinger and Nikolai2010: 116). Even so, overcoming the traditional focus of many European welfare states on passive social transfers will increase the support for social policies among both employers and the (upper) middle class, whose support is crucial to maintaining the financial and political viability of welfare states.

All in all, the social investment approach holds considerable potential as a policy paradigm, when and if it is based on a balanced approach between investment and social protection. Two caveats must be added, though. First, implementing the social investment paradigm is not a free lunch – at least in the short run. Investing in childcare, VET, lifelong learning, and active labor-market policies requires additional funding if the balanced approach to welfare state restructuring is to be taken seriously. In these times of renewed austerity, coming up with additional funds for social investment will be extremely challenging, but not impossible (Diamond & Little Reference Diamond, Liddle, Morel, Palier and Palme2012). The examples of former laggards in childcare provision (e.g., Germany and the Netherlands) catching up with pioneers such as France and the Scandinavian countries show that policy change is possible, despite the constraints of globalization and austerity (Fleckenstein et al.Reference Fleckenstein, Saunders and Seeleib-Kaiser2011; Morgan Reference Morgan, Morel, Palier and Palme2012), even though it may proceed in small incremental steps.

The second caveat is more substantial: reforming welfare states along the lines of the social investment paradigm requires a significant redefinition of the function of social and education policy vis-à-vis the market. Esping-Andersen's (Reference Esping-Andersen1990) concept of de-commodification defines the function of social policy as creating a certain independence of individuals from market forces. The social investment paradigm, by contrast, implies a redefinition of this function. Instead of creating a sphere that is independent of market forces, social policy is believed to contribute to boosting economic performance, for example by maximizing employment for the formerly unemployed or by improving the skill set of low-skilled individuals. From this more critical perspective, the promotion of early childhood education is an attempt to implement the economic goal of making use of all remaining potential for employment, undermining social networks and family structures (Streeck Reference Streeck2008). It is telling that the seemingly obvious policy solution to gender discrimination on the labor market is to expand early childhood education to enable women to participate, although other possibilities, such as reducing working time for both men and women, are imaginable.

The redefinition of education and social policy as instruments for the boosting of economic productivity is gaining momentum at the international level. The expanding involvement of institutions such as the OECD and the EU is associated with the redefinition of the purpose of education from a “decommodified,” Humboldtian conception towards a more functionalist one (Walkenhorst Reference Walkenhorst2008). Both the OECD and the EU need to legitimate their involvement in economic terms: the EU in order to promote the establishment of the European Single Market and the OECD because its mission as a think tank is to promote economic cooperation and development. Thus, the involvement of intra- and international institutions in the social investment debate intensifies the refocusing of welfare state policies and institutions from de-commodification towards re-commodification.

But this criticism, however valid, should not be overestimated, especially if the social investment pillar is complemented by a second pillar focusing on redistribution and social insurance, as outlined above. Furthermore, we should recall that the Keynesian paradigm also served the dual purpose of promoting both social equality and economic development. Expanding social transfers in the name of Keynesianism may contribute to lowering inequality, but it also boosts economic demand. The Swedish welfare state, often used as a role model in the social investment debate (Morel et al.Reference Morel, Palier, Palme, Morel, Palier and Palme2012: 3–4), has always been ambiguous about the real extent of de-commodification. On the one hand, the generosity of benefits in Scandinavian welfare states is much higher than that in other countries (Esping-Andersen Reference Esping-Andersen1990; Allan & Scruggs Reference Allan and Scruggs2004); on the other, employment levels are also much higher, and unemployment levels much lower. This means that the real extent of de-commodification in terms of the labor force that does not participate in employment is actually quite limited. The granting of de-commodification on paper depends on, or at least goes along with, a high level of the commodification of labor on labor markets. In this respect, the social investment paradigm is not all that different from previous paradigms or models in trying to wed concerns about social inequality to economic functionalism. In fact, it may well be the case that in order for welfare states to be sustainable in the long run, they need the support of both the general public and significant parts of the business community.

Avenues for future research

In closing, I wish to briefly highlight some possible avenues for future research. Chapter 5 provided a first glimpse of the complexity of the feedback mechanisms between the macro-level of institutions and policies and the micro-level of attitudes and preferences. While many questions remain, the empirical evidence points to the fact that both positive and negative feedback mechanisms are at work, although it is not clear what factors decide which type of feedback will dominate in a particular case. In addition, the case studies, as well as anecdotal evidence, suggest that policy-makers sometimes care about public opinion but that in other cases organized labor-market and elite partisan interests will dominate. The saliency of a particular issue might be an explanation for the changing relative influence of public opinion and organized interests (Culpepper Reference Culpepper2010), but the obvious follow-up question is when and how issues become salient, and whether saliency can be manipulated by political actors (Baumgartner & Jones Reference Baumgartner and Jones1993). In general, I believe that more committed interaction between comparative welfare state research and political-science research on public opinion and voter participation could open up a new and fruitful research agenda on these issues.

Finally, this book has focused on post-secondary education. Recent policy developments in many European countries, however, have been mainly concerned with the expansion of early childhood education and lifelong learning. In fact, the current period might become a critical juncture for the expansion of early childhood and continuing education, just as the postwar period was for upper and post-secondary education. Despite a general trend across European and OECD countries, we can observe a large degree of cross-country variation in terms of both the rate of expansion and the particular institutions of financing and provision that are being established (Bonoli Reference Bonoli2013; Morgan Reference Morgan, Morel, Palier and Palme2012). Of particular importance is the division of labor between public and private financing and provision – a variable that I found to be of primary importance in the historical development of post-secondary education as well. It is indeed puzzling why voters in one country would be willing to accept, tolerate, and/or support fees of several thousand dollars (United States) or Swiss francs (Switzerland), when voters in another expect free and public provision of education (the Scandinavian countries). Since this is a new policy field, very little is known so far about the factors explaining variation at the macro-level and differences in public attitudes, and much remains to be done.