Partisan strategies and de-industrialization

Transcription

1 Paper prepared for the ECPR Joint Sessions workshop The Politics of Skill Formation: Institutions, Actors, and Change April 2009 in Lisbon Partisan strategies and de-industrialization Carsten Jensen Department of Political Science University of Aarhus DK-8000 Århus C Phone carstenj@ps.au.dk

2 Abstract The paper investigates the partisan strategies to rising de-industrialization in mature welfare states. It presents a theory and empirical test of how rising median voter risk exposure leads governments to increase spending on transfers, which generate immediate protection, and education, which create long-term re-skilling. Critically, however, governments will choose to expand those programs that meet the preferences of their core constituency best. Because education is much less redistributive than transfers, right-wing governments will increase spending on the former and left-wing governments will increase spending on the latter. This partisan behavior is accentuated by the fact that expansion of the overall budget is limited due to fiscal austerity and that governments therefore have to more actively prioritize one over the other. In sum, by integrating the median voter preference for insurance with the partisan preference for redistribution into a single model, the paper promises to reconcile the theory of de-industrialization with the power resource literature. 2

3 Partisan strategies and de-industrialization Welfare state provision expanded tremendously in the first decades following World War II and, as documented by a number of recent studies, de-industrialization has been one of most important drivers of this development. 1 With fewer individuals employed in agriculture and industrial production public demands rose for social protection against the risk of ending up with redundant skills, leading politicians of all party colors to expand provision (Iversen and Cusack 2000; Iversen 2001; 2005). As observed by Pierson (2001), de-industrialization is one of main reasons why welfare programs have been ballooning and, as an unintended consequence, indeed why current-day welfare states are now in permanent fiscal austerity. While the process of de-industrialization has been slowing down in recent decades it has by no means been brought to a halt. Since 1980 the loss of jobs in agriculture and industry has exceeded 8% of the total workforce (cf. Figure 1 below). Compare this yearly drop of around.4% with that from 1961 to 1980 of.5%, and one discovers that there is no reason to expect that deindustrialization has stopped playing a significant causal role in modern-day societies. Following this, the paper asks how the continuing process of de-industrialization is influencing welfare politics in mature welfare state, i.e. in countries where across-the-board expansion is no longer a viable solution to public demands. The two most obvious routes to social protection against de-industrialization are either to provide social transfers, which create immediate protection, or education, which allows for upgrading of skills on the longer run. While both transfers and education can function as a buffer 1 The paper has benefitted from the helpful comments of Michael Becker, Michael Donelly, Henning Finseraas and Jonas Pontusson. 3

4 against skills redundancy arguably a mix between the two is optimal we suggest that left- and right-wing governments will systematically prioritize one over the other: left-wing governments will be expanding transfer programs, whereas right-wing governments will be expanding public education. The diverse partisan responses to de-industrialization are caused by the different redistributive profiles of the two types of programs. As documented in a long line of studies, education is in fact among the least redistributive welfare programs. This is so not only because it is one of the most universal programs in existence, allowing access for both low and high income groups, but also because high income groups systematically use public education more than low income groups. Transfer programs, on the other hand, are generally much more targeted and, as a consequence, also more redistributive (Le Grand 1982; Gooding and Le Grand 1987; Shavit and Blossfeld 1993; Pfeffer 2008). As left- and right-wing governments are forced to make a choice between the two types of programs the former because they cannot have it all in the era of permanent fiscal austerity, the latter because they have to provide something to win the median voter each will choose the type of program that meet the preferences of their core constituency best. For the left-wing governments this will be the transfer programs and for the right-wing governments it will be education. It turns out, in short, that there by closer inspection are clear partisan effects of deindustrialization, but in somewhat unexpected ways. This is documented in a time series-cross section regression analysis on data from 17 countries between 1980 and The finding runs against common wisdom in the welfare state literature that expects left-wing parties to be associated with higher spending on both education and transfers. While this might have been true historically, we show that this no longer is the case and that the relationship is the opposite of the one normally anticipated when it comes to education. Based on this, the final part of the paper discusses the 4

5 implications for both theories of de-industrialization and power resources, i.e. the classic partisan matters argument. It is in particular highlighted how the preferences of specific parties cannot be presumed to be invariant across different programs and time as the social context changes. Appreciating this insight may help move the welfare state literature forward and reconcile those theories that are expecting that policy is driven by preferences for insurance against skill redundancy and those that expect redistributive preferences to matter more. 1. De-industrialization and welfare state growth It is a key observation in the mainstream welfare state literature that the first wave of welfare state building was initiated in the decades around 1900 as a response to the new risks generated by the industrialization that took place at that time (Stephens 1979; Korpi 1983). Although the response varied with the political parties in power, all Western countries saw substantial expansion in this period. As noted by Iversen and Cusack (2000), it is therefore not the least surprising that the even larger welfare state expansion in the post-world War II decades coincided with the encompassing process of de-industrialization, understood as the decrease in available jobs in agriculture and industry. It is, in fact, their argument that just as was the case half a century prior the structural changes of the labor market are one of the main causes of the welfare state boom. The theory of de-industrialization taps the varieties of capitalism approach in assuming that social policy preferences of the public are predominantly determined by the skills specificity of individuals. When individuals possess skills that are in risk of becoming redundant they will be more supportive of extensive welfare spending because such spending offers protection in the event of unemployment (Estevez-Abe et al. 2001; Iversen and Soskice 2001). Deindustrialization entails a continuing reduction in available jobs in agriculture and industry, which 5

6 following this logic is likely to lead to more pro-spending preferences among the public because the skills required for jobs in the growing service sector often will be very different from the skills required for the old jobs. Even in nations where the workforce is endowed with relatively general skills, like the United States or Canada, this basic skill gab is likely to be of crucial importance as noted by Iversen and Cusack (2000, pp ): Whereas skills within agriculture, manufacturing, or services are typically transferable to some degree, most skills acquired in either manufacturing or agriculture travel very poorly to services occupations. Even low-skilled blue-color workers find it exceedingly hard to adjust to similarly low-skilled service sector jobs because they lack something that is vaguely referred to as social skills. The primary element of the theory of de-industrialization is what it dubs the joint structural effect, i.e., the common economic shock hitting the labor market as a consequence of de-industrialization and which raises the risk exposure of the median voter. Because both left- and right-wing parties are forced to capture this voter, both parties will be advocating increased welfare spending. The key expectation of the theory therefore is that both left- and right-wing governments will be associated with welfare expansion a proposition that sets it apart from the conventional power resource theory (Stephens 1979; Korpi 1983). Normally measured as either total government spending or government transfers, it has been documented empirically that de-industrialization from the 1960s to the early 1990s had a direct effect, unmediated by partisan governments, on the size of Western welfare states (Iversen and Cusack 2000; Iversen 2001; 2005). The major contribution of the theory of de-industrialization undoubtedly is the highlighting of the joint structural effect. However, adding to this, it is observed that left-wing 6

7 parties will often try to ensure that the rising spending in particular will be allocated to programs that have strong redistributive effects. This expectation resonates with the power resource theory that posits that the fundamental preference of left-wing parties is to promote economic redistribution because the core constituency of these parties is low income groups. Right-wing parties are, conversely, trying to promote programs that entail as little redistribution as possible as their core constituency is high income groups (Stephens 1979; Korpi 1983; Bradley et al. 2003). According to the theory of de-industrialization, these differences are likely to lead left-wing parties to favor government consumption. Yet, while the argument that left-wing parties will try to maximize redistribution is highly valid, the expectation that this would imply a preference for consumption over transfers is less well-founded. Above all, as we will discuss below, transfer programs are in general more redistributive than consumption and, in particular, education. So although left-wing governments historically may have been associated with higher government consumption, especially of elderly care and childcare, this is hardly a result of wanting more economic redistribution. Much more likely, it is the outcome of a preference for defamilisation, i.e., the reduction of dependence of individuals, notably females, on the traditional family structure and the explicit opposition to this policy by right-wing parties (Esping-Andersen 1999). There is also a certain counter-intuitiveness about the idea that de-industrialization will lead left-wing governments to expand consumption because these new service jobs (e.g., doctors and nurses) will be of the exact kind that cannot be managed by unemployed farmers and industry workers. At least such a notion entails that the government first of all provides re-skilling via education the exact point we will be making below. 7

8 2. Partisan strategies for coping with de-industrialization under fiscal austerity The existing work on de-industrialization has so far overwhelmingly focused on the period of welfare state expansion. Even though the empirical studies have included data until the early 1990s, the studies have never decomposed the time period so any period effect is likely to have been washed out. While, as can be seen from Figure 1, this is not a problem due to any exhausting of deindustrialization, which continues to unfold, albeit at a slightly slower pace and with less variation between the nations, it is curious nevertheless because the economic environment has altered dramatically. FIGURE 1 ABOUT HERE As pointed out by Pierson (2001), the past three decades has been an era of permanent austerity where the previous policy of across-the-board spending as a solution to social problems is no longer feasible. Paradoxically, part of the fiscal hardships has come about because of previous decades willingness by governments of all party colors to respond to de-industrialization by letting the state step in as a buffer. As a consequence of this past generosity, governments are now in a situation where it becomes much more important to prioritize between the various programs. In this section we start out by discussing the two most obvious solutions to rising de-industrialization in modern societies, namely transfer programs and education. Following this, the partisan strategies are outlined. 8

9 2.1. Policy solutions to de-industrialization As emphasized by the previous work on de-industrialization, the major response to rising median voter risk exposure by governments of all party colors has been to expand in particular transfer programs (Iversen and Cusack 2000; Iversen 2001; 2005). Expansion of these programs intuitively makes a great deal of sense when trying to protect the population against loss of skills because the provision of, for instance, unemployment insurance and old-age pensions generates immediate security for the individual. Essentially, of course, transfer programs seldom provide a long-term solution to the problem of skill redundancy because it does not alter the skills themselves. There only exists one type of welfare program that has the capacity to ensure such large-scale changes in the skills distribution of society, namely education. Education is by its very nature aimed at changing the skills of individuals, which sets it apart from all other government programs. 2 To workers in precarious positions on the labor market, education may allow obtaining skills that make shifting from industrial to service production feasible. Extensive public education may also allow children of workers with either low or very industry-specific skills to move to different segments of the labor market than their parents (Ansell 2008). Thus, assuming that workers are both interested in securing immediate protection as well as more long-term social mobility of themselves and their children, it is arguably to be expected that the optimal protection against de-industrialization entails a mix between transfer programs and education. While both transfer programs and education may serve the same insurance function, they have very different redistributive profiles. Given that both are paid for via progressive taxes, 2 With the exception of active labor market policy, but this, however, is, partly, a much more recent type of policy only gaining wide popularity in the 1990s and, partly, of a much smaller magnitude than public education. 9

10 the main distinction stems from how targeted the benefits are. Education is characterized by being one of the most universal welfare programs in the Western world with public services, often free of charge, provided to all quarters of society (Huber and Stephens 2000). Compared to this, almost all transfer programs are considerably more targeted at individuals with low income; even the archtypical universalist old-age pensions of Scandinavia are in fact to a large extent means-tested (Immergut et al. 2007). The fact that the general need for education is related to the lifecycle whereas the need for transfer programs mostly is related to the economic cycle underlines the difference: those individuals most at risk of losing their jobs, i.e. low income individuals, are also the ones most likely to enjoy the protection of core transfer programs (Cusack et al. 2006). Consequently, as noted by Goodin and Le Grand (1987, p. 215), in egalitarian terms [ ] the beneficial involvement of the non-poor in the welfare state is not merely wasteful it is actually counterproductive. The more the non-poor benefit, the less redistributive (or, hence, egalitarian) the impact of the welfare state will be. When it comes to education this basic mechanism is enhanced by the fact that high income groups not only use this service as much as low income groups, but in fact use is even more. This point is forcefully made by Le Grand (1982) in a study of the net benefits received by different income groups from the state, as well as a series of studies of the impact of education on social mobility. The reason is fundamentally that a full utilization of public education demands a large amount of cultural capital, which is something that disproportionately favors members of the high income group (Shavit and Blossfeld 1993; Pfeffer 2008). Although the argument that targeted programs are more redistributive than nontargeted is well-rehearsed, some authors argue that the relationship is in fact the reverse. Korpi and Palme (1998) have introduced the notion of a paradox of redistribution where the inclusion of the middle income groups has created more support for welfare provision historically. This might 10

11 explain why universal welfare states like the Scandinavian have bigger budgets and generate more redistribution than the residual welfare states in the Anglo-Saxon world. Larsen (2008) has recently corroborated the argument by showing that the attitudes towards the poor in universal welfare states are considerably more benevolent than in other types of welfare states. As it is, the two positions do not need to be mutually exclusive. The paradox of redistribution is characterized by, first of all, being a macro-level phenomenon. That is, it concerns the entire set-up of the welfare state rather than individual programs. In Scandinavian welfare states one does on average find less targeted programs as well as more redistribution, but that does not tell us that much about the effect of individual programs and the partisan strategies related to them. The argument does, secondly, not presume that left-wing governments have actively pursued universal programs. In fact, as documented by Esping-Andersen (1985), the universal welfare state type emerged because Scandinavian left-wing parties had to compromise with other political parties to get any programs enacted. Yet, their preferred programs where often targeted because that would benefit their core constituency most. It is, in short, entirely consistent to assume that left-wing governments will be focusing on targeted programs and still find the pattern documented by Korpi and Palme (1998) and Larsen (2008) Partisan strategies Following the theory of de-industrialization, political parties are expected to try capturing the median voter. As de-industrialization creates a joint structural shock on the labor market, increasing the risk exposure of the median voter, both left- and right-wing governments will respond by expanding spending. As just outlined, there are basically two policy solutions that make sense in order to meet the new social risks, namely to provide more transfers or more education. Both are 11

12 arguably valid ways of trying to meet public demands. While no modern countries ever would rely solely on either one of the two, the different redistributive profiles of the programs means that leftand right-wing governments are expected to favor different programs: To left-wing governments transfer programs will be the first choice-policy and education the second choice; to right-wing governments education will be the first choice-policy and transfer the second choice. Figure 2 illustrates the intuition behind the partisan strategies. Line a in Panel A shows the risk distribution at the initial situation. As assumed in the literature, and documented empirically by Cusack et al (2006), there is a negative relationship between risk exposure and income, with low income groups being more at risk of ending up with redundant skills than high income groups. As the joint structural effect of de-industrialization sets in, the risk exposure is raised for all groups to line b (currently we assume that the effect is linear). This increases the risk exposure of the median voter, which triggers the policy changes studied in Panel B. Line c in Panel B is the budget lines of the government in the initial situation. Line d is the total budget constraint imposed. That is, it is not possible to spend more money than that represented by budget line d. On the vertical axis in Panel B the spending on first choice-policy is outlined and on the horizontal axis the second choice-policy is outlined. FIGURE 2 ABOUT HERE Looking at the response of the partisan governments, we see, first, how the joint structural effect forces an expansion of the entire budget line c outwards, ultimately till it meets the budget constraint at line d. Second, as the budget line is pushed outward there will be a tendency to let the expansion benefit the first choice-policy. This entails that left-wing governments will prioritize 12

13 expansion of transfer programs, while right-governments will prioritize expansion of education. This leads to the first and most important hypothesis of the paper. H1: De-industrialization will lead right-wing governments to expand educational spending and leftwing governments to expand transfer spending. In Panel B we assume for reasons of simplicity that there is room for expansion that can fully meet the rising median voter risks, entailing that there is no trade-off between the first and second choice policies: De-industrialization leads to expansion of the first choice policy, but does not influence the provision of the second choice policy. Yet, given the fiscal austerity that many nations has found themselves in since the late 1970s, this seem like a heroic assumption. More likely we may expect there to be a certain trade-off between the two policy solutions as politicians are actively discounting one on account of the other to make ends meet. That is, as de-industrialization leads right-wing governments to expand educational spending it simultaneously lead them to curtail spending on the transfer component and vice versa for the left-wing governments. In the extreme situation where it is impossible to expand further, i.e., the budget line of the government is equal to the budget constraint, there is only room for meeting public demand by shifting spending from one policy area to the other. 3 FIGURE 3 ABOUT HERE 3 Needless to say, governments may choose to curtail spending on third programs, but it lies beyond the scope of this paper to deal with this much more encompassing issue. The logic of the argument holds as long as at least some tradeoff takes place between transfers and education. Given the sheer size of these two programs this is not a particularly farfetched assumption. 13

14 It is well-known that left-wing governments are much more common in some countries than in other ones, presumably due to the highly rigid electoral systems with proportional representation systematically biasing government formation towards the left-wing and Majoritarian systems biasing towards the right-wing (Iversen and Soskice 2006). It is also well-established that long spells of left-wing government in turn lead to a much larger portion of the GDP being allocated to social welfare programs in the postwar decades, all else equal reducing the room for further expansion (Huber and Stephens 2001). Figure 3 shows the relationship between the cumulative share of left-wing governments since 1946 and total social expenditure defined as all cash benefits and benefits in-kind including education between 1980 and The bivariate correlation is.68. This suggests that left-wing governments on average might be more inhibited than right-wing governments in terms of expanding overall spending because the initial budget line c is closer to the budget constraint d than it is for right-wing governments. The de-industrialization-induced trade-off between transfers and education is, in sum, likely to accentuate the partisan strategies of both leftand right-wing governments. Yet, because left-wing governments on average preside over bigger initial budgets than right-wing governments the trade-off ought to be more severe for the left-wing. This is the second hypothesis. H2: The trade-off between transfers and education will enhance the partisan strategies. It will do so especially for left-wing governments. Figure 2 outlines the basic logic of partisan strategies, but assumes that the joint structural effect is linear. In reality, however, this is often not the case because de-industrialization will change the shape of the risk distribution making those already exposed most even worse off (Iversen 2001). Consequently, at relatively low levels of de-industrialization, the median voter has not yet been 14

15 affected, entailing that right-wing governments are not forced to act and left-wing governments are not required to prioritize all that heavily. This logic is illustrated in Panel A with the shaded line e where the only movement takes place to the left of the median voter. Essentially this means that the divergent partisan strategies should become more pronounced as de-industrialization unfolds and gradually comes to affect the voters holding the median position. This is the third hypothesis. H3: The partisan strategies will be accentuated by de-industrialization. Figure 1 above showed the development in de-industrialization. It is evident from the figure that deindustrialization has been unfolding in both the 1980s and 1990s, but Figure 1 also makes it clear that that there has been convergence between the nations. In 1961 the standard deviation was 7.2, dropping to 5.8 in 1980, 4.9 in 1990 and 3 in Especially the 1990s saw a marked convergence as the laggards caught up with the front-runners since there seem to be a ceiling effect at c. 85% de-industrialization that no countries pass. This is important because rising deindustrialization is the motor of the argument presented here. As fewer countries actually experience rising de-industrialization, the argument must be expected to lose explanatory power when applied to the entirety of the Western world. This is the fourth and final hypothesis. H4: The effect of the partisan strategies is diminished over time. 4 Part of this is driven by Australia that reports very low levels of de-industrialization until 1996, but the point is valid even when disregarding Australia. The standard deviations are then 7.4 (1961), 4.8 (1980), 4.1 (1990) and 3.1 (2000). In the statistical tests we control for the influence that this outlier might have without altering the findings. Figure 1 above did not include the data on Australia. 15

16 The expectation of the divergent partisan responses to de-industrialization is new to the welfare state literature. The mainstream literature either equates partisan effects with left-wing governments leading to more welfare across-the-board (Korpi and Palme 2003; Allan and Scruggs 2004), or anticipates that institutional factors have overtaken as the main driver of provision (Pierson 1994; Huber and Stephens 2001). In the relatively few studies in the comparative political economy literature on education the overwhelming expectation is that left-wing governments will expand spending compared to right-wing governments, i.e. the very opposite expectation we have (Boix 1998; Iversen and Stephens 2008; Busemeyer 2009). None of these studies do, however, to our knowledge include measures of de-industrialization, or look at how political decisions regarding education is related to other policy programs. 3. Methodology In order to test the argument of the paper statistical evidence is presented on public spending between 1980 and 2000 for 17 Western countries. 5 We disaggregate the period into the 1980s and 1990s to see if as expected there are period effects. Spending is measured as a percentage of the GDP as is customary in the comparative political economy literature. De-industrialization is measured following Iversen and Cusack (2000) as 100 minus the sum of manufacturing and agricultural employment as a percentage of the working-age population. Partisan government is measured as the share of left parties in government and is taken from Huber et al. (2004), which is one of the most well-established operationalisations in the literature. Given our argument, which assumes that the government is led either by a left- or right-wing party the measure has the 5 Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Sweden, the United Kingdom and the United States. 16

17 advantage that any significant negative coefficients are testament to a positive association between right-wing government and spending (the same logic as in Iversen 2005). To measure how deindustrialization leads partisan governments to pursue different policies, we rely on the interaction term between the partisan government and de-industrialization variables. The interaction term between partisan government and de-industrialization measures the partisan strategies directly. Yet, this does not tell us whether different governments are facing a trade-off between the two policy solutions of transfers and education as de-industrialization rises. We hypothesized that the partisan strategies would be accentuated when taking the trade-off into account because left- and right-wing government not only is distinguished by the expansion of the first choice policy, but also by the curtailing of the second choice policy. In the event that no or only limited expansion is possible, it becomes particularly important to include a measure for the trade-off because the increase in the first choice policy is then directly dependent on the decrease in the second choice policy. To capture the trade-off, an interaction term between de-industrialization and the two alternative policy solutions is included as well. That is, when estimating the determinants of transfers the interaction between de-industrialization and education is included, and when estimating the determinants of education the interaction between de-industrialization and transfers is included. This is a simple, but in our opinion effective way of capturing the deindustrialization-induced need for making trade-offs. Apart from these variables, controls have been included to capture effects from the constitutional structure and various socio-economic and demographic factors. We have deliberately included several measures that capture fiscal austerity at the macro-level, notably net government liabilities and net government interest payment. This may at first seem to rule out finding any effect of the trade-off between the transfers and education because the need for a trade-off is ultimately stemming from the fiscal austerity. Yet, if these austerity measures are not included it becomes 17

18 difficult to know if the trade-off is really just a proxy for a general need to curtail spending in a country, or if it relates to partisan strategies concerning rising deindustrialization more specifically. Table A1 in the appendix lists all the variables with the operationalisations and sources. All variables are measured as levels. This is the most appropriate choice in our opinion because it is the absolute strength of left- and right-wing parties as well as the absolute risk exposure that must be expected to matter. As noted by Huber and Stephens (2001) regarding partisan governments, the impact of a 10% change in left-wing strength is highly dependent on whether the absolute level is 5% rising to 15%, or if it is 45% rising to 55%. The same goes for the impact of de-industrialization because we only expect de-industrialization to matter when the absolute level becomes sufficiently high to affect the median voter. A 10% increase in deindustrialization that does not affect the median voter is unlikely to make right-wing parties respond by expanding education. The regression models follow the convention of the literature by relying on Prais- Winsten estimation with the panel-corrected standard errors and lagged dependent variables suggested by Beck and Katz (1995). Recently some voices of critique have been raised against the use of lagged dependent variables, which risk suppressing theoretically interesting variation in the key variables (Achen 2000). We keep the lagged dependent variables in the model for three reasons. First, because we expect the speed of adjustment to be slow, with the long term effect being bigger than the immediate effect. According to Beck and Katz (2004) a lagged dependent variable is more appropriate than an AR(1) specification in this situation. Second, because it is well-established that social policy is prone to path dependency due to vested interests and learning effects, which may guide how governments respond to new social risks like de-industrialization (Pierson 1994; 2000). A lagged dependent variable is an appropriate technique to factor-out such influences stemming from past policies, thereby isolating the effect of current-day strategic choices of the partisan 18

19 governments (Keele and Kelly 2006). As pointed out by Beck and Katz (2004), it is, third and finally, also the case that the dominating effect of the lagged dependent variable is often overemphasized, implying that the problems of including one is smaller than sometimes expected. The regressions does not include unit dummies, or fixed effects, because unit dummies completely absorb differences in the level of independent variables across units (Plümper et al. 2005, p. 331, emphasis in original). Thus, given that we are using levels of independent variables we cannot simultaneously include unit dummies. As Plümper et al. (2005, p. 334) notes: [I]f a theory predicts level effect on levels or on changes, a fixed effect specification is not the model at hand. If a theory predicts level effects, one should not include unit dummies. In these cases, allowing for a mild bias resulting from omitted variables is less harmful than running a fixed effects specification. Following these considerations, the next section reports the findings of the analysis. 4. Findings Table 1 presents the results of the statistical tests. We are interested primarily in the three variables in the top of the table. We start by looking at the determinants of transfers in the 1980s. Model I shows the effect of partisan governments without taking the trade-off into account, while Model II includes the trade-off. In Model I the partisan government variable is insignificant, but the trade-off makes it highly significant and boosts the coefficients. Model III and IV show the same pattern for the 1990s. In short, when including the trade-off variable left-wing governments are strongly associated with more spending on transfers. Next, Model V and VI report the determinants of educational spending in the 1980s and Model VII and VIII report the determinants for the 1990s. It turns out that right-wing governments is significantly associated with more educational spending 19

20 disregarding the inclusion of the trade-off variable, although the trade-off do accentuate the partisan response as can be viewed from the bigger coefficients in Model VI and VIII compared to Model V and VII, respectively. TABLE 1 ABOUT HERE Figure 4 shows the marginal effects of the main explanatory variable on transfers and education using Model II, IV, VI and VIII, i.e., the full models with the trade-off variables included. A number of things are noteworthy. First, the partisan strategies are accentuated as de-industrialization rises. This follows from the argument above because the risk exposure to the median voter is fairly slight at low levels of de-industrialization, but then increases together with de-industrialization. Left-wing governments are responding to de-industrialization by expanding transfer programs even at fairly low levels of de-industrialization, whereas right-wing governments in the 1980s only expand education at higher levels of de-industrialization. At low levels of de-industrialization leftwing governments are in fact spending more on education than right-wing governments (the two graphs in the bottom row), which is in line with the argument of the paper because the need for prioritizing between the two programs all else equal will be smaller. Another thing to note is that the results are strongest for the 1980s (the two graphs on the left side) and weakest for the 1990s (the two graphs on the right side) where the confidence interval is wider and include more of the relevant observations. This, too, follows from the argument in the sense that we expected the slowdown and convergence in levels of de-industrialization to reduce the explanatory power of the partisan government variable. FIGURE 4 ABOUT HERE 20

21 We have tested for the influence of outliers using the jackknife procedure. It turns out that the results are not driven by outliers. The findings are, in sum, quite robust. They allow us to conclude that de-industrialization leads left-wing governments to expand transfers and right-wing governments to expand education. They show us that these diverse partisan strategies are accentuated by the need for making trade-offs between the two types of programs, but that this effect is particularly pronounced for left-wing governments. We argued above that this is because left-wing governments on average preside over bigger initial budgets that allow for lesser additional expansion. As expected we also found that the partisan strategies are most pronounced when deindustrialization is biggest and that the argument in general faired best in the 1980s. We are, consequently, able to verify all four hypotheses. 5. Ad fontes: Insurance, redistribution, and the partisan strategies of social protection The causal argument of the paper is based on the gradual rise of de-industrialization. Yet, the process of de-industrialization seems to be gradually grinding to a halt with most of the change in 1990s being caused by the catch-up of the laggards. In this narrow sense, the paper merely describes a historical event, however important in its own right, and may not tell us much else beside that. But the paper simultaneously contains some more general lessons that go beyond the idiosyncrasies of the past decades and points towards more universal mechanisms that might be expected to continue to operate and which therefore should inform future research. In this final section we will discuss these lessons in relation to the existing literature. The theory of de-industrialization taps into the varieties of capitalism approach that has won great popularity the recent decade (Estevez-Abe et al. 2001; Iversen and Soskice 2001; 21

22 Martin and Swank 2004). The argument that preferences for insurance against skill redundancy drive social policy provides an alternative to the power resource theory. This latter theory posits that the primary motive of economic and political actors is redistribution, and that partisan differences are likely to play an important role in terms of how much redistribution one will find in a country (Stephens 1979; Korpi 1983; 2006; Bradley et al. 2003). In this paper we took the lead from the theory of de-industrialization in assuming that this process of labor market change constitutes a major motor that is likely to make the median voter more favorable to spending. Critically, however, we simultaneously argued that how the preference for more spending is met depends on the party in government. Left-wing governments will systematically opt for those welfare programs that entail most redistribution, while right-wing governments just as systematically will opt for those programs that redistribute as little as possible. As such, the argument presented here promises to function as a stepping-stone for further integration between the two sets of literature. Iversen (2005) notes that countries with big welfare states generate high levels of both insurance and redistribution, but as argued in the paper it is important not to presume a priori that big spending equal extensive insurance equals extensive redistribution. In future work it will therefore, above all, be important to consider more carefully the theoretical core of the two theories and re-evaluate how that core relates to the function of individual programs in terms of insurance and redistribution. Understanding this gap between insurance and redistribution is vital because it captures why partisan governments can be expected to matter quite often; if insurance and redistribution was the same thing, parties interested in capturing the median voter would have no room working to the benefit of their core constituency. In that case we should only see partisan differences when new social risks affect the work force with a lower risk exposure than the median 22

23 voter, but that would be a different type of effect with right-wing governments refraining from government involvement all together (cf. Kwon and Pontusson 2008). Distinguishing between programs that primarily provide insurance (education) and those that both provide insurance and redistribution (transfers) will be a good starting point, although arguably the transfer component can be disaggregated further. Adding to this, however, are two complications. First, as noted by Pontusson (2005), even with a two-dimensional approach it is probably not possible to account for all welfare programs. Especially programs like elderly care and childcare appear historically to have been driven by a third logic, namely one of gender equality, or defamilisation (Esping-Andersen 1999). Such non-economic preferences are alien to both the varieties of capitalism and power resource theories, and it will be an important task to gauge how they intersect with the preferences for insurance and economic redistribution not least as dropping fertility rates and a soaring age burden calls for increasing female labor force participation. A second complication is constituted by the changing societal context, not least the consequences of the maturation of the welfare states themselves. As argued by Pierson (2001), this has had profound consequences for the options available to politicians and in this paper we showed how it entailed a need for prioritizing between programs as across-the-board expansion no longer is a viable solution. That is, even though expansion in the real world may be feasible sometimes, fiscal austerity nevertheless generates a new policy logic to the extent that rising median voter risk exposure cannot be matched by an expansion of the total budget. In our opinion the need to prioritize in the context of rising median voter risk exposure is therefore likely to accentuate partisan differences at the level of the individual programs, but wash them out at the level of total spending. It is an expectation that gets prima facie validation from the fact that existing studies finding no impact of partisan governments since 1980 have mostly been focused on more 23

24 aggregated spending categories (Huber and Stephens 2001; Swank 2001; Kwon and Pontusson 2008). Hopefully future studies will be able to look into the effects of partisan government over a prolonged period of time at both aggregated and disaggregated levels, getting a better view at this. Importantly, however, the partisan effects are not just direct, but also indirect because they have to be seen in the context of the alternative programs that the partisan governments are actively discounting. Needless to say, this makes getting the causal story of partisan effects right considerably more difficult than has hitherto been appreciated in most of the literature. 24

25 References Achen, Christopher H. (2000) Why Lagged Dependent Variables can Suppress the Explanatory Power of Other Independent Variables. Paper prepared for the Annual Meeting of the Political Methodological Section of the American Political Science Association, UCLA, July 20-22, Allan, J. P. and Scruggs, L. (2004) Political Partisanship and Welfare State Reform in Advanced Industrial Societies, American Journal of Political Science, 48 (3), Beck, Nathaniel and Jonathan N. Katz (1995) What to Do (and Not to Do) with Time-Series Cross- Section Data, American Political Science Review, 89, 3, Beck, Nathaniel and Jonathan N. Katz (2004) Time series-cross section issues: Dynamics. New York University. Working paper. Boix, Carles (1998) Political Parties, Growth, and Equality. Cambridge: Cambridge University Press. Bradley, David, Evelyne Huber, Stephanie Moller, François Nielsen and John D. Stephens (2003). Distribution and redistribution in postindustrial democracies, World Politics 55 (January), Busemeyer, Marius R. (2009). Social democrats and the new partisan politics of public investment in education, Journal of European Public Policy 16 (1), Castles, Francis G. (2004) The Future of the Welfare State. Crisis Myths and Crisis Realities. Oxford: Oxford University Press. Cusack, Thomas, Torben Iversen and Philip Rehm (2006). Risks at work: The demand and supply sides of government redistribution, Oxford Review of Economic Policy 22 (3),

26 Esping-Andersen, G. (1985) Politics against markets. The social democratic road to power. Princeton: Princeton University Press. Esping-Andersen, G. (1999) Social Foundations of Postindustrial Economies, Oxford: Oxford University Press. Estevez-Abe, Margarita, Torben Iversen and David Soskice (2001) "Social Protection and the Formation of Skills: A Reinterpretation of the Welfare State. In Peter Hall and David Soskice eds. Varieties of Capitalism: The Institutional Foundations of Comparative Advantage (London: Oxford University Press. Goodin, Robert E. and Julian Le Grand (1987) Not just for the poor. London: Allen & Unwin. Huber, E., and Stephens, J. D. (2000) Partisan Governance, Women s Employment, and the Social Democratic Service State, American Sociological Review 65 (3): Huber, E. and J.D. Stephens (2001) Development and Crisis of the Welfare State. Parties and Politics in Global Markets. Chicago and London: The University of Chicago Press. Huber, Evelyne, Charles Ragin, John D. Stephens, David Brady and Jason Beckfield (2004). Comparative Welfare States Data Set. Northwestern University, University of North Carolina, Duke University and Indiana University. Immergut, E., Anderson, K., Schulze, I. (red.) The Handbook of West European Pension Politics, Oxford: Oxford University Press. Iversen, Torben. (2001) The dynamics of welfare state expansion: Trade openness, deindustrialization, and partisan politics, in P. Pierson (ed.), The New Politics of the Welfare State, Oxford: Oxford University Press. Iversen, Torben (2005). Capitalism, Democracy, and Welfare. Cambridge: Cambridge University Press. 26

27 Iversen, T. and T.R. Cusack (2000) The Causes of Welfare State Expansion. De-industrialization or Globalization? World Politics, 52, April issue, Iversen, Torben and David Soskice (2001). An asset theory of social policy preferences, American Political Science Review 95 (4), Iversen, Torben and David Soskice (2006). Electoral institutions and the politics of coalitions: Why some democracies redistribute more than others, American Political Science Review 100 (2), pp Iversen, T. and J. D. Stephens (2008) Partisan politics, the welfare state, and the three worlds of human capital formation, Comparative Political Studies, 41, 4/5, Keele, Luke and Nathan J. Kelly (2006) Dynamic models for dynamic theories: The ins and outs of lagged dependent variables, Political Analysis 14 (2), p Korpi, Walter (1983). The democratic class struggle. London: Routledge & Kegan Paul. Korpi, Walter (2006) Power resources and employer-centered approaches in explanations of welfare states and varieties of capitalism, World Politics 58 (January), Korpi, Walter and Joakim Palme (1998) The paradox of redistribution and strategies of inequality: Welfare state institutions, and poverty in the Western countries, American Sociological Review 63 (October), Korpi, Walter and Joakim Palme (2003). New politics and class politics in the context of austerity and globalization: Welfare state regress in 18 countries, , American Political Science Review 97 (3), pp Kwon, Hyeok Yong and Jonas Pontusson (2008). Unions, globalization and the politics of social spending in OECD countries, Paper presented at the workshop Government Partisanship in Comparative Political Economy, September , Princeton University. 27

28 Le Grand, Julian (1982) The strategy of equality. Redistribution and the social services. London: George Allen and Unwin. OCED (various years). Education at a glance. Paris: OECD Publishing. OECD (various years). Labour Force Statistics. Paris; OECD Publishing. OECD (1992): Public educational expenditure, costs and financing: An analysis of trends Paris: OECD Publishing. OECD (2008): OECD.StatExtracts. Paris: OECD Publishing. Downloaded from Pierson, P. (1994) Dismantling the Welfare State? Reagan, Thatcher, and the Politics of Retrenchment. Cambridge: Cambridge University Press. Pierson, Paul (2000). Increasing returns, path dependence, and the study of politics, American Political Science Review 94: 2, pp Pierson, P. (2001) Post-industrial pressures on the mature welfare states, in P. Pierson (ed.), The New Politics of the Welfare State, Oxford: Oxford University Press. Pfeffer, Fabian T. (2008). Persistent inequality in educational attainment and its institutional context, European Sociological Review, 24 (5), Plümper, Thomas, V.E. Troeger and P. Manow (2005) Panel Data Analysis in Comparative Politics: Linking Method to Theory, European Journal of Political Research, 44, 2, Pontusson, Jonas (2005). Varieties and commonalities of capitalism, in David Coates (ed.) Varieties of capitalism, varieties of approaches. London: Palgrave MacMillan. Shavit, Y. and Blossfeld, H.-P Persistent Inequality. Changing Educational Attainment in hirteen Countries, Boulder, Colorado: Westview. Stephens, John D (1979). The transition from capitalism to socialism. Urbana and Chicago: University of Illinois Press. 28