Preferences in Context

Abstract

Political economists have increasingly looked to understand social welfare policy as a product of individual-level demand for social spending. This work hypothesizes that individuals with riskier jobs demand more social spending and that large welfare states emerge where there are more of such individuals. In this article we build on the “policy feedback” literature to argue that existing welfare institutions condition how individual-level factors affect social policy preferences. Specifically, we argue that institutions directly altering the risk of unemployment (employment protection legislation) and those that delink benefits from the labor market create a more uniform system of social risk that reduces the importance of individual-level risk in shaping policy preferences. We test these propositions using multilevel analysis of 19 advanced industrial countries in 2006. We find that individual risk matters for social policy preferences only where employment protection is low and welfare benefits are dependent on employment.

Keywords

policy feedback risk welfare state public opinion

Why do some individuals support spending on social programs more than other people? There has been a resurgence of interest in this question as scholars have looked to differences in popular support for social policy to explain differences in the origins and persistence of welfare states (Brooks & Manza, 2007). Many of these “bottom-up” approaches to explaining welfare state variation rest on deductively derived claims about how individuals’ economic positions shape their attitudes toward the welfare state (Cusack, Iversen, & Rehm, 2005; Iversen & Soskice, 2001; Rehm, 2009, 2011). In this work, welfare states largely work to insure against individual risks. Thus, individuals in jobs with high unemployment risks or with skills specific to that particular job are more likely to support social spending as a hedge against their own labor market risks. Where such jobs proliferate, the state develops more extensive social insurance: bigger welfare states follow the distribution of individual preferences.

Although this work has advanced a number of crucial insights explaining public support for the welfare state, it hinges on strong assumptions about the nature of individual risk and preferences. For these scholars, individual risk comes from the inherent riskiness of one’s job or skills. However, this claim raises the question of whether individual risk and thus individual spending preferences exist prior to welfare state institutions. Both labor market institutions and the benefits that individuals receive from the state can substantially modify the inherent riskiness of a given job. If systems of social protection shape how much economic risk individuals face, then they should also matter for determining individual demand for social welfare.

The claim that policies powerfully shape public preferences has a long history in the study of comparative welfare states (Jaeger, 2009; Jordan, 2010; Korpi & Palme, 1998; Larsen, 2008; Linos & West, 2003; Rothstein, 1992). These “policy feedback” approaches argue that welfare policies alter the cost–benefit calculus for individuals. Larger welfare states, in offering more generous benefits, tie more citizens to the state and increase support for it (Korpi & Palme, 1998). Although cross-national variation in levels of support for social spending provides some evidence for these arguments, empirical analyses of the distribution of individual opinion produce more mixed results (Bleckesaune & Quadagno, 2003; Svallfors, 1997). These ambiguous empirical findings raise questions about the precise connection between the distribution of individual support for the welfare state and existing policy structures

In this article, we build on both the feedback and the risk-based literatures to present a new understanding of how individuals’ policy contexts affect their social policy preferences. We argue that what citizens want from the government depends on how existing policies shape their experience of individual risk. Different social policy institutions create different systems of social risk. Social risk refers to the baseline risk citizens have of needing social benefits provided by the state, regardless of individual characteristics. Where welfare policies create a more uniform system of social risk, they reduce the importance of individual risk in preference formation.

Uniform systems of social risk can emerge for two reasons. First, existing institutions can alter the underlying risk of unemployment for a given group by making it harder for employers to fire and hire employees. Many countries have significant employment protection legislation that reduces differences across employees in terms of the risks of both unemployment and reemployment. Second, some countries provide benefits to individuals independent of the labor market. Countries that provide individual benefits based on citizenship or residence, rather than employment status, make all citizens equally reliant on the state, and varying individual economic risk does not translate into varying risk of needing public benefits. In both cases, policies create more uniformity across citizens in the risks they face; thus, preferences should accordingly also become more homogenous.

We test the connection between individual attributes and social policies by examining individual spending preferences over health care, unemployment benefits, and industrial aid. We look at spending preferences across contexts that directly modify individual labor market risk (varying levels of employment protection legislation) and those that do so indirectly by modifying the link between employment and benefits (employment dependent and independent health care benefits). We show that where institutions create more uniform social risk, they decrease the importance of individual risk in determining spending preferences and increase overall support for spending.

This finding has important implications for both the risk-based literature and work on policy feedback. First, it shows that the individual-level factors purported to explain support for large welfare states actually have a weaker causal impact in large welfare states than in smaller ones. Second, it presents a new argument about how policies affect public preferences. Institutions affect preferences not only directly—by creating new groups of recipients—but also indirectly—by modifying the way individuals experience risk and thus the importance of risk in shaping overall preferences.

We start in the second section with a brief overview of the existing literature on micro preferences and macro welfare regimes and then develop our argument about how these forces interact with one another. The third section describes our empirical approach and data. In the fourth section we examine the determinants of preferences over social insurance spending (unemployment, industrial aid, and health care). The fifth section concludes.

Connecting Macro and Micro Theories of Social Policy Preferences

The idea that welfare states not only play a redistributive role but also insure against the risk of income loss is well accepted in the social policy literature (Baldwin, 1990; Moene & Wallerstein, 2001). In recent years, scholars examining popular support for the welfare state have applied this insight to understand individual demand for social insurance. This research investigates how an individual’s position in the labor market helps shape his or her preferences over spending, with the basic finding that individuals with more inherent labor market risk support more public social spending (Iversen & Soskice, 2001; Rehm, 2009, 2011).

One particularly influential line of argument focuses on the risks associated with possessing different types of skills. Iversen and Soskice (2001) argue that individuals with specific skills are more likely to support social spending. Specific skills are skills that are used in only one industry or firm and are not transferable across different types of jobs. Broadly, those with specific skills face more risk of losing income should they become unemployed as they face greater difficulty in finding new work that is equally remunerative. According to this logic, because social insurance programs protect workers’ income during unemployment, those with specific skills should support them as a compensatory measure. Rehm (2011), conversely, argues that individuals’ occupationally specific unemployment risk determines their policy preferences, with risk again underpinning support for social policy.

These “bottom-up” models draw a line between individual economic attributes, support for social spending, and ultimately the size of the state itself. Estevez-Abe, Iversen, and Soskice (2001) suggest that individual skills aggregate to national outcomes: “[T]he greater the significance of a specific skill system … the greater the number of workers who possess the ‘key skills,’ and the more likely it is the median voter would be someone with an interest in supporting generous social protection” (p. 161). Rehm (2011) argues that countries with more homogeneity in individual-level occupational risk have a larger coalition in favor of social spending. In each case, the distribution of risks at the micro level shapes the distribution of popular attitudes toward the welfare state, ultimately producing different types of macro-level social programs.

Casting the micro–macro link along these lines is potentially problematic. The theory and empirical analysis of individual preference formation focus only on individual risk. Yet, although citizens face risk that is “individualized,” they demand institutions that will socialize risk. If these institutions already exist and work to moderate risk, a question emerges as to why individual risk should be abstracted from its institutional context. Indeed, a core assumption of the “varieties of capitalism” literature that this work draws on is that some societal institutions do shape the risks and choices individuals face in the labor market. But if societal institutions already moderate risk, how can we abstract individuals from those same institutions to determine their preferences?

Many scholars studying the welfare state argue that institutions do indeed affect individual preferences. Work studying “policy feedback” argues that welfare states not only are the product of political struggles but also are crucial determinants of welfare politics in their own right. Drawing on the insights of “historical institutionalist” scholars, this research sees institutional structures as constructing individual social policy preferences. This work suggests two mechanisms that create policy feedback. Some scholars see welfare structures as either tying together or dividing class loyalties by producing benefits that are universalistic or residual (Esping-Andersen, 1990; Korpi & Palme, 1998). Individuals belonging to different classes will have social policy preferences reflecting the structure of welfare benefits. Others scholars argue, instead, that policies do not create class divisions but rather their own specific constituencies of “policy takers” who become its chief supporters (Pierson, 1996). In both cases, the underlying argument is that citizens support programs or policy regimes that benefit them, thus more extensive programs construct more extensive political support.

These claims, though highly influential, have met both empirical and theoretical hurdles. Although cross-national surveys do show systematically varying levels of support for spending across countries, the hypothesis that individual preferences should mirror institutionally produced class cleavages has found mixed support (Bleckesaune & Quadagno, 2003; Jaeger, 2009; Svallfors, 1997). The “policy-taking” hypothesis has also met mixed results in empirical studies. Soss and Schram (2007) and Lynch and Myrskala (2009) find little evidence that policies of, respectively, cash welfare provision and public pensions have reshaped public opinion or created strong support bases among their recipients. Thus, questions remain about the mechanisms linking individual attributes and macro-level institutions.

An emerging literature has begun to resolve these empirical and theoretical dilemmas by moving beyond broad regime or programmatic effects and looking at the way policies condition individual preferences in narrower terms. Studies such as those by Linos and West (2003), Larsen (2008), and Jordan (2010) examine how particular programs or “social policy regimes” condition how individual attributes such as gender, attitudes toward others, and class matter for policy preferences. Here macro regimes alter the impact of specific micro mechanisms, offering a clearer causal mechanism underlying public support for the welfare state. However, this literature does not engage with the insurance function of the welfare state or directly connect with the literature focusing on individual risk, leaving open questions as to how institutions interact with individual-level characteristics in shaping demand for social insurance.

We argue that social policy contexts matter because they modify the degree to which individuals experience direct labor market risk and thus the importance of this risk in shaping individual preferences. The core mechanisms shaping individual preferences in the risk-based literature are (a) underlying differences in unemployment risks associated with a particular job or (b) underlying differences in the risk of needing public benefits because of prolonged unemployment. Where individuals face higher risks of becoming unemployed or of prolonged unemployment, they are more likely to support extensive social spending than their fellow citizens in less risky jobs. We argue that these individual-level mechanisms are likely to matter where social policy compensates for these underlying individual-level risks but does not directly modify them. However, where institutions alter the incidence of risk in the labor market or the risk of needing public benefits, we expect they will have a feedback effect by reducing the importance of individual risk (occupational or skill differences) in shaping preferences. We focus on two ways institutions can do this.

First, social policy institutions can directly affect the distribution of unemployment risk. Many countries have labor market institutions designed to protect employment for existing workers. These institutions are alleged to encourage more investment in specific skills because they modify the inherent riskiness of such investment (Estevez-Abe et al., 2001). One particularly important set of institutions is employment protection legislation (EPL). EPL makes it harder for employers to fire existing workers, reducing the risk of unemployment for the already employed. EPL should thus make differences in individual risk less important in determining the risk of unemployment and thus the need for social insurance.

These same institutions, however, are also associated with problems of job creation—and more controversially unemployment (Lazear, 1990; Organisation for Economic Co-operation and Development [OECD], 2004). Although scholars debate whether EPL causes high aggregate unemployment, there is evidence that EPL disrupts both the flow out of and the flow into employment (Blanchard & Portugal, 2001; Nickell, 1997). In countries with high unemployment and high EPL, even those with general skills can have trouble entering or reentering the labor market. This problem, dubbed as an “insider–outsider” problem, creates a logic of social risk that counteracts individual labor market risk (Lindbeck & Snower, 1989; Rueda, 2005). Regardless of skill specificity, individuals with jobs face a lower likelihood of becoming unemployed, but these same individuals face a uniformly higher risk of prolonged unemployment if they do become unemployed.

Second, social programs can indirectly alter individual-level risks by changing the likelihood that individuals will need public benefits. Where benefits are provided to people as citizens, rather than employees, higher than average risks of prolonged unemployment do not translate into higher risk of needing public spending, and vice versa. For instance, all individuals have guaranteed access to health care in Sweden, meaning that both those with high and those with low risks of unemployment in Sweden depend on the state for benefits, whereas in the United States those with a higher risk of long-term unemployment also have a higher risk of losing access to health insurance (and thus demanding public spending). Although no welfare states truly “decommodify” people, some go much further than others in how far they base benefits on labor market activity (Esping-Andersen, 1990). Where institutions provide benefits independent of the labor market, citizens with risky and nonrisky jobs are at equal risk of needing social benefits.

Taken together, we see that social policy institutions can, but do not necessarily, homogenize both the risk of prolonged unemployment and the costs of unemployment. Where institutions modify social risk by creating more uniformity across the population, they reduce the importance of individual risk in determining individual preferences. Given that differences in risk exposure are crucial to Iversen and Soskice (2001) and Rehm’s (2009, 2011) claims about differences in individual preferences, this contextual effect is potentially important. Indeed, for Rehm (2011), risk homogeneity is crucial to determining spending outcomes—populations with more homogenous risks tend to be more supportive of the welfare state, thus causing larger welfare states. However, if policies themselves construct risk homogeneity, rather than it being inherent in particular economies, this complicates the idea that micro preferences seamlessly drive macro policy outcomes. Our theoretical approach suggests that by changing the level of social risk, and thus modifying the impact of individual risk, national welfare institutions have powerful feedback effects on individual preferences over welfare spending.

Our claim raises a second question. If institutions homogenize the experience of risk across individuals, do they work to make low-risk individuals behave more like high-risk individuals or vice versa? We argue that either pathway is theoretically possible, but, in line with existing empirical work (Finseraas, 2008; Rehm, 2011), the former seems more likely. The homogenization of risk through social programs makes all individuals equally dependent on the state for benefits. In an unmodified market, low-risk individuals, such as those with general skills, are (a) able to find new jobs easily and (b) less reliant on the public sector because they receive many occupationally provided benefits (e.g., health care). As institutions homogenizing risk come into place, these same individuals face a different environment. First, where EPL not only modifies the distribution of unemployment risk but also increases barriers to entry into jobs—as in countries with insider–outsider labor markets—general-skilled individuals now face an equivalent risk to specific-skilled individuals of prolonged unemployment if they lose their jobs. Thus, as in Saint-Paul (2000), EPL will become associated with higher aggregate demand for unemployment insurance as all citizens now face longer spells of unemployment.

Second, where benefits are provided independent of employment, public spending now benefits low-risk individuals directly and their lower level of labor market risk does not translate into lower risk of needing the public system. Indeed, rather than having private occupational benefits, they must now rely on the public sector for benefits. As such, they should support spending. In both cases, as risk is decoupled from inherent features of an individual’s job and cast at a societal level, those who are “winners” in a less regulated labor market are now more dependent on the state for their well-being, making them more similar to their colleagues in inherently risky jobs.

These claims suggest a very different take on the micro–macro connection than much of the literature. We are arguing that societal institutions can either reflect or modify individual risk. Where these institutions modify risk, they tend to reduce individual differences based on risk and increase the likelihood of overall support for the welfare state. It is problematic to argue that large welfare states exist today because of current individual demands based on risk—the structure and impact of individual risk is at least partly endogenous to existing institutions. Indeed, individual risk should matter least in generating demand for higher spending in countries that already have high spending and a more homogenous system of social risk. By contrast, where social protection is lacking, individuals do face substantial risk based on their labor market position. Consequently, we expect individual risk to be most important as a predictor of support for public spending in countries where social protection is comparatively low. In the long run, it is true that macro-level institutions may be endogenous to public opinion. Nonetheless, in the medium run, when citizens form social policy preferences they are partly doing so in response to a fixed policy environment.

Equally, the argument here contributes to the policy feedback literature. Our argument is broadly congruent with the existing literature, but it specifies the micro–macro connection differently. Although the core mechanism linking policies to individual preferences in the existing feedback literature is benefit receipt, we argue that risk modification is more important with regard to social insurance preferences. Institutions such as EPL that do not directly affect what an individual receives from the state can still have a feedback effect by modifying labor market risk.

To test these claims, we examine two mechanisms of social risk modification: the extent of EPL and the existence of employment independent benefits. EPL directly modifies the distribution of individual labor market risk, and thus is likely to matter for spending on programs directly related to the labor market, such as unemployment insurance and industrial aid. We expect the effect of individual risk factors such as skill specificity on support for unemployment insurance and industrial aid to be much smaller in countries with robust EPL. We further expect that since EPL is often associated with a higher aggregate risk of lengthy unemployment, general skills individuals will look more like their specific skilled colleagues and thus demand more spending. Controlling for unemployment levels should remove this effect.

Hypothesis 1a: Higher levels of EPL will reduce the importance of individual risk factors such as skill specificity in determining individual social insurance spending preferences.

Hypothesis 1b: EPL will tend to increase the overall level of support for social insurance. This effect should be reduced once unemployment levels are taken into account.

Employment-independent benefits indirectly alter individual risk. Here we focus on the risk of losing job-related benefits, for example health care, rather than income. Some countries provide health care to all citizens, regardless of their employment status (Hacker, 2004). These countries may provide universal health insurance through a “single payer” (as in Canada) or a national health service where the government actually produces health care (as in Sweden and the United Kingdom). In each case, benefit receipt is not tied to work and thus is unaffected by spells of unemployment. By contrast, other countries have a connection between work and health benefits. Although very few OECD countries (the United States is an exception) tie health benefits directly to work, the extent of this coverage and the associated contributions are tied to employment in many Continental European countries, Japan, and Ireland. Skill specificity should thus matter more in determining spending preferences where benefits are dependent on employment and less where they are not.¹ Because employer-independent systems make individuals equally dependent on the state, we expect they will increase the overall level of support. Since individual health risks are not dependent on the level of unemployment, this outcome should not be affected by aggregate unemployment.

Hypothesis 2a: Employment independent benefits will reduce the importance of individual risk factors such as skill specificity in determining individual spending preferences.

Hypothesis 2b: Employment independent benefits will tend to increase the overall level of support for the welfare state. This effect should not be affected by aggregate unemployment.

In summary, welfare institutions create a system of social risk that directly modifies individual risk. Tracing the interaction between an individual’s labor market position and spending preferences requires taking this social risk into account.

Data and Statistical Approach

In this section we describe the data and statistical techniques we use in our empirical tests of the propositions developed above. The data set we use is the 2006 Role of Government International Social Survey Programme (ISSP 2006), which asks a variety of questions about whether individuals would prefer the government to increase or decrease spending on a series of social policies. Our sample covers between 18 and 19 countries and around 15,000 individuals. The previous iteration of the Role of Government survey—the ISSP 1996 data set—was key to the development of hypotheses about skill specificity and policy preferences (Iversen & Soskice, 2001). The ISSP 2006, also used by Rehm (2011), covers a broader range of countries than the ISSP 1996, producing greater variation among national social policies.²

Unlike most cross-national surveys, the ISSP asks questions about government spending on specific social programs. Although previous authors have typically combined these disparate measures into single or multiple dimensions of aggregate spending preferences, we retain the disaggregated measures since our interest is in how specific national-level policy institutions affect individual preferences about specific policies.³ The dependent variables we examine are unemployment benefits, support for declining industries, and health care spending.⁴ Individuals are asked the following question:

Listed below are various areas of government spending. Please show whether you would like to see more or less government spending in each area. Remember that if you say “much more,” it might require a tax increase to pay for it.

Thus, this survey question directly addresses the issue of fiscal trade-offs—individuals are asked to compare increased spending with increased taxes. Respondents are provided with the following choices: spend much more, spend more, spend the same, spend less, and spend much less. Following Rehm (2011), we use a binary scale of 0 for spend much less, spend less, and spend the same, and 1 for spend more and spend much more.⁵

As independent variables we employ individual-level variables, national-level variables, and the cross-level interactions of individual- and national-level variables. Our chief individual-level independent variable is a measure of individual risk: skill specificity. We use the measure of skill specificity developed by Iversen and Soskice (2001) and Cusack et al. (2005). This measure derives the specificity of an individual’s skills from his or her occupation, using the 1988 International Standard Classification of Occupations (ISCO88)—a hierarchical categorization of occupations that produces occupational codes one to four digits long. The skill-specificity index uses the number of four-digit occupation subcategories in each two-digit category as a proxy for the specialization of skills. For example, ISCO code 82, “machine operators and assemblers,” has 36 subcategories, whereas ISCO code 52, “models, salespersons, and demonstrators,” has only 3 subcategories. Yet both groups account for around 5% of the workforce. Iversen and Soskice (2001) argue that the former group has much more specific skills than the latter. We use the occupational categories provided in the ISSP 2006 and follow the instructions provided in Cusack et al. (2005) to produce a measure of their s1 skill specificity index, which ranges from 0.48 to 4.11. In the online appendix (available at http://www.polisci.umn.edu/~ansell/papers.htm), as a cross-validation measure we also examine Rehm’s (2011) individual occupational risk variable, coded as the national unemployment rate for each of the 9 one-digit ISCO codes.

To capture general skills, we use the ISSP’s 7-point scale of formal education to ascertain whether individuals have received a university degree, which is the highest category included. We create a dummy variable measuring whether individuals have attained a university education: We refer to this variable as degree. This variable is an important control in our analysis of social insurance since the risk-based literature predicts individuals with general skills will be less supportive of social insurance spending. We use this variable because countries in the sample differ dramatically in the structure of secondary education. Only when we turn to university degrees do we see a clear commonality and thus a comparable indicator of high general skills. Nonetheless, our results are very similar if the 7-point education scale is used.

At the national level, our core variables are EPL and health care structure. Our EPL index is taken from the OECD (2006) and is produced by aggregating measures of (a) regulation on temporary employment, (b) specific requirements for collective dismissal, and (c) protection of regular workers against dismissal.⁶ The EPL measure varies between 0.65 (United States) and 3.5 (Portugal). For our analysis of the health care spending question, the national policy we examine is whether the country has an employment-independent health care system. We code Australia, Canada, Denmark, Finland, New Zealand, Norway, Portugal, Spain, Sweden, and the United Kingdom as employment-independent systems and France, Germany, Ireland, Israel, Japan, the Netherlands, Switzerland, and the United States as employment-dependent systems.⁷ To create our cross-level interaction variables, we multiply our individual-level variables skill specificity and degree by the relevant national variable (EPL for unemployment and industrial aid preferences, health care system for health preferences).⁸ In Table 4 below, we also examine the effects of other national variables: Here we use the national unemployment level in 2005, taken from OECD (2006), the proportion of secondary school students in vocational education from OECD (2008), and the Gini coefficient for occupational unemployment from Rehm (2011). The online appendix provides descriptive statistics and correlations among these variables.

We also include a battery of control variables, following the specifications used in Iversen and Soskice (2001) and Rehm (2011). We begin with a measure of income. The ISSP does not provide a direct interval measure of family income but instead uses a 10-point scale.⁹ We also include a series of employment and demographic variables: age, gender (1 = male), and dummy variables for unemployment, nonemployment, self-employment, part-time work, and whether the individual belongs to a trade union. We also include a variable measuring political information to control for exposure to pro- or antispending arguments in the media. To control for the anti-redistribution effect of religiosity hypothesized by Scheve and Stasavage (2006), we use an 8-point measure of religious attendance. Finally, we use a 5-point scale of partisanship ranging from far left to far right, drawn from party identification. This latter variable dramatically reduces the size of the data set since independents and nonvoters are excluded. Like Iversen and Soskice (2001), we impute partisanship through multiple imputation to retain individuals who would otherwise drop out of the analysis.¹⁰

We now turn to our estimation techniques. We perform three different types of multilevel binary dependent variable analysis: a logit model with country fixed effects, a logit model with random effects, and a two-step model with random intercepts and coefficients. All estimations use heteroscedasticity-robust standard errors clustered by country.¹¹ The models presume that observed survey answers reflect an unobserved variable y^*_ij, measuring underlying support for the policy, with i indexing individuals and j indexing countries. The equation for the fixed effects model is,

Y_{IJ}^{*} = α_{0} + β_{1} S_{IJ} + β_{2} D_{IJ} + β_{3} Z_{J} S_{IJ} + β_{4} Z_{J} D_{IJ} + W_{ij} b + U_{J} + ε_{IJ}

This equation has an intercept α₀, the skill specificity variable for each individual s_ij and its coefficient β₁, the university degree variable for each individual d_ij and its coefficient β₂, the interaction of skill specificity with the national policy z_j and the coefficient β₃, the same for the degree variable with coefficient β₄, the matrix of remaining controls and their coefficients W_ijb, a country fixed effect u_j, and an individual error term ϵ_ij. By contrast, the equation for the random effects model is,

Y_{IJ}^{*} = α_{0} + β_{1} S_{IJ} + β_{2} D_{IJ} + β_{3} Z_{J} S_{IJ} + β_{4} Z_{J} D_{IJ} + β_{5} Z_{J} + W_{ij} b + v_{J} + ε_{IJ}

This equation differs in two ways. First, we directly estimate the effect of the national policy z_j, and, second, the country-specific error term υ_j is randomly drawn from a normal distribution, with mean zero and variance σ²_v.

These two multilevel techniques, used in Tables 1 through 3, have complementary properties. The fixed effects model, also used in Iversen and Soskice (2001) and Hellwig (2001), controls for cross-country heterogeneity in responses and for variation in spending preferences related to the existing level of spending, a potentially important omitted variable. However, since the national-level policy z_j is invariant within countries, we cannot estimate the level effects of national policies on aggregate spending preferences. Conversely, the random effects approach, also used in Finseraas (2008) and Solt (2008), permits the estimation of the effects of national policies such as EPL. What the random effects approach loses in its ability to control for national-level idiosyncrasies it gains in its ability to examine cross-sectional variation at the national level.

Table 1.

Unemployment Preferences and Employment Protection Legislation (EPL)

	(1)	(2)	(3)	(4)	(5)	(6)
	FE	RE	FE	RE	FE	RE
Individual
Income	−0.461***	−0.486***	−0.417***	−0.446***	−0.417***	−0.446***
	(0.047)	(0.034)	(0.049)	(0.036)	(0.048)	(0.036)
Skill specificity	0.309***	0.335***	0.278***	0.320***	0.323***	0.360***
	(0.064)	(0.080)	(0.070)	(0.082)	(0.098)	(0.103)
Univ. degree	−0.148	−0.109	−0.179	−0.167	−0.110	−0.125
	(0.187)	(0.150)	(0.178)	(0.153)	(0.245)	(0.199)
Unemployed	0.969***	0.887***	0.962***	0.875***	0.964***	0.876***
	(0.189)	(0.111)	(0.188)	(0.120)	(0.188)	(0.120)
Nonemployed	−0.000	0.007	0.028	0.039	0.029	0.039
	(0.107)	(0.061)	(0.103)	(0.065)	(0.103)	(0.065)
Self-employed	−0.350***	−0.345***	−0.339***	−0.331***	−0.339***	−0.332***
	(0.078)	(0.071)	(0.067)	(0.075)	(0.068)	(0.075)
Part-time	0.118	0.103	0.106	0.091	0.108	0.092
	(0.103)	(0.069)	(0.105)	(0.073)	(0.105)	(0.073)
Age	0.006**	0.005***	0.006**	0.005***	0.006**	0.005***
	(0.003)	(0.002)	(0.003)	(0.002)	(0.003)	(0.002)
Gender	−0.095*	−0.080*	−0.092	−0.071	−0.092	−0.071
	(0.055)	(0.045)	(0.064)	(0.048)	(0.064)	(0.048)
Religiosity	0.016	0.015	0.008	0.005	0.008	0.005
	(0.014)	(0.011)	(0.013)	(0.011)	(0.013)	(0.011)
Union member	−0.175***	−0.166***	−0.128**	−0.121***	−0.128**	−0.121***
	(0.050)	(0.029)	(0.052)	(0.031)	(0.052)	(0.031)
Political info.	0.057	0.060***	0.064	0.071***	0.064	0.072***
	(0.039)	(0.021)	(0.039)	(0.023)	(0.039)	(0.023)
Partisanship			−0.211***	−0.213***	−0.211***	−0.213***
			(0.061)	(0.021)	(0.061)	(0.021)
National
EPL		0.523*		0.455*		0.239
		(0.267)		(0.259)		(0.312)
EPL × skill spec.	−0.101***	−0.117***	−0.082***	−0.108***	−0.059*	−0.088*
	(0.026)	(0.036)	(0.031)	(0.037)	(0.036)	(0.049)
EPL × univ.	−0.133	−0.140**	−0.126	−0.117	−0.099	−0.100
	(0.098)	(0.071)	(0.095)	(0.073)	(0.119)	(0.089)
Unemployment						0.129
						(0.109)
Unemp. × skill spec.					−0.015	−0.013
					(0.017)	(0.021)
Unemp. × univ.					−0.021	−0.013
					(0.051)	(0.039)
Constant	−1.167***	−2.557***	−0.616***	−1.803***	−0.606***	−2.142***
	(0.147)	(0.596)	(0.216)	(0.585)	(0.213)	(0.636)
N	14,432	14,432	13,120	13,120	13,120	13,120
Countries	18	18	18	18	18	18

Standard errors in parentheses. Models 3 through 6 use multiple imputation.

p < .10. **p < .05. ***p < .01.

Table 2.

Industrial Aid Preferences and Employment Protection Legislation (EPL)

	(1)	(2)	(3)	(4)	(5)	(6)
	FE	RE	FE	RE	FE	RE
Individual
Income	−0.339***	−0.344***	−0.304***	−0.307***	−0.303***	−0.306***
	(0.044)	(0.032)	(0.041)	(0.034)	(0.041)	(0.034)
Skill specificity	0.387***	0.361***	0.397***	0.361***	0.390***	0.363***
	(0.091)	(0.086)	(0.091)	(0.089)	(0.103)	(0.104)
Univ. degree	−0.642***	−0.722***	−0.689***	−0.763***	−0.499***	−0.612***
	(0.125)	(0.126)	(0.145)	(0.129)	(0.175)	(0.162)
Unemployed	0.136	0.098	0.129	0.117	0.127	0.115
	(0.136)	(0.121)	(0.133)	(0.129)	(0.131)	(0.129)
Nonemployed	0.096*	0.085	0.104*	0.096*	0.106*	0.097*
	(0.054)	(0.055)	(0.056)	(0.057)	(0.056)	(0.057)
Self-employed	−0.294***	−0.282***	−0.259***	−0.242***	−0.258***	−0.242***
	(0.098)	(0.059)	(0.089)	(0.062)	(0.089)	(0.062)
Part-time	0.060	0.077	0.042	0.061	0.044	0.063
	(0.073)	(0.060)	(0.073)	(0.062)	(0.073)	(0.062)
Age	−0.009***	−0.010***	−0.009***	−0.010***	−0.009***	−0.010***
	(0.002)	(0.001)	(0.002)	(0.001)	(0.002)	(0.001)
Gender	−0.603***	−0.580***	−0.560***	−0.543***	−0.559***	−0.543***
	(0.043)	(0.040)	(0.041)	(0.042)	(0.041)	(0.042)
Religiosity	−0.012	−0.019*	−0.023**	−0.030***	−0.023***	−0.030***
	(0.009)	(0.010)	(0.009)	(0.010)	(0.009)	(0.010)
Union member	−0.126***	−0.115***	−0.091***	−0.078***	−0.090***	−0.078***
	(0.030)	(0.026)	(0.029)	(0.027)	(0.029)	(0.027)
Political info.	0.180***	0.170***	0.194***	0.184***	0.194***	0.184***
	(0.025)	(0.020)	(0.027)	(0.021)	(0.027)	(0.021)
Partisanship			−0.111***	−0.117***	−0.111***	−0.117***
			(0.028)	(0.022)	(0.028)	(0.022)
National
EPL		0.463**		0.416**		0.317
		(0.191)		(0.187)		(0.227)
EPL × skill spec.	−0.100***	−0.092**	−0.104***	−0.088**	−0.114**	−0.091*
	(0.038)	(0.041)	(0.039)	(0.043)	(0.047)	(0.053)
EPL × univ.	−0.014	0.032	0.006	0.043	0.075	0.101
	(0.098)	(0.071)	(0.095)	(0.073)	(0.119)	(0.089)
Unemployment						0.061
						(0.079)
Unemp. × skill spec.					0.004	0.000
					(0.018)	(0.019)
Unemp. × univ.					−0.059*	−0.047
					(0.034)	(0.030)
Constant	−0.006	−1.160***	0.306*	−0.705	0.307*	−0.871*
	(0.167)	(0.433)	(0.177)	(0.431)	(0.177)	(0.473)
N	14,376	14,376	13,082	13,082	13,082	13,082
Countries	18	18	18	18	18	18

Standard errors in parentheses. Models 3 through 6 use multiple imputation.

p < .10. **p < .05. ***p < .01.

Table 3.

Health Care Preferences and Single-Payer Systems

	(1)	(2)	(3)	(4)	(5)	(6)
	FE	RE	FE	RE	FE	RE
Individual
Income	−0.163***	−0.163***	−0.132***	−0.132***	−0.132***	−0.132***
	(0.039)	(0.038)	(0.044)	(0.040)	(0.044)	(0.040)
Skill specificity	0.247***	0.220***	0.265***	0.218***	0.235*	0.148
	(0.053)	(0.054)	(0.082)	(0.058)	(0.129)	(0.124)
Univ. degree	−0.177**	−0.177**	−0.183**	−0.172*	−0.258	−0.213
	(0.088)	(0.084)	(0.094)	(0.090)	(0.310)	(0.185)
Unemployed	0.156	0.101	0.230	0.174	0.229	0.173
	(0.170)	(0.141)	(0.194)	(0.157)	(0.193)	(0.157)
Nonemployed	0.076	0.081	0.097	0.105	0.097	0.105
	(0.064)	(0.066)	(0.067)	(0.070)	(0.067)	(0.070)
Self-employed	−0.318***	−0.285***	−0.290***	−0.248***	−0.289***	−0.246***
	(0.093)	(0.065)	(0.096)	(0.069)	(0.094)	(0.069)
Part-time	0.066	0.045	0.079	0.051	0.078	0.049
	(0.053)	(0.070)	(0.052)	(0.074)	(0.051)	(0.074)
Age	0.000	−0.001	0.000	−0.001	0.000	−0.001
	(0.003)	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
Gender	−0.420***	−0.391***	−0.376***	−0.348***	−0.376***	−0.348***
	(0.071)	(0.047)	(0.077)	(0.050)	(0.077)	(0.050)
Religiosity	0.044***	0.045***	0.042***	0.042***	0.042***	0.042***
	(0.013)	(0.011)	(0.015)	(0.012)	(0.015)	(0.012)
Union member	−0.115***	−0.129***	−0.088**	−0.103***	−0.088**	−0.103***
	(0.035)	(0.030)	(0.036)	(0.032)	(0.036)	(0.032)
Political info.	0.137***	0.132***	0.165***	0.161***	0.165***	0.161***
	(0.035)	(0.023)	(0.033)	(0.025)	(0.032)	(0.025)
Partisanship			−0.098**	−0.113***	−0.098**	−0.113***
			(0.040)	(0.021)	(0.040)	(0.021)
National
Single payer (SP)		0.995***		1.085***		1.061***
		(0.317)		(0.305)		(0.302)
SP × skill spec.	−0.241***	−0.232***	−0.227**	−0.197**	−0.224**	−0.194**
	(0.064)	(0.073)	(0.094)	(0.079)	(0.088)	(0.079)
SP × univ.	−0.314**	−0.320***	−0.307**	−0.328***	−0.303*	−0.326***
	(0.137)	(0.107)	(0.146)	(0.113)	(0.159)	(0.113)
Unemployment						−0.063
						(0.069)
Unemp. × skill spec.					0.005	0.012
					(0.025)	(0.018)
Unemp. × univ.					0.013	0.007
					(0.046)	(0.028)
Constant	0.374*	−0.010	0.574**	0.165	0.589**	0.551
	(0.208)	(0.283)	(0.261)	(0.293)	(0.250)	(0.515)
N	15,064	15,064	13,350	13,350	13,350	13,350
Countries	19	19	18	18	18	18

Standard errors in parentheses. Models 3 through 6 use multiple imputation.

p < .10. **p < .05. ***p < .01.

The effects of social risk policies on policy preferences can be seen in two ways. First, in the random effects models the direct level effect of a policy can be read off its coefficient as the average effect of the policy on spending preferences across all citizens. Second, in both techniques, the policy-conditional effect of individual-level variables can be understood as the coefficient of the individual-level variable when the national-level variable equals zero plus the coefficient on the interactive variable multiplied by the value taken by the national-level variable.

Ideally, one would test a feedback model of preference formation using panel data, but unfortunately no cross-national panel survey with appropriate questions currently exists, and, moreover, we would need information on preferences before and after changes in EPL and health care policy and such changes are rather rare. Nonetheless, if policy feedback arguments are correct, then cross-sectional analyses should show that contextual effects come only through the relevant policy and not other national-level variables. We take this approach in Table 4 by estimating a series of “two-stage” models. This technique allows us to compare the relative importance of a variety of different national contextual variables—EPL, health care structure, unemployment, skills profile, and occupational unemployment inequality. Accordingly, we can show it is the social risk policy specific to the spending question that drives our results, not the national composition of the labor market nor the composition of skills.

Table 4.

Two-Stage Analyses

	Unemployment spending
	Skill specificity	Skill specificity	Skill specificity	Skill specificity	Skill specificity
Control	—	Single payer	Unemployment	Vocational	Rehm Gini
Corr. with EPL	—	.09	.59	.42	−.36
EPL	−0.095***	−0.105***	−0.070*	−0.083**	−0.081**
	(0.030)	(0.034)	(0.039)	(0.033)	(0.030)
Control		0.039	−0.016	−0.001	0.620
		(0.059)	(0.016)	(0.001)	(0.617)
Constant	0.292***	0.289***	0.339***	0.300***	0.112
	(0.067)	(0.068)	(0.081)	(0.075)	(0.188)
N	18	18	18	17	15

	Constant	Constant	Constant	Constant	Constant
Control	—	Single payer	Unemployment	Vocational	Rehm Gini

EPL	0.528*	0.582**	0.366	0.628**	0.672*
	(0.258)	(0.244)	(0.322)	(0.266)	(0.318)
Control		−0.671*	0.096	−0.017	7.358
		(0.377)	(0.113)	(0.010)	(6.284)
Constant	−2.057***	−1.780***	−2.308***	−1.427**	−4.167**
	(0.558)	(0.545)	(0.635)	(0.579)	(1.882)
N	18	18	18	17	15

	Health Care Spending
	Skill specificity	Skill specificity	Skill specificity	Skill specificity	Skill specificity
Control	—	EPL	Unemployment	Vocational	Rehm Gini

Corr. with single payer (SP)	—	.09	−.10	.04	−.12
SP	−0.170**	−0.157**	−0.158**	−0.180**	−0.184***
	(0.065)	(0.068)	(0.067)	(0.068)	(0.059)
Control		−0.026	0.009	0.000	2.084**
		(0.046)	(0.015)	(0.002)	(0.848)
Constant	0.171***	0.212**	0.109	0.155*	−0.331
	(0.047)	(0.098)	(0.107)	(0.084)	(0.210)
N	19	18	18	17	15

	Constant	Constant	Constant	Constant	Constant
Control	—	EPL	Unemployment	Vocational	Rehm Gini

SP	0.589**	0.821***	0.808***	0.862***	0.878***
	(0.279)	(0.249)	(0.249)	(0.250)	(0.271)
Control		−0.030	−0.025	−0.007	7.622*
		(0.179)	(0.057)	(0.006)	(3.783)
Constant	1.111***	0.934**	1.022**	1.202***	−1.034
	(0.206)	(0.405)	(0.398)	(0.358)	(0.932)
N	19	18	18	17	15

EPL = employment protection legislation. Standard errors in parentheses.

p < .10. **p < .05. ***p < .01.

At the “first stage” we run a separate logit analysis within each country, and then at the “second stage” we use coefficients from the first stage as the dependent variable and national-level variables as predictors. This is equivalent to running a model with both random intercepts and random coefficients for all independent variables. In the second stage analyses we adjust for the fact that the dependent variable, being a regression estimate, is measured with error, using the technique and code developed by Lewis and Linzer (2005). We compare the effects of relevant national policy variables on the coefficient for skill specificity and that for the constant term.

Individual Risks, Social Risk, and Support for Social Spending

We begin our empirical analyses by examining the role of EPL in shaping individual preferences over unemployment spending (Table 1) and aid to declining industries (Table 2) before turning to the role of employment-independent benefits in shaping health-spending preferences (Table 3). In each case we are focusing on the effect of individual risk—proxied by skill specificity—conditional on the national level of social risk. We argue that where EPL is high, or benefits are employment-independent, social risk is more uniform and individual-level risk will hence matter less in determining social insurance preferences. This implies a negative coefficient on the interaction between skill specificity and EPL (or employment independent benefits). We also expect that EPL and employer independent benefits will be associated with a higher aggregate level of support for spending.

Tables 1 through 3 all contain six models. Models 1 and 2 are baseline fixed and random effects models. Models 3 and 4 add a control for partisanship and accordingly use multiply imputed data sets. Models 5 and 6 further add a national-level measure for unemployment and its interaction with both skill specificity and degree. These final models control for an important conditioning variable that is strongly correlated with EPL (at .59) and that might affect the salience of individual skills (as in Iversen & Soskice, 2001, who examine this specification). We examine other potential conditioning national effects in Table 4.

Table 1 examines the interactive effect of EPL and skill specificity on preferences over unemployment spending. In Models 1 and 2 skill specificity has a statistically significant positive direct effect on unemployment spending preferences—as in Iversen and Soskice (2001)—but also a statistically significant negative coefficient for its interaction with EPL. This implies that as EPL increases, skill specificity has less of a positive impact on support for unemployment spending.¹² This is true whether we control for country-specific effects (Model 1) or include cross-sectional variation and the direct effect of EPL (Model 2).¹³ The direct effect of EPL is positive and statistically significant, implying average support for unemployment spending is higher in countries with greater EPL. Finally, we see statistically significant negative effects of income and of self-employment and positive effects of unemployment, age, union membership, and being female. In the online appendix, we show that this pattern holds up when we use the 5-point version of the dependent variable and if we replace skill specificity with Rehm’s (2011) measure of occupational unemployment risk.

Figure 1a presents the estimated probability of supporting increased spending on unemployment benefits for different levels of skill specificity, using the random effects specifications from Model 2.¹⁴ Two predicted probabilities (with 95% confidence intervals) are plotted, one for a country with a low EPL of score of 0.65—for example, the United States—and one for a country with a high EPL score of 2.89—for example, Germany. Only in the low EPL case is there a robust positive effect of skill specificity on unemployment spending preferences. In the high EPL case, the effect is indistinguishable from zero. The level effect of EPL is also apparent: General-skilled citizens are much more supportive of increased unemployment spending in high EPL countries than they are in low EPL countries. By contrast, specific-skilled individuals have similarly high levels of support across both systems. Figure 1b shows the declining effects of skill specificity as EPL increases: Once EPL exceeds 2 (its median), skill specificity ceases to have a significant effect. Substantively, we see it is only where social risk is less uniform that individual risk matters in determining unemployment spending preferences.

Figure 1.

Unemployment spending preferences and employment protection legislation (EPL)

Models 3 and 4 add partisanship, using the multiply imputed data sets. We see a strong negative effect of right-wing partisanship on spending preferences, but the effects of skill specificity and its interaction are consistent. Models 5 and 6 add the national level of unemployment and its interactions with skill specificity and degree. Here we do find a somewhat reduced coefficient for the interaction of skill specificity and EPL, unsurprising given the correlation of .59 between EPL and unemployment. The interaction remains significant at the 10% level, and, moreover, the interaction of unemployment with skill specificity has no effect: It is EPL not unemployment that conditions the effects of skill specificity. We do, however, find that including the level of unemployment wipes away the level effect of EPL. Since we hypothesized that general skills citizens support unemployment spending in high EPL countries because EPL makes reemployment more difficult, this finding is reassuring.

In Table 2 we examine the industrial support question. With regard to the interactive effect of skill specificity and EPL, we find very similar results to the previous analysis. Across all the models there is a robust negative effect of this interactive variable. Figures 2a and 2b show that when EPL is low, a 1-point shift in skill specificity is estimated to produce a 7% increase in support for industrial aid. However, this effect declines rapidly, and when EPL is high, there is no statistically discernable effect. Again, individual risk matters less, if it all, when the government creates more uniform social risk—here, by propping up declining industries that might otherwise make workers with specific skills redundant.

Figure 2.

Industrial aid preferences and employment protection legislation (EPL)

In Table 3 we examine the demand for social protection from a different angle: preferences over health spending. We distinguish between countries with employment-independent health care systems (single-payer systems) and those where health benefits are partially tied employment (multipayer systems). We expect skill specificity to matter for health spending preferences only where benefits are tied to employment—that is, in multipayer not single-payer systems. We also expect higher aggregate support for health spending in single-payer systems: Where the government is the only “game in town,” all citizens must rely on it regardless of their employment risk.

Across all six models there is a strong and statistically highly significant conditional effect of skill specificity and type of health system. In employment-dependent multipayer systems, skill specificity is positively associated with health care spending preferences, whereas in employment-independent single-payer systems there is no estimated effect of skill specificity on health spending preferences. Moreover, in the online appendix we show that this remains true when using the 5-point preference scale as a dependent variable and if individual occupational unemployment risk is used instead of skill specificity. Finally, in the random effects models we see a sizable and statistically significant direct positive effect of single-payer health care systems on support for health spending, as hypothesized.

Figures 3a and 3b show the estimated effects of skill specificity on the probability of supporting increased health spending in single versus multipayer systems. Although all individuals, regardless of skill specificity, have an average probability of support of 83% in single-payer systems, in multipayer systems individuals with low levels of skill specificity have a support probability of 60% compared to 80% for those with high skill specificity. Finally, as hypothesized above, unlike the case of unemployment and industrial aid, Models 5 and 6 show that the unemployment rate affects neither the interactive nor direct effects of single-payer health systems.

Figure 3.

Health spending preferences and health systems

In Table 4 we provide further confirmation that it is the specific policies that condition social risk that affect the importance of individual skill specificity rather than other national institutions discussed in the comparative political economy literature. We conduct a series of two-stage analyses, first running logit models for the dependent variables for each country and then using the coefficient estimates as dependent variables. In the second stage analysis we examine the effects of a variety of national policy and economic variables on two sets of coefficients. First we examine the estimated coefficients on individual skill specificity—that is the effect of individual risk on policy preferences. Second we examine the constant term—which measures the “level” effects of the context variables—when all continuous individual-level independent variables are mean centered and all categorical variables to are set to zero, thereby reflecting the preferences of a “typical” citizen.¹⁵ We do this process in turn for unemployment insurance and health care spending.¹⁶ By adding a variety of differing context variables, we are able to determine whether the effects of EPL and of health care system are the national factors driving our results, or if other contexts have an important role in shaping individual risk and “typical” support for social spending.

Beginning with unemployment insurance, we see that EPL on its own is a strong and statistically significant negative predictor of the coefficient on skill specificity (as we found in Table 1) and that EPL has a positive direct effect on “typical” support for unemployment spending (using the constant term as the dependent variable). When we include single-payer health care system as a control, it has almost no impact on EPL’s estimated effect on either the skill specificity coefficient or the constant. This is not surprising given the low correlation of .09 between EPL and health care system, but it is reassuring that these policy institutions are not interchangeable and that health policies do not affect unemployment preferences. When we add the level of unemployment as a context variable, it does somewhat reduce EPL’s affect on skill specificity, but it is significant at conventional levels, whereas unemployment has no effect on the skill specificity coefficient. As in Table 1, we see that the direct effect of EPL is, however, reduced dramatically—consistent with our hypothesis that EPL’s level effect is through its relationship with aggregate unemployment.

Estevez-Abe et al. (2001) argue that the aggregate level of vocational training will produce increased demand for social insurance. Accordingly, we examine the proportion of secondary students in vocational streams to see if the composition of skills in the economy is driving our results. However, despite a high correlation of .42 with EPL, once again we see a significant effect for EPL and no effect of the vocational control, either on the coefficient for skill specificity or on the “typical” level for unemployment spending.¹⁷ Finally, Rehm (2011) argues that the inequality of unemployment across occupations produces varied demand for social insurance. Although we do have a reduced sample, we find no effect of Rehm’s “unemployment Gini” on either the effect of EPL on the skill specificity coefficient or the level effect.¹⁸

Table 4 repeats these analyses with health care spending as the dependent variable and single-payer health care as the main independent variable. Single-payer health care remains a significant predictor of both the skill specificity coefficient and the “typical” level of support, regardless of whether EPL, the unemployment rate, vocational education profile, or the unemployment Gini is included as a “rival” context variable. In the final model, there is some evidence that unemployment inequality accentuates the role of skill specificity in determining support for health care spending: This is consistent with our account of the interplay between individual and social risk since unemployment inequality heightens rather moderates the role of individual labor market status.

In summary, these results suggest a powerful conditioning effect of social risk on individual policy preference formation. Both EPL and employment independent benefits, by creating more uniform social risk, attenuate the importance of individual risk in shaping demands for social protection. Although we cannot entirely discount the role of endogeneity, the results in Table 4 showing that only the relevant policy context variable affected individual preferences, along with the fixed effects results, suggest that our results are not being driven by some unobserved national sentiment for social policy. Policy supply fundamentally shapes policy demand.

In Conclusion

The turn to examining the relationship between individual preferences and the structure of the welfare state has been a foundational step in comparative political economy. Although earlier works—for example Esping-Andersen (1990)—focused on the role of group action in demanding and securing social policies, research over the past decade has provided a salutary opening of the “black box” of preferences that lay implicit in many of these classic works. This literature provides an account of how national-level institutions might have emerged from the aggregation of individual characteristics such as skill specificity or occupational risk. However, something has been lost in this new program—the important conditioning effect of national policies on individual preferences. The “policy feedback” literature, for its part, reminds us that welfare states are not inert products but themselves shapers of political identification and policy demands.

The findings in this article provide a new way to link together the micro and macro literatures of policy preferences. We show that that national-level institutions affect preferences by altering the degree to which individuals experience economic risk. Where national institutions create more uniformity in risks and more dependence on the state, individual risk matters less in shaping preferences, and even those with low levels of individual risk are more supportive of the welfare state.

The article leaves open questions about the dynamic story of preference formation and welfare state change. We have argued that existing welfare regimes condition the importance of individual attributes in determining preferences. Indeed, we find that less extensive systems of social protection regimes accentuate the importance of individual risk in determining policy preferences. Given enough individuals with specific skills, this pattern could in the long run lead to realized demands for greater social protection. Or, conversely, as highly regulated labor markets reduce EPL, the politics of individual risk might become more salient in countries where it plays a more limited role today. Accordingly, scholars of institutional change might themselves want to turn to examine the micro dynamics of preference formation.

Footnotes

Acknowledgements

We would like to acknowledge Teri Caraway, Kathleen Collins, Silja Hausermann, Lisa Hilbink, David Rueda, Phil Shively, and the editors and reviewers of Comparative Political Studies for their extremely helpful comments, critiques, and clarifications. We would also like to thank Philipp Rehm for his comments and for his generosity in giving us access to his data.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Bios

Jane Gingrich is assistant professor of political science at the University of Minnesota. Her work focuses on comparative political economy and governance, particularly in Western Europe. She is the author of Making Markets in the Welfare State: The Politics of Varying Market Reforms (Cambridge University Press, 2011) and coeditor of Social Policy in the Small European States, forthcoming with Berghahn Books.

Ben Ansell is associate professor of political science at the University of Minnesota. His work broadly covers issues of inequality in comparative politics. He is the author of From the Ballot to the Blackboard: The Redistributive Political Economy of Education (Cambridge University Press, 2010) and articles on education policy in World Politics and International Organization. Together with David Samuels he has written on inequality and democracy, including an article published in Comparative Political Studies. He is also coeditor of Social Policy in the Small European States, forthcoming with Berghahn Books.

References

Baldwin

Peter

. (1990). The politics of social solidarity: Class bases of the European welfare state 1875–1975. Cambridge, UK: Cambridge University Press.

Blanchard

Olivier

Portugal

Pedro

. (2001). What hides behind an unemployment rate: Comparing Portuguese and U.S. labor markets. American Economics Review, 91, 187-207.

Bleckesaune

Morten

Quadagno

Jill

. (2003). Public attitudes toward welfare state policies. European Sociological Review, 19(5), 415-427.

Brambor

Thomas

Clark

William

Golder

Matt

. (2006). Understanding interaction models: Improving empirical analyses. Political Analysis, 14, 63-82.

Brooks

Clem

Manza

Jeff

. (2007). Why welfare states persist: The importance of public opinion in democracies. Chicago, IL: University of Chicago Press.

Cusack

Thomas

Iversen

Torben

Rehm

Philipp

. (2005). Risks at work. Oxford Review of Economic Policy, 22(3), 365-389.

Esping-Andersen

Gøsta

. (1990). The three worlds of welfare capitalism. Princeton, NJ: Princeton University Press.

Estevez-Abe

Margarita

Iversen

Torben

Soskice

David

. (2001). Social protection and the formation of skills: A reinterpretation of the welfare state. In Hall

Soskice

(Eds.), Varieties of capitalism (pp. 145-183). New York, NY: Oxford University Press.

Finseraas

Henning

. (2008). Income inequality and demand for redistribution. Scandinavian Political Studies, 32(1), 94-119.

10.

Hacker

Jacob

. (2004). Dismantling the welfare state? Political institutions, public policies and the comparative politics of health reform. British Journal of Political Science, 34, 693-724.

11.

Hellwig

Timothy

. (2001). Interdependence, government constraints, and economic voting. Journal of Politics, 63(4), 1141-1162.

12.

Iversen

Torben

Soskice

David

. (2001). An asset theory of social policy preferences. American Political Science Review, 95, 875-893.

13.

Jaeger

Mads Meier

. (2009). United but divided: Welfare regimes and the level and variance in public support for redistribution. European Sociological Review, 25(6), 723-737.

14.

Jordan

Jason

. (2010). Institutional feedback and support for the welfare state: The case of national health care. Comparative Political Studies, 43, 862-885.

15.

Korpi

Walter

Palme

Joakim

. (1998). The paradox of redistribution and strategies of equality. American Sociological Review, 63(5), 661-687.

16.

Lazear

Edward

. (1990). Job security provisions and employment. Quarterly Journal of Economics, 105(3), 699-726.

17.

Larsen

Christian Albrekt

. (2008). The institutional logic of welfare attitudes: How welfare regimes influence public support. Comparative Political Studies, 41, 145-168.

18.

Lewis

Jeffrey B.

Linzer

Drew A.

(2005). Estimating regression models in which the dependent variable is based on estimates. Political Analysis, 13(4), 345-364.

19.

Lindbeck

Assar

Snower

Dennis

. (1989). The insider-outsider theory of employment and unemployment. Cambridge, MA: MIT Press.

20.

Linos

Katerina

West

Martin

. (2003). Self interest, social beliefs, and attitudes to redistribution. European Sociological Review, 19(4), 393-409.

21.

Lynch

Julia

Myrskala

Mikko

. (2009). Always the third rail? Pension income and policy preferences in European democracies. Comparative Political Studies, 42, 1068-1097.

22.

Moene

Karle Ove

Wallerstein

Michael

. (2001). Inequality, social insurance, and redistribution. American Political Science Review, 95, 859-874.

23.

Nickell

Stephen

. (1997). Unemployment and labor market rigidities: Europe versus North America. Journal of Economic Perspectives, 11(3), 55-74.

24.

Organisation for Economic Co-operation and Development. (2004). Employment outlook. Paris, France: Author.

25.

Organisation for Economic Co-operation and Development. (2006). Employment outlook. Paris, France: Author.

26.

Organisation for Economic Co-operation and Development. (2008). Education at a glance. Paris, France: Author.

27.

Pierson

Paul

. (1996). The new politics of the welfare state. World Politics, 48(2), 143-179.

28.

Rehm

Philipp

. (2009). Risks and redistribution: An individual-level analysis. Comparative Political Studies, 42, 855-881.

29.

Rehm

Philipp

. (2011). Social policy by popular demand. World Politics, 63(2), 271-299.

30.

Rothstein

. (1992). Labor-market institutions and working-class strength. In Steinmo

Sven

Thelen

Kathleen

Longstreth

Frank

(Eds.), Structuring politics: Historical institutionalism in comparative analysis (pp. 33-56]). New York, NY: Cambridge University Press.

31.

Rueda

David

. (2005). Insider outsider politics in industrial democracies. American Political Science Review, 99(1), 61-74.

32.

Saint-Paul

Gilles

. (2000). The political economy of labour market institutions. Oxford, UK: Oxford University Press.

33.

Scheve

Kenneth

Stasavage

David

. (2006). Religion and preferences for social insurance. Quarterly Journal of Political Science, 1(3), 255-286.

34.

Solt

Frederick

. (2008). Economic inequality and democratic political engagement. American Journal of Political Science, 52(1), 48-60.

35.

Soss

Joe

Schram

Sanford

. (2007). A public transformed? Welfare reform as policy feedback. American Political Science Review, 101(1), 111-127.

36.

Svallfors

Stefan

. (1997). Worlds of welfare and attitudes to redistribution: A comparison of eight Western nations. European Sociological Review, 13, 283-304.