Abstract
How does the structure of social policy institutions shape the level of public support for the welfare state? The policy feedback literature predicts that highly inclusive welfare institutions generate larger bases of public support by shifting the focus away from redistribution and toward common market insecurities felt across classes, while more selective strategies erode support by highlighting the conflicts of interest imbedded in clearly redistributive social programs. This paper expands on existing research by adopting a disaggregated approach to measuring both welfare state structure and public support, uncovering important cross-program variations in public attitudes and welfare state design masked by traditional measures of universality and public support. This project applies this method to public opinion data in 17 advanced capitalist democracies across three policy areas: healthcare, pensions, and unemployment. The findings offer evidence of policy feedback effects.
Introduction
Recent years have witnessed a reawakening of interest in the factors that create cross-national variation in support for the welfare state. Aided by a number of new datasets that allow for comparisons across much larger samples of countries than previously possible, welfare state researchers have once again turned attention toward the interactions between public attitudes and welfare state design. Public opinion research has traditionally sought to identify individual level determinants of support for the welfare state or to link cross-national variations in ideological and cultural values to different patterns of welfare state politics. Scholars interested in the ‘new politics’ of the welfare state invert the conventional causal relationship between public opinion and the welfare state by seeing the structure of social policy institutions as an important independent variable that influences public support for welfare programs. Concerned with the resilience of social programs in the face of ongoing economic, demographic, and political pressure for retrenchment, the new politics perspective argues that social policy institutions once in place reshape welfare state politics by creating new political constituencies of welfare beneficiaries and altering the political discourse surrounding the welfare state. Depending upon how they incorporate citizens into core social programs, welfare states may build large constituencies in favor of further social policy development or generate hostility to further welfare state expansion (Korpi and Palme, 1998; Rothstein, 2002). Over time, these policy feedback effects may produce path dependencies in which generous and highly inclusive welfare states solidify strong bases of public support, while more targeted welfare states that emphasize poverty reduction and redistribution erode public support by intensifying the conflicts between beneficiaries and contributors.
Efforts to empirically demonstrate a link between the overarching structure of welfare states and public attitudes have met with mixed results (Andreß and Heien, 2001; Arts and Gelissen, 2001; Bean and Papadakis, 1998; Blekesaune and Quadagno, 2003; Edlund, 1999; Jæger, 2006; Larsen, 2008; Svallfors, 1997). The inconsistent results in existing research may stem from a reliance on welfare state typologies and indicators of public support that do not adequately account for the complexity of welfare state design or public attitudes. The standard approach seeks to locate variations in public support for redistribution across different welfare regime types as defined by Esping-Andersen’s (1990) ‘three worlds of welfare capitalism.’ This approach may not provide a sufficiently nuanced characterization of welfare state structure to account for important cross-national and within-country variations in attitudes toward specific sectors of the welfare state of relevance to policy feedback theory. In addition, the heavy dependence on attitudes toward redistribution as a measure of welfare state support may fail to capture the complexity of public attitudes toward particular aspects of the welfare state. Reliance on a single indicator of support for the welfare state may miss important variations in public support across different social policy sectors within countries. Support for redistribution may also be a particularly poor measure of public support for those social programs that are principally concerned with social insurance rather than overt redistribution. The combination of overly broad measures of both the independent and dependent variables combined with the theoretically questionable use of support for redistribution as a proxy for welfare support may generate a significant measurement error that prevents an accurate appraisal of policy feedback effects.
One strategy for overcoming these common theoretical and methodological problems is to disaggregate both the structure of the welfare state and public support by examining specific social policy sectors. Rather than seeing the welfare state and public attitudes as unified wholes, this approach links public support for specific social programs to the unique structural characteristics of those programs. This programmatic approach allows for the investigation of support for specific programs that may not perfectly fit into a country’s broad welfare regime type, avoiding the measurement error produced by overly broad regime categories. In addition, by emphasizing support for specific welfare programs rather than redistribution or broad welfare state support, disaggregation offers a more theoretically meaningful examination of the effects of welfare state structure on public attitudes.
The next section examines the existing theoretical and empirical work on policy feedback effects and public support for the welfare state, highlighting areas where there is a potential lack of connection between the theoretical predictions and the empirical models used to test them. The second and third sections develop an empirical model of policy feedback theory emphasizing programmatic variations in the structure of welfare state programs and patterns of public support. The final section applies this programmatic approach to public support for healthcare, pensions, and unemployment insurance in 17 advanced capitalist democracies. The project concludes with a discussion of the implications of this research for policy feedback theory and debates concerning welfare state politics in the current age of austerity.
Welfare state structure and public attitudes
Traditionally, research on public opinion and the welfare state has been principally concerned with how variations in ideological and cultural values across countries produce different models of the welfare state (Alesina and Glaeser, 2004; Brooks and Manza, 2006). Historical institutionalism challenges this approach by calling attention to the ways in which popular attitudes are themselves shaped by the institutional and policy environment. Skocpol’s (1992) work on the early development of survivors’ pensions in the United States identified a policy feedback effect in which policies, once in place, transform the surrounding political environment by reshaping the interests, identities, and political capabilities of interest groups and individuals (Thelen and Steinmo, 1992). By emphasizing the ability of policy institutions to ‘influence individuals perceptions about what their interests are, whether their representatives are protecting those interests, who their allies might be and what political strategies are promising’ (Pierson, 1993: 621), the policy feedback approach reverses the traditional cause and effect relationship between public attitudes and policy design. This new perspective demands greater exploration of the role policy institutions play in generating patterns of cross-national variation in public attitudes (Mettler and Soss, 2004).
One of the most clear applications of the insights of the policy feedback perspective to the study of welfare politics comes from Pierson’s (1994; 1996) influential work on the ‘new politics’ of the welfare state. Seeking to explain the tenacity of the welfare state in the face of significant economic, demographic, and political pressures for retrenchment, Pierson argues that the development of welfare state transformed the political environment in ways that made serious retrenchment difficult. Once in place, welfare programs generate new political constituencies of beneficiaries who become a significant obstacle to change, insulating the welfare state from reform. The creation of the welfare state thus reshaped welfare politics by creating new political alliances and altering how individuals interpreted their interests.
Though all social programs are expected to generate support among their beneficiaries, the structure of particular programs may have significant consequences for the size and nature of these new constituencies. Rothstein (1998, 2002) distinguishes between two distinct approaches to the incorporation of individuals into the welfare state. Selective welfare state strategies are designed to create a social safety net to ensure that those on the lowest rungs of the income ladder do not fall too far below some socially defined minimum standard of living. Also known as targeting, the selective strategy applies means-tested benefits to focus welfare state resources on the poor. In contrast, the universal welfare state provides benefits as a right of citizenship, emphasizing shared risks and common needs for social services across the income scale. If the selective welfare state is defined by means-testing and public assistance, direct government service provision and generous social benefits for the middle classes are the hallmarks of the universal welfare state.
Korpi and Palme (1998) argue that universal and selective welfare state strategies produce significantly different political dynamics by affecting the size of beneficiary populations and the relationship between beneficiaries and contributors to the welfare state. By their very nature, selective welfare states carve out relatively small constituencies of new welfare beneficiaries in comparison to more universal approaches. Furthermore, the means-testing of selective welfare states highlights the redistribution of income from rich to poor, drawing a clear distinction between net beneficiaries and net contributors to the welfare state. The selective welfare state thus presents welfare state politics as a zero-sum game in which middle and upper class families must pay into a system that only directly benefits poor and working class families. The universal welfare state produces a very different political dynamic. The emphasis of the universal welfare state on social services and benefits as a right of citizenship broadens the population of welfare state beneficiaries to include the middle classes. For example, the British National Health Service, which provides effectively free access to the healthcare system for all citizens, offers an immediate and tangible benefit to individuals across the income scale in a way that the targeted system of Medicaid in the United States does not. Although generous social provision clearly increases the total tax burden on the middle classes and involves significant cross-class redistribution (Organisation for Economic Co-operation and Development, 2008), it may be difficult in practice for middle-class families to determine whether they are net beneficiaries or net contributors to the welfare state. By blurring the lines between beneficiaries and contributors, the universal welfare state reduces the appearance of redistribution and shifts the focus of political debates away from zero-sum class conflicts toward a discourse of common solutions to common problems.
While Korpi and Palme focus on the larger constituencies of welfare beneficiaries created by universal programs, Rothstein (1998, 2002) argues that universalism generates popular support through a moral logic that moves beyond direct calculations of self-interest. Rothstein claims that universal welfare states are more popular because they are perceived to be more just. By singling out particular individuals for benefits while demanding contributions from others, selective welfare policies may offend public notions of both substantive and procedural justice. With the selective strategy, the state does not treat all individuals equally, but rather it adopts a clear choice to promote the welfare of some (the poor) over that of others (the middle and upper classes). By contrast, universalism fits more comfortably with public expectations concerning equal treatment and fairness by spreading both the benefits and the burdens of the welfare state more broadly. For Rothstein, the greater perceived fairness of universalism is what generates broader public support for more inclusive welfare states, rather than the immediate self-interests of potential welfare beneficiaries.
Though emphasizing different causal mechanisms, the work of Korpi and Palme (1998) and Rothstein (1998, 2002) both propose a policy feedback effect in which the structure of social programs reshapes welfare state politics in ways that may foster path dependency. By building broad, cross-class coalitions in favor of the welfare state, universalism may produce large constituencies of welfare state supporters resistant to retrenchment, effectively locking universal welfare states into place. Alternatively, selective strategies may undermine political support for the welfare state by reducing the number of beneficiaries and highlighting the conflict of interest between the middle class and the ‘transfer’ classes. In such systems, welfare state politics is likely to be characterized by the demands of middle-class voters to reduce the burdens of the welfare state by maintaining meager benefits and minimizing the size of the beneficiary population (Heclo, 1986).
Efforts to test empirically the prediction that universal welfare states generate higher levels of public support than more selective strategies have generally attempted to link variations in cross-national support for the welfare state to broad welfare state typologies such as Esping-Andersen’s (1990) influential Three Worlds of Welfare Capitalism. Esping-Andersen clusters welfare states into three broad regime types according to how they institutionalize different norms of social solidarity, justice, and the relationship between citizens, the state and the market. Scholars interested in policy feedback effects have attempted to demonstrate that patterns and levels of public support vary systematically across these regime types. These efforts to test policy feedback theory through application of the regimes approach have met with mixed results. A number of studies have found patterns of welfare state support across the worlds of welfare capitalism consistent with the predictions of policy feedback theory (Andreß and Heien, 2001; Edlund, 1999; Jakobsen, 2011; Larsen, 2008; McClosky and Zaller, 1984; Svallfors, 1997), while others find no significant evidence for a link between public attitudes and welfare regime type (Arts and Gelissen, 2001; Bean and Papadakis, 1998). In addition, efforts to demonstrate a relationship between welfare regime type and class polarization in welfare state attitudes have found limited evidence of distinct cleavage patterns across different welfare regimes (Edlund, 2007; Jæger, 2006; Larsen, 2008; McClosky and Zaller, 1984; Svallfors, 2004).
One potential cause for these mixed results may lie in the reliance of existing research on overly broad welfare state typologies that fail to account for the complexity of welfare state structure (Jæger, 2006). The reliance on Esping-Andersen’s typology of welfare regimes is problematic in that it may fail to capture the internal diversity of welfare state design. Though countries may fall into broad regime categories, specific sectors of the welfare state may diverge significantly from the overarching structure of the welfare state (Bambra, 2005). The British National Health Service and Canadian Medicare provide two prominent examples of how clearly universal programs may exist in otherwise targeted welfare states. If welfare states have significant internal variation in their structure, public attitudes may also vary across these policy areas. For example, it is possible that in Britain the universal healthcare system may enjoy significantly more popular support than more targeted sectors of the welfare state. The potential for cross-sectoral variation in welfare state design and public support cannot be captured by broad welfare state typologies.
A second problem with the welfare regimes approach lies in the typology itself. Esping-Andersen’s original formulation of the worlds of welfare capitalism clustered states into regime types based on a single snapshot in time, using data on welfare state design from around 1980. At the beginning of the 21st century, the worlds of welfare capitalism typology may no longer accurately describe cross-national variations in welfare state design. Scruggs and Allan (2006) challenge the worlds of welfare capitalism typology by demonstrating significantly less internal coherence to each of the three worlds of welfare capitalism, particularly in more recent years, than generally assumed. Using new and updated measures of welfare design and generosity, Scruggs and Allan find little evidence of states clustering into three distinct regime types. The internal variation within welfare states and the uncaptured policy change of intervening years have combined to undermine the descriptive power of standard regime classifications, producing potentially significant measurement error in the core independent variable.
A second cause for the mixed results in previous research may lie in the typical measures of welfare state support. Existing research on policy feedback generally measures welfare state support through levels of public support for redistribution (Andreß and Heien, 2001; Jæger, 2006; Svallfors, 1997) or more rarely an index measure of support for particular sectors of the welfare state (Larsen, 2008). This poses two problems. First, if significant internal variation exists within countries in how they approach different sectors of the welfare state, as proposed above, a single measure of welfare state support may not capture variations in public attitudes across different policy areas. Second, support for redistribution is particularly problematic because it measures only one limited aspect of public attitudes toward the welfare state. As suggested by the discussion above, questions concerning support for redistribution essentially probe public attitudes toward a particular type of welfare state program, namely targeted programs that emphasize redistribution from rich to poor. Though clearly redistribution is a significant part of all welfare state programs, many sectors of the welfare state such as pensions and healthcare may not be viewed through the lens of redistribution. What is more important, policy feedback theory argues that universal welfare state programs are more popular than targeted programs because they reduce the appearance of cross-class redistribution, effectively delinking attitudes toward redistribution from support for particular social policies. In other words, policy feedback theory is predicated on the argument that in universal welfare states support for redistribution is not a measure of support for the welfare state. Accordingly, support for redistribution is not an appropriate measure of welfare state support as it measures only one element of welfare state attitudes and assumes a link between support for redistribution and support for the welfare state explicitly challenged by policy feedback theory.
One approach to resolving these common methodological problems is to disaggregate the welfare state into different policy areas in order to connect the structure of specific policy institutions to support for those institutions. This approach eliminates the need for redistribution as a measure of welfare support and avoids the problems associated with overly broad and problematic measures of welfare regime type. Jordan (2010) applied this strategy to support for national healthcare in a limited sample of countries, finding a link between the structure of healthcare institutions and support for national healthcare. This project extends on this research by substantially increasing the number of cases under consideration and examining two additional areas of the welfare state: pensions and unemployment insurance. The next section discusses how the distinction between universal and selective welfare states is operationalized across different sectors of the welfare state.
Universalism across different sectors of the welfare state
The discussion above suggests that variations in the structure of different welfare state programs within countries not captured in standard welfare typologies may generate different patterns of public support for particular programs. To examine the possibility that the structure of particular programs produce unique patterns of public support within and across countries, it is important to first isolate important policy fields and then to determine how the distinction between targeted and universal welfare strategies is put into practice in these policy sectors. Because different sectors of the welfare state involve different constituencies, offer different types of benefits, and protect against different sets of risks, universal and targeted strategies may look very different within particular policy areas. As a result, different sectors of the welfare state may require unique typologies that capture how the principles of a universal or selective welfare state strategy are applied in that area.
Of the three policy areas discussed here, healthcare poses the most significant measurement problem due to the sheer size and complexity of the healthcare sector and national healthcare systems. Unlike most cash benefit programs, national healthcare involves a complex set of interlocking policies regulating a wide range of issues including private and public insurance options, doctor payment schemes, pharmaceutical and medical technology pricing and availability, coverage lists, facilities construction and maintenance, and certification of medical professions. The size and complexity of national healthcare systems defies easy categorization (Freeman, 1999; Wessen, 1999).
Despite the incredible complexity of healthcare as a policy field, it is possible to distinguish between two broad strategies of incorporating citizens into the national healthcare system of relevance to policy feedback theory. Hacker’s (2004) distinction between hierarchical and decentralized healthcare systems offers a useful starting point. Hierarchical healthcare systems adopt a single payer approach in which the government becomes the dominant purchaser or provider of healthcare. In this model, all individuals in society are incorporated into a single healthcare system financed directly by the government through general tax revenue without regard for their class status or ability to pay. The hierarchical healthcare system is perhaps the most quintessential example of the universal welfare state strategy.
While hierarchical healthcare systems directly purchase or provide care, decentralized healthcare systems focus on improving or guaranteeing citizen access to health insurance in highly regulated markets. Decentralized models are significantly more diverse than hierarchical healthcare systems. In social insurance systems like those of Germany, the government requires all individuals to enroll in non-profit insurance funds organized by occupational class. The social insurance model ensures widespread access to the healthcare system while allowing both benefit levels and contribution rates to vary across different occupational and income groups. An alternative strategy, adopted by countries such as the United States and Switzerland, relies more heavily on the promotion of private health insurance to ensure greater access to the healthcare system. Though decentralized systems may vary significantly in their internal design, what binds them together and distinguishes them from hierarchical healthcare systems is that individuals are enrolled into distinct insurance programs, which may vary in their contribution rates and benefit levels. Unlike universal systems then, decentralized systems do not unite all citizens into a single, universal healthcare system, but instead divide citizens into distinct beneficiary groups and emphasize individual responsibility for securing health insurance.
The distinction between hierarchical and decentralized healthcare systems is measured as direct government spending on healthcare, excluding social insurance, as a percentage of total healthcare spending (public and private). This measure distinguishes between highly centralized systems in which direct government spending dominates the healthcare market and decentralized systems where multiple payers including social insurance funds and private health insurance companies are more significant. This continuous measure captures the difference between hierarchical and decentralized healthcare systems, while also allowing for a few hybrid models like those of Australia, Ireland, and Portugal in which large, single-payer healthcare systems operate alongside a more decentralized healthcare system for some segment of the population. 1
The measurement of universalism for pensions and unemployment insurance is significantly more straightforward than that of the healthcare system. The essential claim of policy feedback theory is that welfare state programs are more popular when they incorporate the middle classes and reduce the appearance of being targeted at the poor. Unlike in healthcare, which concerns the provision of a service, in cash transfer programs, the universalism is not synonymous with the provision of an identical benefit to all income classes. As the Swedish Social Democrats famously came to realize in the 1950s (Baldwin, 1990; Esping-Andersen, 1992), universal benefits that provide a meager payout to all income groups are not sufficient to induce middle-class participation in or support for the welfare state. For welfare programs to be considered as more than mere redistribution to the poor, they must offer sufficiently generous replacement rates to ensure that middle-class families will view them as forms of social insurance against market insecurities. Welfare state programs that fail to scale-up with income may be functionally and politically equivalent to targeted welfare state programs because they do not offer sufficient benefits to the middle classes to offset their costs. As such, whether the middle class may formally access cash benefit programs is less relevant to the question of inclusivity than whether the programs are structured so that the middle classes actually see a direct and meaningful benefit from them. For cash benefit programs, universality engages both a question of equal access and of benefit generosity.
Scruggs and Allan (2006) provide perhaps the clearest indicator of universalism understood as a combination of both universal access and sufficiently generous benefits in their generosity indexes. The generosity indexes score countries on the basis of how they compare to other advanced industrial democracies on core components of the benefit system including the net replacement rates, the percent of the population covered by the program, and any limits to participation to create a single measure of the generosity of both pension and unemployment insurance systems. The inclusion of both the percentage of the population covered and the replacement rate provide important information concerning the degree to which the middle classes may be covered and meaningfully benefit from the system. As such, the generosity indexes capture the essential question of the extent to which middle-class families are incorporated into the pension and unemployment insurance systems.
Measurement, methods, and data
The previous section developed a scheme for measuring the universalism of particular sectors of the welfare state that account for the potential differences created by different policy areas. An empirical test of the predictions of policy feedback theory must link these measures of inclusion to public support for particular welfare state programs. The International Social Survey Programme’s (ISSP) 2006 Role of Government IV survey provides the necessary data on public attitudes toward particular sectors of the welfare state. The ISSP interviewed individuals across 33 countries concerning their attitudes towards the responsibility of government for managing the economy and providing social protection. Seventeen countries met the initial classification criteria of advanced industrial democracies in which sufficient data were available: Australia, Canada, Denmark, Finland, France, Germany, Ireland, Japan, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, the United Kingdom, and the United States. Data limitations in the generosity indexes require excluding Portugal and Spain in the analysis of pension and unemployment insurance, reducing the number of cases under consideration in these two policy areas to 15.
Support for each sector of the welfare state is measured from a single response question that asks respondents to determine whether each policy area is the responsibility of government. For example, for unemployment insurance respondents were asked, ‘On the whole, do you think it should be or should not be the government’s responsibility to provide a decent standard of living for the unemployed?’ Respondents were given four potential responses to the question ranging from ‘Definitely should be’ to ‘Definitely should not be.’ For all three policy areas, strong opposition was uncommon in the sample, with less than 1 percent of respondents expressing strong opposition in the cases of pensions and healthcare. Given the difficulties associated with estimating rare events in probit models (Hamilton, 1992), the bottom two categories were collapsed into a single opposition category. The support variables thus contain three values that represent opposition, moderate support, and strong support for each welfare sector. 2
In addition to the indicators of universalism discussed above, the analysis includes social expenditure as a percent of GDP as a country level control. This variable controls for the overarching size and generosity of the welfare state, which may be correlated with programmatic variations in the structure of particular social programs. This helps to ensure that the effects of programmatic differences are not merely reflections of the broader structure of the welfare state, providing further support to the suggestion that support for welfare state programs can vary independently based on the structure of those programs.
When examining unemployment insurance, it is important to consider the role of structural differences in the economies of states that might influence public attitudes toward the unemployment system. In particular, cross-national public opinion research suggests that support for unemployment insurance grows during periods of high unemployment (Blekesaune, 2007; Blekesaune and Quadagno, 2003). High unemployment is thought to increase public support for unemployment insurance both by increasing the percentage of the population dependent upon benefits and by fostering a belief among the public that unemployment is caused by structural features of the economy outside the control of individuals. To control for this possibility, the models of unemployment insurance contain a control for the unemployment rate in the country at the time of the survey.
At the individual level, one of the most important and consistent determinants of welfare state support is income and class status (Edlund, 2007; Jæger, 2006; Svallfors, 2004). Class status is measured according to the European Socio-Economic Classification (ESeC) system (Erikson and Goldthorpe, 1992; Rose and Harrison, 2007). The ESeC typology provides an internationally comparable measure of class based on individuals’ occupational status. The analysis here utilizes Rose and Harrison’s (2007) three-category scheme, which differentiates between the ‘salariat’ (large employers, managers, and high end professionals), the intermediate (self-employed, white and high-grade blue-collar workers), and the working class (lower white-collar and skilled and unskilled manual laborers). 3 Married respondents were assigned the higher class status associated with their or their spouse’s occupation. The analysis uses a dummy variable for each of the upper two classes relying on the working class as the reference category. Each of these two variables is expected to be negatively correlated with support for the welfare state.
In addition to the class variables, the analysis includes a measure of the respondent’s household income. Unfortunately, the ISSP income measure does not allow for easy comparison across countries. For each country, income is coded using different scales, numbers of categories, and top and bottom coding conventions. To partially correct for this problem, income was standardized within country so that the mean income in each country is set to zero and the variable reflects the number of standard deviations away from the mean. Given that higher income individuals are expected to receive fewer net benefits from the welfare state, income should be negatively associated with support for the welfare state.
In addition to these more central independent variables, the analysis includes a number of additional individual level controls. Standard controls include sex, age, and education as they may affect individuals’ ideological perspective or their interpretation of their relationship to the welfare state (Edlund, 1999; Svallfors, 1997). Education is measured on a five point-scale assessing the highest level of education attained by respondents at the time of the survey and ranges from no formal education to the completion of a university degree. Dummy variables for retirees and those who are unemployed or have an unemployed spouse help control for ‘transfer classes’ that may be current welfare recipients. Finally, current union membership is used to further control for class and ideological differences between union and non-union members.
The analysis of cross-national survey data poses a unique challenge due to the multi-level structure of the data. Because individuals within the same country are likely to be more similar to one another than to those in different countries, cross-national survey data violates the standard assumption that observations are independent. Failure to correct for this clustering of respondents by country may artificially depresses standard errors (Steenbergen and Jones, 2002). Cluster-robust standard errors provide the most simple and direct correction for this problem. As with other methods for correcting for clustering, cluster-robust errors may also produce artificially low standard errors in cases where the number of upper-level cases (countries) falls below 30. Following Franzese (2005), the analysis applies a small-sample correction to reinflate the standard errors, minimizing the additional risks of Type I errors produced by using cluster-robust standard errors with a relatively small number of upper-level cases. 4
Finally, the ordinal nature of each of the independent variables demands the use of ordered probit. Ordered probit is an extension of the standard probit model that allows for the existence of multiple values of the dependent variable through the calculation of unique intercepts (or thresholds) for each category of the dependent variable (Daykin and Moffatt, 2002). Unfortunately, ordered probit results are often difficult to interpret. In order to simplify the presentation of results, where appropriate the analysis includes predicted probabilities of different levels of welfare state support using model coefficients and their standard errors (King et al., 2000).
Results
Do more inclusive social programs produce higher levels of public support for the welfare state? Before examining the ordered probit models, the analysis begins with a comparison of the mean level of popular support for each welfare state sector to the corresponding measure of inclusiveness across the sample of countries. The first column of Table 1 presents the correlations between welfare state structure and mean levels of national support for each policy sector. The evidence from column 1 conforms to the expectations of policy feedback theory. In each of the three policy areas, the indicators of inclusion are positively correlated with support for that sector of the welfare state. Despite the relatively small sample size, the correlations coefficients are high and statistically significant at the 0.05 level. This simple bivariate analysis provides clear evidence of a link between the structure of particular welfare state programs and overall public support for those policy areas.
Cross-national comparison of mean support for the welfare state and indicators of universalism.
Notes: Correlation coefficients comparing cross-sectional mean values. P values are in parentheses.
Correlations with mean levels of support for the corresponding welfare state sector.
Table 1 also presents the correlations between each of the measures of universalism and mean support for redistribution across the sample of countries. Support for redistribution was measured for each country by summing individual responses to a question asking respondents whether they believed it was the responsibility of the government to redistribute income from the rich to the poor. The measures of universality in the healthcare and pension sectors are not statistically correlated with mean levels of support for redistribution. This evidence strengthens the claim that the links between support for these programs and their structure is not a simple reflection of cross-national variations in attitudes about redistribution, which may produce more generous welfare states. Unlike the other two sectors of the welfare state, the structure of the unemployment system does appear to be correlated with redistribution attitudes (r = 0.44), although it narrowly fails traditional standards of statistical significance (p = 0.104). Given the more clearly redistributive nature of unemployment insurance when compared to pensions and healthcare this finding is not surprising, but it suggests the need for further analysis.
For a more complete test of policy feedback theory, the analysis now turns to the ordered probit models. Table 2 presents the results from the ordered probit models for each of the three sectors of the welfare state. As with the bivariate correlations above, the results conform to the expectations of policy feedback theory. Each of the three measures of inclusiveness is positively and statistically significantly related to support for the corresponding welfare state sector. In other words, all else held equal, more inclusive welfare state programs are associated with higher levels of public support. Though the models are all statistically significant, it is not easy to determine from these results the substantive relevance of institutional structure on popular attitudes.
Determinants of support for specific welfare state programs.
Notes: All results from ordered probit analysis using cluster-robust standard errors. Standard errors corrected for small upper-level sample size in parentheses (see note 4 for full description).
p<0.05, **p<0.01, ***p<0.001.
To explore the substantive impact of institutions on support for the welfare state, Table 3 presents the predicted probabilities of support for each welfare state sector for a ‘typical’ individual in both selective and universal systems. 5 For the purposes of this comparison, for each policy area, the representative case for the selective welfare state was defined as the country scoring in the 25th percentile of the corresponding measures of inclusion. Similarly, the representative case for the universal welfare state was defined as the country scoring in the 75th percentile on each measure of inclusion. For example, in the case of unemployment insurance, this strategy amounts to comparing the universal system of Denmark to the more selective unemployment insurance system of Britain. This approach mirrors that of the regimes literature by comparing countries that are clear representatives of their regime types, while avoiding the comparison of the most extreme values.
Notes: p>Predicted probabilities based on Models 1, 3, and 5 presented in Table 2, assuming all dichotomous controls set to median values and continuous controls set to means.
‘Universal’ defined as a country in the 75th percentile of each measure of inclusion.
‘Selective’ defined as a country in the 25th percentile of each measure of inclusion.
‘Institutional effect’ is defined as the difference in predicted support between universal and selective systems. Upper and lower bounds of the 95% confidence interval are in parentheses.
For both pensions and healthcare, the opposition category represents a relatively small proportion of the total population. As a result, in these two policy areas, the main effect of inclusive welfare state institutions was to shift more individuals from the moderate to the strong support categories. For pensions, a shift from a selective to a universal system is expected to increase the probability of strong support by 10 percentage points. For healthcare, the effect is nearly twice as large with the universal healthcare systems increasing the probability of highest support by 19 percentage points. Though support for pensions and healthcare remain high in all countries, universalism does appear to significantly increase the number of those strongly committed to these programs.
Support for unemployment insurance is much more contentious than that of the other two policy sectors, with a much greater number of respondents expressing opposition to the program. In the case of unemployment insurance, universalism increases the predicted probability of the highest level of support by 12 percentage points. Universalism is also expected to reduce the predicted probability of opposition to unemployment from 0.42 under the typical selective system to 0.24 in a more universal model. When scaled up to the full population, these results suggest substantial differences in overall support for unemployment insurance in the two systems.
The relationship between the measures of inclusion and support for the welfare state are independent of both the structural and individual level controls. Social expenditures as a percent of GDP performed poorly as a predictor of welfare state attitudes, suggesting that the structure of individual welfare state programs are stronger determinants of welfare state support than the overall size and generosity of the welfare state. The unemployment rate was found to have a statistically significant effect on support for unemployment insurance, but it did not overwhelm the substantive or statistical importance of institutional design. The individual level controls all generally conformed to expectations. The class variables demonstrated that the upper and middle classes were less likely to support each welfare state institution than those in the working class. This class effect is further confirmed by the consistent negative relationship between income and welfare state support as well as the positive effect of union membership on welfare attitudes.
The strong statistical relationship between the three measures of inclusiveness and support for the welfare state in the face of both individual and structural controls provides solid evidence of policy feedback effects. In each of the three policy areas, more inclusive welfare state strategies are associated with higher levels of public support for the welfare state independent of the overall size and generosity of the welfare state. In addition, the predicted probabilities demonstrate that these institutional differences have substantively meaningful effects on the levels of support across countries.
Conclusions
Policy feedback theory draws attention to the role policy institutions play in shaping public attitudes. Working from this perspective, scholars interested in the new politics of the welfare state argue that social programs once in place become integral parts of a new status quo, fundamentally transforming welfare state politics by creating new political constituencies and shaping how the working and middle classes interpret their relationship to the welfare state. Yet, the effect of the welfare state is not uniform, but instead depends upon how it incorporates citizens into core benefit programs. Highly inclusive welfare institutions generate larger bases of public support by shifting the focus of the welfare state away from redistribution and toward common market insecurities felt by both the middle and working classes, while more selective strategies erode support by making clear the conflicts of interest imbedded in clearly redistributive social programs. Over time, these political dynamics may produce path dependencies in which universal welfare programs become locked in place by their broad based political support, while selective programs become increasingly targeted in order to limit the burden of redistributive programs on middle class voters.
This research sought to overcome common methodological problems associated with empirical tests of policy feedback theory in order to identify the proposed links between the structure of welfare state institutions and popular attitudes. This disaggregated approach connected popular support for particular sectors of the welfare state to the unique structure of those programs. The analysis revealed solid evidence that more inclusive welfare state strategies produce higher levels of public support for the welfare state. Even after a controlling for both individual and cross-national factors, the evidence suggests that universalism builds larger bases of support for the welfare state than more targeted strategies.
The results provide further evidence in favor of the claims of the ‘new politics’ school that the welfare state has become an important independent variable by altering public attitudes and influencing political preferences. These findings have implications for the analysis of the current age of austerity by suggesting that the vulnerability of particular welfare state programs to retrenchment is influenced by how they incorporate citizens. More selective social programs may be at higher risk of retrenchment compared to more universal programs that have significantly larger bases of public support. What is important is that the disaggregated approach adopted here suggests that these effects are not likely to be uniform within countries as suggested by research which relies on overly broad measures of welfare regime type; rather, variations in program design may produce unique political dynamics across each sector of the welfare state within countries.
Finally, the findings presented here support the claims of historical institutionalist scholars of a policy feedback effect whereby welfare politics evolves over time as early policy choices influence later political conflicts and policy debates. Although this research is limited in scope, it is consistent with a story of welfare state politics that sees the political preferences of voters and the structure of welfare state institutions as both being endogenously reshaped by one another in a complex political dynamic not captured by models of welfare state politics which assume a static relationship between the political preferences and social policy institutions.
Footnotes
Funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.