Decomposing the effect of social policies on population health and inequalities: An empirical example of unemployment benefits

Abstract

Aim: The purpose of this study is to discuss and empirically contrast different conceptualizations and operationalizations of social policies in analysing health and educational differences in health cross-nationally. Methods: Country-level institutional and expenditure data on unemployment benefit schemes and individual-level data from the EU-SILC for 23 countries were used to analyse the association between unemployment benefits and self-assessed health for individuals with different educational attainment. Results: The analyses indicate that higher coverage rate (i.e. the proportion of the relevant population eligible for benefits) is associated with better self-related health among both low- and high-educated individuals, but is not linked to smaller educational differences in health. In contrast, replacement rate (i.e. the amount of benefits received) in isolation is not related to self-assessed health. However, in countries where coverage rates are high, higher replacement rates are associated with better health among both low- and high-educated individuals and smaller educational differences in health. Conclusions: Decomposing unemployment benefit programmes into two main dimensions – the proportion in the labour force covered by such programmes and the replacement rate received in case of unemployment – may present further insights into institutional mechanisms linking macro-level social policies to individual-level health outcomes.

Keywords

coverage institutional approach replacement rate self-assessed health unemployment benefits welfare state

Introduction

In the last decades, increased attention has been directed towards structural or macro-explanations of health inequalities at the individual level. The role of the welfare state and social policies has quite naturally come to occupy a prominent place in this line of research. Recent reviews suggest, however, that the evidence for such policies having a positive effect on population health and reducing socio-economic differences in health is rather inconsistent [1]. One possible reason for this is that the conceptualization and operationalization of the central independent variable, the welfare state and social policies display considerable variation between studies. Bergqvist et al. [1], building on Dahl and van der Wel [2], argue that studies analysing the impact of welfare states and social policies on health and health inequalities can be divided into three groups based on the way in which such policies are conceptualized and operationalized.

The regime approach, owing much of its current popularity to the path-breaking work of Esping-Andersen [3], identifies clusters of qualitatively distinct welfare-state types. At the heart of this approach is the idea that there exist institutional complementarities between different social policies, and that the social policy model of a particular country therefore is something more and qualitatively different to the sum of individual policies and programmes – it is the result of a complex interaction between historical and political driving forces, concrete policies and institutions and outcomes. In this way, such an approach provides a valuable summary of different social policies including the social, economic and historical factors that have produced these policies and their effect on a wide range of living conditions. However, when social policies are captured in broad and analytically fairly blunt groupings of nations in welfare clusters, rather than in terms of specific programmes and policies, it is difficult to draw theoretically and empirically valid conclusions about the mechanisms linking macro-level characteristics to individual-level outcomes [4]. This problem is further aggravated by the fact that different programmes and policies are often organized along different principles within countries, and countries belonging to the same welfare state model may therefore display a high degree of variation with regard to how specific programmes are organized.

The expenditure approach operationalizes the welfare state and social policies in terms of public spending on different programmes and services. The main advantage of this approach is that public data are readily available from international organizations such as the OECD and the EU for a relatively large number of countries and years. Moreover, in many instances, economic resources in the form of public expenditure are indeed important for the amount and quality of services and benefits provided by the welfare state. However, the same amount of economic resources is often spent in very different ways: on benefits and services for a relatively small and privileged group in society, or on increasing the availability and coverage of services and income maintenance schemes to broader population groups. These two approaches are likely to have different effects on population health and socio-economic inequalities in health: whereas the first approach will probably do little to address such inequalities, empirical evidence suggests that the universality (or coverage) of welfare policies is a key dimension in this respect [5].

The third approach, the institutional approach, focuses on the institutional organization of specific policies and how this affects population health. With regard to income maintenance programmes, such organizational features include, for example, eligibility criteria, the duration of benefits, the amount of benefits received (replacement rate) and the proportion of the relevant population eligible for benefits (coverage rate). It can be argued that such organizational features can be summarized into two central dimensions: coverage and replacement rate [6]. This approach enables researchers to separate the effects of different organizational features on population health. An additional advantage with this approach is that it is sensitive to the fact that different programmes are often organized along different principles within countries.

With this as a background, the aim of the present article is to contrast the institutional and expenditure approach when analysing self-assessed health and educational differences in self-assessed health among the working-age population in 23 European countries. More specifically, we analyse the association between two central aspects of unemployment benefit programmes – replacement and coverage rates – and health and compare these results to an approach that conceptualizes and operationalizes support to the unemployed in terms of real expenditure per capita.

Why, then, should the coverage and replacement rate of unemployment benefits matter for the subjective health of individuals? First, and perhaps most obviously, such benefits may be used to transfer economic resources from rich to poor (or from those who have work to the unemployed), which may be converted into goods and services that are beneficial in improving and maintaining good health or avoiding health hazards. Second, unemployment-benefit schemes may also affect inequality in society and the psychological consequences associated with being relatively deprived. Unemployment is often associated with loss of social status, and by reducing inequalities between those in and out of work, unemployment benefits may reduce the stress reactions associated with unemployment as well as increase unemployed individuals’ sense of autonomy and control over their lives [7,8]. This idea may also be linked to the classical writings on social citizenship by TH Marshall [9], who argued that universalism is closely connected to class solidarity and solidarity of the whole nation. Social policies providing protection for all against major risks, such as the risk of income loss during times of unemployment, could then be considered as a form of psychological mechanism through which a sense of community and solidarity is constructed. Moreover, narrowly targeted policies require the identification of the “poor” or “needy” and therefore also often create feelings, on the part of the recipients, of being stigmatized, with resulting loss of self-respect and self-esteem. Loss of self-respect and self-esteem in turn has been shown to affect health adversely [10]. Third, job insecurity is an important stressor and as such associated with a range of adverse health outcomes [11]. Unemployment benefits may lessen the perceived economic consequences of job loss and also the negative effects of job insecurity on employed individuals’ subjective health. Unemployment benefits may therefore contribute to both stability and predictability for both individuals and society at large. Such benefits might be viewed as a collective resource that does not need to be utilized in order to have a welfare-enhancing effect; the mere knowledge that they exist may reduce the negative consequences of insecurity [11].

Accordingly, it could be argued that increased coverage of unemployment benefit schemes will also be associated with better self-assessed health at low replacement rates since it will provide the unemployed with at least a minimum level of material resources, and possibly because it creates bonds of solidarity and avoids stigmatizing those in need of such benefits. However, to ensure that individuals can uphold a socially acceptable standard of living, the drop in income and consumption cannot be too large during times of unemployment. This implies that coverage rates will have the largest impact on health when it is combined with high replacement rates and, hence, the existence of an interaction effect between coverage and replacement rates.

The present study is, to our knowledge, the first to analyse the association between both the separate and combined effects of unemployment replacement and coverage rates and self-assessed health cross-nationally. There are however other studies that are relevant to this purpose. Veenhoven [12] and Ouweneel [13] find no evidence of an association between overall social insurance expenditure and subjective health in samples of 42 and 41 nations, respectively, neither among the general population nor among the unemployed. Bambra and Eikemo [14] find that the relationship between unemployment and health in Europe varies by welfare state regime, suggesting that levels of social protection may indeed have a moderating effect. Nordenmark et al. [15], analysing mental well-being among unemployed in Sweden, Ireland and Great Britain, find that type of unemployment benefit received is an important determinant of mental distress with income replacement benefits being more beneficial than flat rate benefits. According to Rodriguez [16], means-tested benefits do not seem sufficient to reduce the impact of unemployment on health status in Britain, Germany and the United States, and interprets this in terms of the stigma associated with means-tested benefits that may add more stress to the recipients and exacerbate their vulnerability to health deterioration.

Material and methods

Data on replacement (RR) and coverage (Cov) rates for unemployment benefits come from the Social Citizenship Indicator Program (SCIP), which provides type-case benefit and coverage data on state-legislated unemployment benefits [17]. To allow for comparisons across countries, replacement rates reflect the benefits an average worker would receive in case of unemployment. This average worker is 30 years of age, has worked for 10 years, has not been unemployed for the last two years (before the present unemployment spell) and earns an average production worker’s wage. Benefits levels are expressed as net replacement rates, that is, the ratio between the net benefit and the after-tax wage. Expenditure on unemployment benefits is from Eurostat [18] and is measured as purchasing power parities per capita. Dahl and van der Wel [4] show that the operationalization of expenditure has very little effect on the results when analysing educational differences in self-assessed health cross-nationally. Nominal GDP in purchasing power standards per capita is used as a control variable at the national level to remove the effect of national differences in wealth. All country-level variables refer to the year 2005. Basic descriptive statistics is provided in Table I.

Table I.

Descriptive statistics (proportions, mean and range).

	Individual-level variables		Country-level variables
Education	Primary	23.9	Coverage, mean (range)	73.0 (50–100)
	Secondary	53.1
	Tertiary	23.0	Replacement, mean (range)	61.6 (17–92)
Age, mean		37.2
Gender	Men	48.4	Unemployment benefit	293.8 (23–917)
	Women	51.6	expenditure, mean (range)
Birth country	Country of residence	93.2
	Other EU-country	1.5	GDP, mean (range)	20,709 (5700–52,600)
	Country outside EU	5.3
Household type	Single	8.4
	Couples, no children	23.6
	Single, children	5.3
	Couples, children	62.7
Activity status	Employed	70.4
	Unemployed	6.7
	Retired	1.2
	Inactive	21.7
Able to face unexpected expenses	Yes	65.2
	No	34.8
Number of observations		175,367		23

Individual-level data come from the European Union Statistics on Income and Living Conditions (EU-SILC) cross-sectional national surveys from 2006, and includes 23 countries (Austria, Belgium, Czech Republic, Germany, Denmark, Estonia, Spain, Finland, France, Greece, Hungary, Ireland, Italy, Lithuania, Latvia, the Netherlands, Norway, Poland, Portugal, Sweden, Slovenia, Slovakia and the United Kingdom; Bulgaria, Cyprus, Iceland, Luxembourg and Malta are excluded because of missing data on one or more macro-level variables). Implementation and data confidentiality of the EU-SILC is regulated by regulation EC 1177/2003. Analyses are restricted to respondents between 18 and 54 years of age.

The dependent variable is self-assessed general health, constructed from a question asking respondents to rate their health on a 1–5 scale. This variable was dichotomized into “very good or good” health (assigned the value 1) versus “less than good” health (“fair”, “bad” and “very bad”). Self-rated health is a strong predictor of mortality and other health outcomes even after controlling for a wide variety of health-related measures that cover medical, physical, cognitive, emotional and social status [19], although this predictive power has been shown to vary somewhat between socioeconomic groups [20]. However, it should be pointed out that our understanding of this indicator and what it exactly measures and why it has such a strong and constant association with mortality is rather unclear [20]. There are also some well-known validity problems with this indicator in comparative research, such as the fact that groups may have different reference levels to which they compare their own health [21].

Educational attainment (Edu) is based on ISCED 1997, and the original variable has been re-grouped into: (i) lower secondary education or below (ISCED 0–2); (ii) upper secondary and post-secondary non-tertiary education (ISCED 3–4); and (iii) tertiary education completed (ISCED 5–6) [22]. For convenience, these groups will be referred to as primary, secondary and tertiary education. For illustrative purposes, we will focus only on respondents with primary and tertiary education. Based on previous research on predictors of self-rated health [23] and data availability in the EU-SILC, the following individuals covariates were controlled for in all models to account for composition and confounding effects: age; sex; birth country (distinguishing between native-born, immigrants from an EU-country, and immigrants from a non-EU country); household type (distinguishing between single, couples with no children, couples with children and single with children); activity status (distinguishing between those who are employed, unemployed, retired or inactive for at least six months during the last year); and whether the respondent believes that the household can face unexpected financial expenses. Scaled weights are used to correct for selection bias [23].

Logistic regression analysis with cluster-robust standard errors are used to correct for the fact that observations in the data are nested within countries and therefore not independent. To assess whether unemployment benefits modify the association between education and health, cross-level interactions between the individual-level variable education and the country-level variables measuring different aspects of unemployment benefits schemes (expenditure, replacement rate and coverage) were estimated (controlling for other individual-level covariates and GDP at the country level). With interaction terms, one has to be very careful when interpreting any of the terms involved in the interaction. In a simple linear-additive regression, the effect of a variable x on the dependent variable y is simply its coefficient, β_x. However, the interaction terms of interactive models have multiple effects, and to assess properly the full effect of such interaction terms, we must simultaneously consider how they affect the coefficients of the single variables that comprise the interaction terms, as well as how the coefficient of the interaction term depends on the other values of the other variables with which it interacts. To assess the full effect of the interaction variables, we will thereafter graphically present predicted probabilities for low- and high-educated individuals.

Equation 1 describes the model when testing the interaction effect between education and unemployment benefit expenditure (Figure 1) and coverage and replacement rate (Figure 2), where UB stands for these three dimensions of unemployment benefits schemes.

L n (\frac{P}{1 - p}) = a + β_{1} E d u + β_{2} E d u \times U B + β_{3} U B + β_{4} G D P + β_{i} I n d . C o v a r i a t e s + ϵ

(1)

Although most textbooks argue for the inclusion of all constitutive terms when specifying multiplicative interactive models, this advice should be weighed against the relatively small number of countries available in the present analysis. In the final analysis, when testing the interaction effect between education, coverage and replacement rates (Figure 3 and Figure 4), two alternative models were estimated: one containing all constitutive terms (Equation 2) and one containing only the three-way interaction term between education, replacement levels and coverage and the main effects of these variables (Equation 3). Goodness-of-fit statistics indicates that Equation 2 performs significantly better than Equation 3: the reduction in the chi²-value when including all constitutive terms is 60.5, which is significant at the 0.1%-level. The presentation of the results will therefore be based on estimates from Equation 2, and differences in results between Equation 2 and Equation 3 will be discussed in the text.

L n (\frac{P}{1 - p}) = a + β_{1} E d u + β_{2} E d u \times C o v \times R R + β_{3} E d u \times C o v + β_{4} E d u \times R R + β_{5} C o v \times R R + β_{6} C o v + β_{7} R R + β_{8} G D P + β_{i} I n d . C o v a r i a t e s + ϵ

(2)

L n (\frac{P}{1 - p}) = a + β_{1} E d u + β_{2} E d u \times C o v \times R R + β_{3} C o v + β_{4} R R + β_{5} G D P + β_{i} I n d . C o v a r i a t e s + ϵ

(3)

Graphical illustrations of the results from Equation 2 will illustrate how predicted probabilities of having good health varies along the empirical distribution of replacement rates in two contexts, low (50%) and high (100%) coverage rates. To make the presentation more comparable, the empirical distribution of coverage, replacement rates and expenditure have been rescaled to vary between 0 and 100 in the figures. The results presented are based on the total analytical sample. In addition, analyses were stratified by gender and activity status. Comparisons between the whole sample and the stratified samples are based on simple ratios between predicted values for the whole sample and the stratified samples.

Results

Figure 1 displays predicted probabilities of having good health for individuals with primary and tertiary education over the empirical distribution of expenditure on unemployment benefits, and Figure 2 displays the corresponding predicted values over the empirical distribution of coverage and replacement rates.

The results for unemployment benefit expenditure are very similar to those presented by Dahl and van der Wel [4]: higher expenditures increase the probability of having good health for individuals with both primary and tertiary education (Figure 1). Moreover, as this association is stronger for respondents with primary education, educational differences in self-assessed health tend to decrease as social expenditure increase. Also, higher coverage rates seem to be associated with better self-assessed health (Figure 2), but as this association is very similar for individuals with primary and tertiary education, it does little to affect educational differences in self-assessed health. In contrast, higher replacement rates do not appear to be associated with better self-assessed health, irrespective of respondents’ level of educational attainment (Figure 2).

Figure 3 provides evidence for the existence of an interaction effect between coverage and replacement rates. Moreover, this interaction effect is more pronounced for individuals with primary education. In countries with high coverage rates, higher replacement rates are associated with an increased probability that individuals with both primary and tertiary education will state that they are in good health. Since this association is more marked for individuals with primary education, higher replacement rates in combination with high coverage rates tend also to be associated with a decrease in educational differences in self-assessed health. In countries with high coverage rates, the ratio between the predicted probability of having good health for individuals with tertiary education vs. individuals with primary education at the lowest replacement level is 1.33 (0.76/0.57) and at the highest replacement level, it is 1.05 (0.92/0.88). The narrowing of educational differences in self-assessed health along increases in replacement levels at high coverage levels is also present when estimating a model containing only the three-way interaction term and the main effects (i.e., Equation 3). In this model, the ratio between the predicted probabilities of having good health for individuals with tertiary education vs. individuals with primary education is reduced from 1.16 to 1.05 as replacement levels increase.

Stratified analyses indicate that these ratios are rather stable across groups. For women, the corresponding ratios are 1.33 and 1.06, and for men, 1.32 and 1.05. For the unemployed, the ratio between predicted probabilities for respondents with tertiary vs. respondents with primary education is comparatively large a high replacement rates (1.18), but since the ratio at low replacement levels is also high (1.49), the relative decrease in educational differences in self-rated health is as large for this group ([1.49–1.18]/1.49 = 21%) as it is for the whole sample ([1.33–1.05]/1.33 = 21%).

Analyses were also performed to see whether these predicted probabilities differed according to household income. For individuals living in a household with an equalized household income below 40% of the median income in the full analytical sample, the ratio between the predicted probabilities of having good health for individuals with tertiary education vs. individuals with primary education is reduced from 1.38 to 1.05 as replacement levels increase. For individuals living in households with an equalized household income above 60% of the median income in the full sample, this ratio is decreased from 1.26 to 1.03. For individuals living in households with an equalized median income between 40 and 60% of the median income, the pattern resembles closely that found for the full analytical sample.

That the general pattern in Figure 3 is rather similar across different socioeconomic groups may indicate that unemployment benefits function as form of collective resource. By decreasing insecurities and increasing stability and predictability, they do not only affect the health of the group it is primarily intended for, the unemployed, but also the health of groups with a more stable position on the labour market and higher incomes.

In contrast, in countries with low coverage rates the self-assessed health of respondents with both primary and tertiary education becomes worse as replacement rates increase (Figure 4). This pattern is surprising and difficult to understand intuitively. One possible explanation is that as replacement rates increase, the discrepancy between a system that formally promises significant economic support during unemployment and the concrete outcome of such a system, where relatively few individuals get access to this support, gives rise to feelings of distrust and disappointment that in turn contribute to worse self-assessed health. This, in combination with the stigmatization that more narrowly targeted schemes may entail, could explain the pattern found in Figure 4.

Conclusions

This article has demonstrated that focusing on the institutional organization of welfare states may provide important insights into the mechanisms linking macro-level social policies to population health and socio-economic differences in health. Results show that taken in isolation, the coverage of unemployment benefit schemes is positively related to the self-assessed health of respondents with both primary and tertiary education, but do little to affect educational differences in health. In contrast, unemployment benefit replacement rates in isolation appear to have no relationship to respondents’ self-assessed health irrespective of their level of educational attainment. However, high replacement rates in combination with high coverage rates is associated with better self-assessed health among respondents with both primary and tertiary education, and since this association is stronger for the former group, it also tends to be associated with a decrease in educational differences in self-assessed health. Previous research has indicated that unemployment, and possibly, the risk, thereof, is more related to mental than to physical health [25]. Therefore, an outcome variable focusing on mental health may have strengthened further the associations presented here.

In relation to previous findings that social expenditures reduce educational differences in self-assessed health, we have been able to institutionally decompose this effect, and to show that the association between unemployment benefit replacement rates and health and educational differences in health is contingent upon a large proportion of the population being covered by these benefits. This finding is likely to be of relevance for both policy-makers and researchers interested in the effect of macro-level social policies on individual-level health outcomes.

Footnotes

Conflict of interest

The authors declare that there is no conflict of interest.

Funding

The research leading to these results was carried out within the framework of the DRIVERS project () coordinated by EuroHealthNet, and has received funding from the European Union (FP7 2007-2013) under grant agreement no. 278350.

References

Bergqvist

Åberg Yngwe

Lundberg

. Understanding the role of welfare state characteristics for health and inequalities – an analytical review. BMC Public Health 2013;13:1234–54.

Dahl

van der Wel

. Educational inequalities in health in European welfare states: a social expenditure approach. Soc Sci Med 2013;81:60–9.

Esping-Andersen

. The three worlds of welfare capitalism. Cambridge: Polity Press, 1990.

Lundberg

. Commentary: politics and public health—some conceptual considerations concerning welfare state characteristics and public health outcomes. Int J Epidemiol 2008;37:1105–8.

Lundberg

Åberg Yngwe

Kölegård Stjärne

. The role of welfare state principles and generosity in social policy programmes for public health: an international comparative study. Lancet 2008;372:1633–40.

Palme

. Welfare states and inequality: institutional designs and distributive outcome. Res Soc Stratif Mobil 2006;24:387–403.

Marmot

. The status syndrome: how your social standing affects your health and life expectancy. London: Bloomsbury, 2004.

Wilkinson

Pickett

. The spirit level: Why more equal societies always do better. London: Allen Lane, 2009.

Marshall

. Citizenship and social class and other essays. Cambridge: Cambridge University Press, 1950.

10.

Marmot

. Self esteem and health: autonomy, self esteem, and health are linked together. BMJ 2003;327:574–757.

11.

Sjöberg

. Social insurance as a collective resource: unemployment benefits, job insecurity and subjective well-being in a comparative perspective. Soc Forces 2010;88:1281–1304.

12.

Veenhoven

. Well-being in the welfare state: Level not higher, distribution not more equitable. J Comp Pol Anal 2000;2:91–125.

13.

Ouweneel

. Social security and well-being of the unemployed in 42 nations. J Happiness Stud 2002;3:167–92.

14.

Bambra

Eikemo

. Welfare state regimes, unemployment and health: a comparative study of the relationship between unemployment and self-reported health in 23 European countries. J Epidemiol Commun H 2009;63:92–98.

15.

Nordenmark

Strandh

Layte

. The impact of unemployment benefit system on the mental well-being of the unemployed in Sweden, Ireland and Great Britain. Eur Soc 2006;8:83–110.

16.

Rodriguez

. Keeping the unemployed healthy: the effect of means-tested and entitlement benefits in Britain, Germany, and the United States. Am J Public Health 2001;91:1403–11.

17.

Ferrarini

Nelson

Korpi

. Social citizenship rights and social insurance replacement rate validity: pitfalls and possibilities. J Eur Public Policy 2013; 20: 1251–66. For access to data, see https://dspace.it.su.se/dspace/handle/10102/7 (accessed 24 September 2013).

18.

http://epp.eurostat.ec.europa.eu/portal/page/portal/eurostat/home/ (accessed 25 September 2013).

19.

Idler

Benyamini

. Self-rated health and mortality: a review of twenty-seven community studies. J Health Soc Behav 1997:21–37.

20.

Jylhä

. What is self-rated health and why does it predict mortality? Towards a unified conceptual model. Soc Sci Med 2009;69:307–16.

21.

Jürges

. True health vs response styles: exploring cross-country differences in self-reported health. Health Econ 2007;16:163–78.

22.

Schneider

Müller

. Measurement of education in EU-SILC – preliminary evaluation of measurement quality. Equalsoc Working Paper 2009/5.

23.

Marmot

Wilkinson

. Social determinants of health. New York: Oxford University Press, 2006.

24.

Carle

. Fitting multilevel models in complex survey data with design weights: recommendations. BMC Med Res Methodol 2009;9:49–62.

25.

Paul

Moser

. Unemployment impairs mental health: meta-analyses. J Voc Behav 2009;74:264–82.