Sociology of Health & Illness Vol. 34 No. 6 2012 ISSN 0141–9889, pp. 858–879 doi: 10.1111/j.1467-9566.2011.01433.x

Welfare state regimes, health and health inequalities in adolescence: a multilevel study in 32 countries Matthias Richter1, Katharina Rathman2,3, Saoirse Nic Gabhainn4, Alessio Zambon5, William Boyce6 and Klaus Hurrelmann2 1

Institute of Medical Sociology, Martin Luther University Halle-Wittenberg, Halle, Germany 2 Hertie School of Governance, Berlin, Germany 3 Berlin Graduate School of Social Sciences, Humboldt University Berlin, Germany 4 Health Promotion Research Centre, National University of Ireland, Galway, Ireland 5 Department of Public Health, University of Turin, Italy 6 Department of Community Health & Epidemiology, Queen’s University, Kingston, Canada

Abstract

Comparative research on health and health inequalities has recently started to establish a welfare regime perspective. The objective of this study was to determine whether different welfare regimes are associated with health and health inequalities among adolescents. Data were collected from the ‘Health Behaviour in School-aged Children’ study in 2006, including 11- to 15-year-old students from 32 countries (N = 141,091). Prevalence rates and multilevel logistic regression models were calculated for self-rated health (SRH) and health complaints. The results show that between 4 per cent and 7 per cent of the variation in both health outcomes is attributable to differences between countries. Compared to the Scandinavian regime, the Southern regime had lower odds ratios for SRH, while for health complaints the Southern and Eastern regime showed high odds ratios. The association between subjective health and welfare regime was largely unaffected by adjusting for individual socioeconomic position. After adjustment for the welfare regime typology, the country-level variations were reduced to 4.6 per cent for SRH and to 2.9 per cent for health complaints. Regarding cross-level interaction effects between welfare regimes and socioeconomic position, no clear regime-specific pattern was found. Consistent with research on adults this study shows that welfare regimes are important in explaining variations in adolescent health across countries.

Keywords: welfare regimes, adolescence, subjective health, socioeconomic status, HBSC

Introduction Socioeconomic differences in health and longevity are well documented. Innumerable studies have shown that adults with lower education, occupational status and income are more likely to suffer from adverse health and die earlier than those who are better-off (Mackenbach et al.  2012 The Authors. Sociology of Health & Illness  2012 Foundation for the Sociology of Health & Illness/Blackwell Publishing Ltd. Published by Blackwell Publishing Ltd., 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA

Welfare state regimes, health and health inequalities in adolescence

859

2002, Bartley 2004, Mackenbach 2006). A more complex picture of socioeconomic inequalities in health emerges among adolescents. Even though the social gradient is less persuasive and less visible than among adults, several studies have found clear health differences for individual (West 1997, Goodman 1999, Chen et al. 2002, Starfield et al. 2002), as well as school- and country-level socioeconomic status in adolescence (Torsheim et al. 2004, 2006). Socioeconomic differences in health and health behaviour among adolescents within countries now represent an active and important area of research (Richter et al. 2009). However, although an increasing number of studies have begun to integrate a welfare regimespecific perspective on the social determinants of health (Raphael 2006), little is known about the relationships between different aspects of the welfare state, health and health inequalities in adolescence. Welfare regimes The welfare state has been subject of academic interest for several years and across a range of disciplines (Titmuss 1974, Esping-Andersen 1990, 1999). Various classifications have been suggested to categorise the provision of welfare into different ‘regime types’ (Bambra 2005, 2006, Eikemo and Bambra 2008). Although Esping-Andersen’s ‘three worlds of welfare’ typology is probably the most well-known welfare state classification, it has been criticised by many scholars (Castles and Mitchell 1993, Ferrera 1996, Arts and Gelissen 2002, Bambra 2006, 2007). Esping-Andersen’s typology of welfare states is based on three principles (decommodification, social stratification and private-public mix) which according to the extent of their presence allow the identification of three ‘ideal’ regime types: liberal, conservative, and social democratic (Eikemo et al. 2008a, 2008b). Welfare states belonging to the liberal regime (e.g. UK, USA, Ireland and Canada) are characterised by minimal state provision of welfare, modest benefits and strict entitlement criteria (Eikemo and Bambra 2008). Further, welfare recipients are usually means-tested and stigmatised (Esping-Andersen 1990, Bambra 2007). In conservative welfare states (Germany, France, Austria, Belgium, Italy and, to a lesser extent, the Netherlands) welfare programmes and benefits are often ‘status differentiating’, related to income levels, and administered through employers (Bambra 2007). Thus, such welfare politic is geared towards maintaining existing social patterns (Eikemo and Bambra 2008). The social democratic welfare regime (Norway, Sweden, Denmark and Finland) represents the smallest cluster and is characterised not only by its universal, egalitarian and comparatively generous benefits, but also by a commitment to full employment and income protection. Further, this is a regime which strongly intervenes to promote equality through a redistributive social security system (Esping-Andersen 1990). As a result of an extensive debate about Esping-Andersen’s typology, Ferrera (1996, 2005) introduced a fourth type: Scandinavian (social democratic), Bismarckian (conservative), Anglo-Saxon (liberal) and the Southern welfare regime. According to Ferrera (1996) the Southern European countries, such as Italy, Greece, Portugal and Spain, comprise a distinctive, southern, welfare state regime. Their welfare programmes are described as ‘rudimentary’ because they are still characterised by their fragmented system of welfare provision and welfare services (Eikemo and Bambra 2008). Moreover, reliance on the family and voluntary sector remains a prominent defining feature. More recently, former ‘Eastern European’ countries, such as Poland and Bulgaria, are being considered as another distinctive regime type (Esping-Andersen 1999, Kovacs 2002). These countries have experienced economic upheaval after the breakdown of their communist regimes and are still undergoing extensive social reforms which are frequently indicated by a shift towards the Liberal welfare regime. Their welfare benefits are  2012 The Authors Sociology of Health & Illness  2012 Foundation for the Sociology of Health & Illness/Blackwell Publishing Ltd

860

Matthias Richter et al.

characterised by limited health service provisions and the overall population health is relatively poor compared to other member states of the European Union (Eikemo and Bambra 2008). These five regime types represent the most frequent classification in current approaches to comparative regime-specific research (Bambra 2006, Eikemo et al. 2008a, 2008b, 2008c). Welfare state regimes and health An increasing number of studies have shown that population health differs substantially by welfare regimes (Navarro et al. 2006, Chung and Muntaner 2006, 2007, Eikemo et al. 2008a, 2008b, 2008c, Bambra et al. 2009). Most of these studies focus on infant mortality and low birth weight. There is no study that investigates adult mortality in a regime comparative perspective. Navarro et al. (2006) focused on life expectancy, while Eikemo et al. (2008c) examined differences in morbidity using a welfare state regime approach. In almost all analyses it has been shown that social democratic countries rank more positively on various population health indicators than the other regimes, especially those characterised as being liberal. These findings have been consistent for different welfare regime typologies irrespective of whether they are based on Esping-Andersen’s classification or on typologies of political traditions (for example Navarro et al. 2006, Borrell et al. 2007) that are closely related to Esping-Andersen’s typology, as countries with social democratic traditions usually also have a social democratic welfare system. Welfare state regimes and health inequalities Recently a welfare state regime perspective was also introduced to the analyses of crossnational differences in the magnitude of health inequalities for adults (Eikemo et al. 2008a, 2008b, 2008c, Bambra et al. 2009). Eikemo et al. (2008a), for example, showed that the Southern European welfare regime had the largest and the Bismarckian regime the smallest disparities in self-rated health and longstanding illness, while the Scandinavian regime was less favourably placed than the Anglo-Saxon and Eastern regimes. This exemplary finding is found in many studies: the social democratic welfare states do not have the smallest health inequalities even though they have the highest level of population health. In general, conservative countries seem to perform better and show the smallest inequalities in health (especially for self-rated health). Although there is increasing evidence for this patterning of health and health inequalities it should be acknowledged that the so-called ‘Scandinavian paradox’ is highly debated among scholars (Lahelma and Lundberg 2009). This finding represents an important challenge for public health policy and practice (Dahl et al. 2006). So far, comparative research on welfare regimes and health has rarely focused on adolescents (Zambon et al. 2006). It is reasonable to assume that not only individual determinants but also macrolevel factors and welfare regime characteristics influence adolescent health and wellbeing (Zambon et al. 2006, Ravens-Sieberer et al. 2008, Holstein et al. 2009). These welfare arrangements are likely to work through and along with social and income inequalities to affect intermediary determinants such as material circumstances, parental support or health behaviour of both parents and children (CSDH 2008, Beckfield and Krieger 2009). Thus, macrolevel determinants do not necessarily need to be directly associated with adverse health but could act indirectly as a stratifying mechanism through other determinants of health (Torsheim et al. 2004). Further, research has shown that socioeconomic differences in health in adolescence are less consistent and less pronounced (West 1997, Spencer 2006, Richter et al. 2009), as compared to health inequalities in adulthood. In this context it can be expected that the specific characteristics of welfare  2012 The Authors Sociology of Health & Illness  2012 Foundation for the Sociology of Health & Illness/Blackwell Publishing Ltd

Welfare state regimes, health and health inequalities in adolescence

861

infrastructure and arrangements of a country influence health and health inequalities cumulatively over the life course (Graham and Power 2004, Blane 2006). Zambon et al. (2006) were among the first who examined whether different types of welfare regimes mediate the effect of socioeconomic position on adolescents’ health using data from the ‘Health Behaviour in School-aged Children (HBSC)’ in 2002. They hypothesised that countries with stronger redistributive policies (i.e. social democratic and conservative welfare regimes) are more effective in weakening the association between socioeconomic position and health. Their findings showed that, even after adjusting for family socioeconomic position (SEP), levels of general subjective health for adolescents are lower in countries belonging to the Anglo-Saxon and Eastern European regimes. The findings from Zambon et al. (2006) reinforce that welfare regimes are also important determinants of adolescents’ health. However, it is not known whether the challenging findings observed for adults (i.e. better overall population health in social democratic regimes but not the smallest health inequalities) also hold for young people. We decided to update and to extend Zambon et al.’s analyses with more recent data from the HBSC study by applying multilevel analyses. The objective of our study is therefore to examine whether different types of welfare regimes are associated with differences in subjective health and with socioeconomic differences in health among adolescents in 32 European and North American countries.

Methods Population Data were obtained from the Health Behaviour in School-aged Children (HBSC) study 2005 ⁄ 2006, a cross-national survey conducted in collaboration with the World Health Organization (Currie et al. 2008a). The aim of the HBSC study is to describe young people’s health and health behaviour and to analyse how these outcomes are associated with social contexts. Cross-sectional surveys of 11-, 13- and 15-year-old children and adolescents are carried out every four years in a growing number of countries based on an internationally agreed protocol (Currie et al. 2009). The latest survey in 2005 ⁄ 06 included a total of 41 countries from Europe and North America. The aims and theoretical framework of the study have been described elsewhere (Currie et al. 2007, Roberts et al. 2009). The data were collected by means of standardised questionnaires, administered in school classrooms according to standard instructions. Students were selected using a clustered sampling design, where the initial sampling unit was the school class. The recommended minimum sample size for each country was 1536 students per age group (i.e. 11, 13 and 15 year olds), to assure a 95 per cent confidence interval of + ⁄ )3 per cent for prevalence estimates of around 50 per cent. The sample size included a design factor of 1.2 because of the cluster sampling (the design factor of 1.2 was based on analyses of the 1993 ⁄ 1994 and 1997 ⁄ 1998 HBSC surveys). In some of the participating countries, HBSC was exempt from the requirement for ethical approval. In other countries that required approval this was obtained by different institutional review boards. The present analysis is based on N = 141,091 11 to 15 year olds (66,964 male and 74,127 female students) from 29 European countries, as well as Canada, the United States and Israel, which were categorised into four regimes according to Ferrera’s welfare typology (1996) (Scandinavian, Bismarckian, Anglo-Saxon, Southern), with an additional fifth category for Eastern European countries (Table S1, available in the supplementary files accompanying the online version of this paper). Data from England, Scotland, and Wales  2012 The Authors Sociology of Health & Illness  2012 Foundation for the Sociology of Health & Illness/Blackwell Publishing Ltd

862

Matthias Richter et al.

were merged to represent the United Kingdom as were data from the French and Flemishspeaking parts of Belgium to represent Belgium. Iceland, Malta, Greenland and Turkey were excluded from the analyses because it was not possible to categorise them into one of the regime clusters. Furthermore, we excluded Lithuania and Slovakia due to a high number of missing values (more than 10%) for both SEP indicators. Health outcomes The outcome measures were self-rated health and health complaints (Haugland and Wold 2001, Haugland et al. 2001, Hetland et al. 2002, Ravens-Sieberer et al. 2008). In terms of selfrated health (SRH) students were asked to describe their health: ‘would you say your health is…?’ with response codes of ‘excellent’, ‘good’, ‘fair’ and ‘poor’. The response categories were dichotomised into ‘very good or good’ versus ‘less than good’ (‘fair’, ‘poor’) health (Currie et al. 2008a). Health complaints were measured using the HBSC symptom check list (HSCL). Students were asked how often in the last 6 months they had experienced the following symptoms: headache; stomach ache; backache; feeling low; irritable or bad tempered; feeling nervous; difficulties in getting to sleep and feeling dizzy. The response options for each item ranged from ‘about every day’ to ‘rarely or never’. These response options were categorised into ‘two or more symptoms more than once a week’ versus ‘less than two symptoms’ (Currie et al. 2008a). Socioeconomic position Socioeconomic position (SEP) was measured with two different indicators: family affluence and parental occupation. Family affluence was assessed using the family affluence scale (FAS) which is a validated measure based on four different aspects of the household’s material conditions (Boyce et al. 2006, Torsheim et al. 2006, Currie et al. 2008b): Does your family own a car? (0, 1, 2 or more); how many times did you travel away on holiday with your family during the past 12 months? (0, 1, 2, 3 or more); do you have your own bedroom for yourself? (no = 0, yes = 1); and how many computers does your family own? (0, 1, 2, 3 or more). A composite FAS score was calculated by summing the responses to these four items ranging from 0 to 7. The FAS scores were subsequently recoded in three groups: high (6–7 points), middle (4–5 points) and low (0–3). Family affluence has several benefits such as the low percentage of missing responses from young people and documented cross-national comparability (Currie et al. 2008a, 2008b). Parental occupation was assessed by two open-ended questions. Students were asked to indicate separately where their father and mother work and to describe what kind of job they do. Those students whose parents did not work were asked to indicate what they were doing instead: ‘is ill or has retired or going to school’, ‘is looking for a new job’, ‘works as a housewife ⁄ househusband’. Using standard occupational coding guidelines, all countries were required to code this information into five rank-ordered social classes, in accordance with the classical RGSC (registrar general social class) coding scheme. Since many mothers were economically inactive and many responses on parental occupation were missing, information from the father and mother was combined, using the highest category for each couple as the parental indicator. In order to obtain three groups of similar size, the original five categories were recoded into high (I and II), middle (III) and low parental occupation (IV and V). Students were also classified in the lowest category, if neither parent was working for money (‘looking for a job’, ‘full time at home’). Adolescents who lived with a single parent were also grouped in the lowest category, when the single parent was ‘unemployed’, ‘sick, retired, or a student’ or ‘full time at home’.  2012 The Authors Sociology of Health & Illness  2012 Foundation for the Sociology of Health & Illness/Blackwell Publishing Ltd

Welfare state regimes, health and health inequalities in adolescence

863

Statistical analysis In order to assess the extent to which the welfare state regime typology explains the proportional variation in both health outcomes as well as their association with socioeconomic inequalities in health, two-level hierarchical logistic regression models were conducted. The basic principle of hierarchical regression (also known as multilevel modelling) is that the data are structured in hierarchically nested groups (Merlo 2003, Merlo et al. 2005, Rabe-Hesketh and Skrondal 2008, Hox 2010). The level 1 units in our sample are individual students (N = 141,091), the level 2 units are the 32 European and North American countries. By using self-rated health and health complaints as separate outcomes, we were able to analyse the extent to which health varies at the individual level compared to the country level and simultaneously to identify factors (i.e. welfare regime type) that might explain the variation. The analyses with random intercepts for both health outcomes were done in five steps. First, the variation in both health outcomes without using any explanatory variables was analysed in order to decompose the variance of the intercept into variance components for each of the two levels (model 1). Such models are called interceptonly ⁄ empty models or just variance component models (Rabe-Hesketh and Skrondal 2008, Hox 2010). In a second step, a model with individual socioeconomic predictors, namely, parental occupation (model 2a) and family affluence (model 2b) was conducted. In a third step, we analysed the effect of welfare regime dummy variables (the Scandinavian regime was used as reference) separately, in order to identify how both health outcomes vary by welfare state arrangements (model 3). Next, we included parental occupation (model 4a) and family affluence (model 4b) as well as the regime dummy variables, respectively. Finally, in order to assess the association between different welfare regimes and socioeconomic differences in health, we followed an approach by Rostila (2007) and analysed two additional models with cross-level-interaction terms between both SEP indicators (models 5a and 5b) and the four welfare regime dummy variables (high SEP group in the Scandinavian regime as reference). The variation by country was expressed as intraclass correlation coefficient (ICC) (RabeHesketh and Skrondal 2008) which indicates the proportion of variance of the outcome that is attributable to differences between countries. As the level 1 and level 2 variances are not on the same scale, we have used the latent variable approach (Goldstein 2003). We assume a latent underlying continuum of both health outcomes; the ICC was computed using a formula for logistic models (Rabe-Hesketh and Skrondal 2008: 238). In this formula, residuals are assumed to have a standard logistic variance structure at the level 1 (p2 ⁄ 3 = 3.29), and a normal distributed variance structure (u0j) at level 2 (here: country where the subscript j refers to the different countries). At the country-level the ICC may be calculated as u0j ⁄ (u0j + 3.29), indicating the ratio of the random country variance to the total variance. At the individual-level the ICC equals 3.29 ⁄ (u0j + 3.29). We have presented these numbers as a percentage of total variance (ICC · 100). By adding successively individual and country-level predictors to the models, we consider some part of the compositional differences and disentangle some of the country variance discovered in the variance component model (model 1). To assess the goodness of fit of different models, we calculated the deviance values ()2 · log-likelihood) (Albright and Marinova 2010). In each model we controlled for age (dummy variables coded with three age groups: 11, 13 and 15; 11 year olds as reference category) and gender (boys as reference). The conventional 5% level was used to determine statistical significance. All analyses were conducted using STATA version 11.  2012 The Authors Sociology of Health & Illness  2012 Foundation for the Sociology of Health & Illness/Blackwell Publishing Ltd

864

Matthias Richter et al.

Results Descriptive results: regime-specific prevalence rates Table 1 shows the prevalence rates for both health outcomes by welfare state regime and country for boys and girls. Fair ⁄ poor SRH for boys was highest in the Anglo-Saxon welfare state regime (13.6%) followed by the Eastern European regime (13.3%). For girls the highest Table 1 Age-adjusted prevalence rates (PR) for fair ⁄ poor health and two or more health complaints more than once a week by welfare state regime ⁄ country, 11- to 15-year-old students (N = 141091), in percentages Fair ⁄ poor self-rated health (SRH)

Two or more health complaints

PR boys

PR girls

PR total

Country (N = 32) welfare regime

10.8 9.1 14.5 8.5 10.8 9.2 13.8 9.1 11.4 10.3 10.0 5.4 10.3 12.7 9.2 5.8 16.4 17.3 13.6 4.2 5.9 3.8 10.7 5.6 5.8 8.3 10.4 9.7 12.4 18.2 13.3 11.5 11.1 22.8 9.3 21.0 13.3

16.1 11.3 19.9 13.5 15.1 13.0 20.9 15.4 15.9 19.8 18.9 9.3 16.3 16.9 12.0 7.2 24.5 25.3 18.9 7.2 11.8 6.0 16.1 9.0 9.7 12.8 16.6 14.0 14.8 25.0 22.4 16.7 18.2 31.4 14.0 44.7 21.0

13.5 10.2 17.2 11.0 13.0 11.1 17.3 12.2 13.7 15.1 14.4 7.4 13.3 14.8 10.6 6.5 20.4 21.3 16.2 5.7 8.9 4.9 13.4 7.3 7.7 10.5 13.5 11.9 13.6 21.6 17.9 14.1 14.6 27.1 11.7 32.8 17.2

Denmark Finland Norway Sweden Scandinavian Austria Belgium France Germany Luxembourg Netherlands Switzerland Bismarckian Canada Ireland Israel United Kingdom United States Anglo-Saxon Greece Italy Macedonia Portugal Spain Southern Bulgaria Croatia Czech Republic Estonia Hungary Latvia Poland Romania Russian Federation Slovenia Ukraine Eastern

PR boys

PR girls

PR total

16.8 19.8 20.4 22.0 19.6 13.3 24.0 29.0 16.7 22.0 15.1 20.9 20.8 25.5 19.7 46.3 22.8 31.8 26.6 31.4 38.1 26.6 14.6 25.1 26.9 30.8 24.8 26.7 26.4 28.1 27.3 29.1 34.8 31.8 16.8 28.1 27.4

27.7 31.6 30.9 35.8 31.3 22.3 32.9 45.8 26.6 41.7 27.8 33.9 33.1 36.3 29.3 53.6 32.1 48.2 37.6 50.2 54.6 37.5 30.7 39.2 41.7 42.6 37.1 41.2 37.5 39.6 42.3 41.1 51.9 42.0 23.6 50.7 40.8

22.2 25.7 25.6 28.9 25.5 17.0 28.5 37.4 21.6 31.8 21.4 27.4 26.9 30.9 24.5 50.0 27.5 40.0 32.1 40.8 46.4 32.1 22.6 32.1 34.3 36.7 30.9 34.0 31.9 33.9 34.8 35.1 43.3 36.9 20.2 39.4 34.1

 2012 The Authors Sociology of Health & Illness  2012 Foundation for the Sociology of Health & Illness/Blackwell Publishing Ltd

Welfare state regimes, health and health inequalities in adolescence

865

rates were also found in the Eastern European (21.0%) and the Anglo-Saxon regime (18.9%). For both genders, the lowest rates were observed in the Southern welfare regime. Students in Scandinavian and Bismarckian welfare regimes were ranked between these two extremes. With regard to health complaints, boys in the Eastern regime and girls in the Southern regime reported the highest prevalence followed by the Anglo-Saxon (boys) and the Eastern European regime (girls). Students in the Scandinavian and Bismarckian regimes showed the lowest prevalence of two or more weekly health complaints. Generally, girls reported worse ill-health than boys for both health outcomes in all welfare regimes. Multivariate analyses: welfare regimes and health In order to assess whether the prevalence of fair ⁄ poor health and two or more health complaints are significantly different between welfare regimes we calculated logistic multilevel random intercept models. As our research focus was mainly on general regime-specific patterns in health and health inequalities, the multilevel analyses were conducted for the whole sample adjusted for gender (Table 2). In a first step we calculated the intraclass correlation coefficient (ICC) for the empty model without any covariates (model 1). This model indicated that 6.7 per cent of the within-subject variation for fair ⁄ poor SRH is attributable to differences between countries (two or more health complaints: 3.9%). Models 2a and 2b take into account level 1 variables only (age, gender, parental occupation and family affluence). The models show that having fair ⁄ poor SRH and two or more health complaints are both positively correlated with increasing age and being a female student. Compared to the highest group of family affluence and occupational status the odds ratios for fair ⁄ poor health increase with decreasing socioeconomic position (low parental occupation: odds ratio (OR) 1.43, 95% confidence interval (CI): 1.38–1.49; low family affluence: OR 1.92, 95% CI: 1.84–2.00). For two or more health complaints a similar pattern was found (low parental occupation: OR 1.26, 95% CI: 1.22–1.29; low family affluence: OR 1.42, 95% CI: 1.37–1.47). After adjustment for individual-level characteristics, the variation in intercepts between countries indicates that about 7 per cent for fair ⁄ poor health and 4 per cent for two or more health complaints of the total variation of individual health outcomes is due to country-level characteristics. These percentages are related to the country level and might be attributable to contextual factors which are not considered in the model. We therefore included welfare regime dummy variables in our model (Table 3). In line with our descriptive results the Southern regime had the lowest odds ratio of fair ⁄ poor health (OR 0.55, 95% CI: 0.33–0.93) compared to the Scandinavian regime, while for the other regimes no significant association was found. The multilevel model for two or more health complaints also confirms our findings in Table 1, showing that the Southern (OR 1.53, 95% CI: 1.01– 2.32) and East European welfare regimes (OR 1.50, 95% CI: 1.05–2.16) displayed higher odds ratios for health complaints compared to the Scandinavian regime. For the AngloSaxon regime a similar but non-significant effect was found. After adjusting for the welfare regime typology, the unexplained variations between countries were reduced to 4.6 per cent for SRH and to 2.9 per cent for multiple health complaints compared to the second model. In other words, the welfare regime type partly explains the variation in both health outcomes among countries. In models 4a and 4b we added individual SEP indicators to the model already including regime dummy variables. The individual-level associations did not substantially change from the second model. Further, the association between the Southern welfare regime and SRH was reduced, but remained significant after adjustment for SEP indicators. For two and more health complaints the effect for the Eastern welfare regime was still significant after adjusting for parental occupation (model 4a). This indicates that a considerable part of this association  2012 The Authors Sociology of Health & Illness  2012 Foundation for the Sociology of Health & Illness/Blackwell Publishing Ltd

 2012 The Authors Sociology of Health & Illness  2012 Foundation for the Sociology of Health & Illness/Blackwell Publishing Ltd

32

32