Article Text
PDF
Abstract
Background This study investigates social inequalities in self-rated health dynamics for working-aged adults in four nations, representing distinct welfare regime types. The aims are to describe average national trajectories of self-rated health over a 7-year period, identify social determinants of cross-sectional and longitudinal health and compare cross-national patterns.
Methods Data are from national household panel surveys in Britain, Germany, Denmark and the USA. The self-rated health of working-age respondents is measured for the years 1995–2001. Social indicators include education, occupational class, employment status, income, age, gender, minority status and marital status. Latent growth curve models are used to estimate both individual change and average national trajectories of self-rated health, conditioned on the social indicators.
Results Ageing-vector graphs reveal general declines in health as people age. They also show differential patterns of change for specific national cohorts. Older cohorts in Denmark had poorer health and young cohorts in the USA had better health in 2001 than 1995. Social covariates predicted baseline health in all four countries, in ways that were consistent with welfare regime theories. Once inequalities in baseline health were accounted for, the few determinants of mean health decline occurred mainly in the USA, again in line with theoretical expectations. Finally, trajectories of health for those in average and advantaged social circumstances were similar, but disadvantaged individuals had much poorer health trajectories than ‘average’ individuals. The differences were greatest in the countries with lower levels of public transfers.
Conclusion National differences in self-rated health trajectories and their social correlates may be attributed partly to welfare policies.
- ageing
- comparative study
- longitudinal data analysis
- self-rated health
- socioeconomic factors
- social differences
- social welfare
Statistics from Altmetric.com
- ageing
- comparative study
- longitudinal data analysis
- self-rated health
- socioeconomic factors
- social differences
- social welfare
The association between social disadvantage and health is well documented in social research: people from disadvantaged households die younger and have more illnesses than those who are better off. To make full use of this knowledge, though, we need to understand the long-range implications—for example, whether the picture gets better or worse, and for whom—and how social policy can improve or aggravate the situation.1 Existing gaps in knowledge come partly from the usual practice of measuring health and social position contemporaneously. Therefore, although social position is widely known to be associated with health in cross-sectional studies,2 its relationship with health change is less well researched. Recent research using a dynamic or life course approach challenges this state of affairs by showing that both health and social position are much more complex and diverse than previously thought.3 4 Long-term studies show that health, on average, deteriorates slowly over time, but in some people it actually improves steadily over the years, in some it remains stable or fluctuates, and in some it declines quite rapidly. Taken together, this work suggests that a dynamic approach to health and social position will tell us more than a static approach. It will broaden the view of who is at risk, and offer a more informative picture of long-range implications.
Ecological systems theory also tells us that health and social dynamics are shaped by the institutional settings that people live in,5 whereas the welfare states literature provides a framework for understanding how a nation's social policies shape these settings—and which groups are likely to be most vulnerable. This study addresses these issues by examining health and social position over a 7-year period (1995–2001) in four OECD countries characterised as distinct welfare regime types. Castles and Mitchell's typology, which builds on Esping-Andersen's well-known ‘worlds of welfare capitalism’ thesis6 and emphasises differences in the redistributive properties of social transfers and the provision of welfare services, is particularly apposite for health inequalities research.7 8 These authors classify: the USA as a liberal welfare state with little redistribution of wealth and minimal welfare provision; Britain as a radical welfare state because of its low expenditure combined with a more equalising health and benefits system; Germany as a conservative welfare state with higher expenditure on welfare services but an emphasis on status differentiation leading to negligible equalisation; and Denmark as a non-right hegemony with high expenditure and strong egalitarian policies (see supplementary appendix table A1, available online only).
Data from the USA, Britain, Germany and Denmark will be used to examine: (1) how people's experiences of health change over time; (2) how social determinants relate to cross-sectional and longitudinal health; and (3) whether the picture is different from one welfare state to another. We hypothesise that the strength of the relationship between the social determinants and health will depend on the universalism of social transfers and welfare services in each country.9 Social influences will thus be most apparent in the US data and least apparent in the Danish data.
Methods
Data
The data come from the 1994–2001 waves of the US Panel Study of Income Dynamics (PSID),10 the British Household Panel Survey (BHPS),11 the German Socio-Economic Panel Survey (SOEP)12 and the Danish panel from the European Community Household Panel Survey (ECHP).13 The PSID started with a national sample of nearly 5000 households in 1968. Individuals were interviewed every year until 1997, after which the interviews were carried out biennially. The BHPS, initiated in 1991, is an annual survey of private households containing approximately 5500 households and 9000 men and women. The SOEP is also a representative longitudinal study of private households, which started in 1984 with almost 6000 households in ‘SOEP West'. The ‘SOEP East’ sample, with a further 2000-odd households, was added in 1990. It now provides information on all household members living in the old and new German states, including Germans, foreigners and recent immigrants to Germany. The ECHP is a survey based on a standardised questionnaire that involves annual interviewing of a representative panel of households and individuals in each country. The total duration of the ECHP was 8 years, running from 1994 to 2001. We use the data for Denmark in this analysis. (Background information on the panels is available in supplementary appendix table A2, available online only).
The analysis samples comprise household heads and partners who were of working age over the course of the study, that is, 25–57 years old at the beginning of the observation period, with complete covariate data in 1994 and health data for at least one wave from 1995 to 2001. The upper age limit for women at baseline in the British sample was 52 years because most retire at age 60 years. These selection criteria yielded the following sample sizes: 5202 from the PSID; 4477 from the BHPS; 4991 from the SOEP and 3553 from the Danish panel of the ECHP.
Measures
Self-rated health
In the USA, household heads were asked ‘Would you say your health in general is excellent, very good, good, fair, poor?’ Household heads were also asked to rate the health of their partners, when applicable. This question was asked each year from 1995 to 2001, excluding 1998 and 2000. In Britain, respondents were asked annually from 1994 to 2001: ‘Please think back over the last 12 months about how your health has been. Compared to people of your own age, would you say that your health has on the whole been excellent, good, fair, poor, very poor?’ In 1999, British respondents were asked a different question as part of the Short Form 36 so that wave was not included in the analysis. In Germany, the respondents were asked each year ‘How would you describe your current health: very good, good, satisfactory, poor, bad?’ and in Denmark the question was ‘How is your health in general: very good, good, fair, bad, very bad?’, again asked annually.
The following social predictors of health were measured in 1994. Additional details are given in table A3 of the supplementary appendix, available online only.
Cohabitation status
Respondents were classified as: married or living as a couple (reference group); formerly partnered (widowed, divorced, or separated); or single.
Education
Using the 1997 International Standard Classification of Education (ISCED-97) scheme,14 education is categorised as lower secondary (levels 0–2), upper secondary (level 3), or post secondary education (levels 4–7; reference category).
Minority status
Responses to questions on race, ethnicity and country of origin were used to code majority (reference) or minority status.
Gender
A measure of gender was coded 1 for women and 0 for men (reference).
Employment status
Current employment status was categorised as employed (reference), unemployed and seeking work, or out of the labour force.
Occupational class
Current or most recent occupational class was coded 1 if a respondent had a routine or semiroutine occupation and 0 otherwise. The variable was derived from 1988 International Standard Classification of Occupations (ICSO-88) major groupings (8 and 9 coded 1; all others coded 0).15 In order to include in the analysis the large numbers of cases with missing occupational data (primarily women, especially in the USA), a variable representing missing information on occupation was coded 1 if missing and 0 otherwise.
Income
Income was coded 1 if less than the country-specific median adjusted household income for the entire population and 0 otherwise.
Age
Continuous measures of age and age squared at wave 0 were centred on 40 years of age.
Analysis
Latent growth curve (LGC) methodology was used to examine trajectories of change over time in self-rated health.16 17 The model assumes a continuum underlying the observed ordinal measures of self-rated health. That is, given a vector y of observed repeated measures, there is a corresponding continuous underlying variable vector y*. The LGC are then estimated for the y* rather than the original y. Latent factors representing the intercept (baseline status) and slope (rate of change) components are extracted from the observations across time. The metric of y* is a Z-score with the mean at zero. Factor loadings of the latent intercept component to all observations are fixed to 1, and the linear slope component is defined by fixing the parameters from 0 (1995) to 6 (2001). The unconditional model is expressed as:
where
We first examined the LGC model of linear change in self-rated health in each country with the latent intercept and slope regressed on age (centred on 40 years), a quadratic effect of age and main effects for the social variables (all time-invariant and centred on their within-country means). Social variables were removed from the model if their main effects were not significant in any country. This applied to the effects of education, occupational class and household income on the health slope. As social influences on health may depend on age, we then tested two-way interaction terms with age and the retained social variables one at a time. The final conditional model included main effects for the retained social variables and all age interaction terms that were significant in at least one country. Interactions were retained for the effects of minority status by age and employment status by age effects on the intercept. This model is expressed as:
Definitions are as for equations (1)–(3). The intercept and slopes are also functions of baseline factors x (x1=age; x2=age2; x3=female; x4=minority; x5=single; x6=formerly partnered; x7=lower secondary education; x8=upper secondary education; x9=unemployed; x10=out of the labour force; x11=routine occupational class; x12=missing occupational class; x13=household income).
The growth curve models were estimated in Mplus version 5.1,18 using the weighted least squares estimator with mean and variance adjusted χ2 statistic, which computes parameter estimates on the basis of all available data, including the incomplete cases. Inverse probability weights accounted for differential sampling in the four surveys, and standard errors were adjusted for the clustered sample design in the USA, Britain and Germany (direct single-stage sampling was used in Denmark).
Ageing-vector graphs are presented to illustrate graphically the level of self-rated health in each country at baseline and the direction and amount of change throughout the age range of our samples. Ageing-vector graphs reveal age and cohort trends in self-rated health over the 7-year period and any age by cohort interactions.19 The ageing-vector graphs were fitted using Stata 10.2.
Results
Covariate distributions for the four countries are provided in table 1. Between-country differences are evident for the covariates. Model fit statistics and parameter estimates are given in table 2 for the final conditional growth curve model. The χ2 statistic is known to be oversensitive to sample size,21 yet by this criterion the model fits the US data (p=0.46). Values over 0.95 on the confirmatory fit index and Tucker–Lewis index are considered a good fit.22 The root mean square error of approximation gives a measure of the discrepancy in fit per degrees of freedom, with values below 0.05 considered to be a good fit. Based on these criteria, the model is an excellent fit to the data in all four countries.
Distribution of covariates reported in 1994 by country*
Estimates, standard errors and fit statistics for the conditional growth curve model
Given the different wording of the self-rated health question and the different response category labels for each country, it is difficult to make between-country comparisons about mean levels and change over time.23 24 Nevertheless, some observations can be made about the self-rated health trajectories within each country. There were significant declines in self-rated health over the 7-year period, but also significant variance around this average trajectory, for both initial health and its rate of change. The vector graphs (figure 1) show changes within birth cohorts as they age over 6 years, and changes in age-specific health between cohorts who reach the same age at different periods in time. The arrows represent the predicted origin and change in health for 2-year birth cohorts, starting with those aged 25 years and ending with those aged 57 years in 1994.
Ageing-vector graphs for (A) USA, (B) Britain, (C) Germany and (D) Denmark at average* covariate values. *Mean gender, minority status, marital status, employment status, education; median income. SRH, self-rated health.
The vector graphs for the USA and Germany show that self-rated health remained relatively stable for young adults, declined as adults became middle aged and then became more stable again. The graphs for Britain and Denmark indicate a steady decline throughout working life. The ageing vectors also suggest that in all countries except Denmark there was little change between cohorts in age-specific health. The Danish model implies an unfavourable trend in self-rated health: ratings were lower for persons of a given age in 2001 than for persons of the same age in 1995. By contrast, the graphs illustrate a favourable trend in self-rated health for those at the beginning of their working life in the USA. There is also evidence that the wording of the self-rated question could be influencing the shape of the trajectories. The British graph shows a less steep age-related decline in health, consistent with being asked to rate health ‘compared to people of your own age’.25 26 Relative to Germany and Denmark, the USA shows slower declines in health. It is possible that the more positive response labels in the USA elicit answers that are biased towards better health.27
The regression estimates in table 2 provide information on the determinants of cross-sectional and longitudinal health. These vary cross-nationally. For the USA, all the covariates with the exception of gender were independently related to the state of health at baseline. For Britain, only gender and cohabitation status had no effect independent of the other covariates, whereas education had a weaker relationship with baseline health than in the USA. The German data showed weaker covariate associations with initial health than was the case for the USA or Britain, and there was even less evidence that the social covariates had an impact on self-rated health in Denmark: unemployment, occupational class and income in 1994 had substantially smaller effects on health 1 year later than they did in the other three countries. Conversely, education had a stronger association with baseline health in Denmark than elsewhere. The interactions with age showed that in each country, the negative relationship between baseline health and being out of the labour force was stronger for older respondents, especially in Britain and Denmark. American and British minority respondents who were older at baseline also had poorer initial health than their younger compatriots.
For all countries, the social covariates had a much weaker relationship with longitudinal health than with baseline health. In the USA, being a woman and of minority status increased the rate of decline. Respondents in the USA and Britain who were out of the labour force in 1994 had slower declines in health, on average. The covariates had no independent association with the rate of health change in Germany. Health declined faster for those who were no longer in a cohabiting relationship in Denmark.
To compare the joint effect of the social influences cross-nationally, we constructed health trajectories for individuals with distinct types of social profiles. Figure 2 gives mean trajectories for three groups distinguished on the basis of advantaged, average and disadvantaged social profiles (profile specifications are provided in the footnote to the figure). For all four countries, there is little difference in the ageing trajectories for those who were advantaged and average on a range of social characteristics. By contrast, being disadvantaged has a strong, and cumulative, effect on ageing health trajectories. Differences were already apparent at 25 years of age in the USA and Britain, and gaps widened with age, especially in the USA and Denmark. At the start of working life, inequalities were most marked in the USA and least evident in Germany and Denmark. By the end of working life, Germany had the smallest health gap between the advantaged and disadvantaged.
Growth curves by levels of advantage* for (A) USA, (B) Britain, (C) Germany and (D) Denmark. *Advantaged: male, majority status, cohabiting with a partner, educated to tertiary level, employed in a non-routine occupation, and with an adjusted household income above the within-country median; Average: see footnote to figure 1; Disadvantaged: female, minority status, no longer living with a partner, lower secondary education, unemployed, in a routine occupational class, and an income below the median adjusted household income. SRH, self-rated health.
Discussion
The results of this study add to a small but growing body of research on self-rated health trajectories,28–32 and offer four new insights on the comparative health of working-age populations. First, they present a differentiated profile of long-term health; although on average self-rated health declined with age, this did not apply to everyone. A second contribution is that we characterised long-term health trajectories in structural terms, and found that social markers signifying differential opportunities and privileges affected health trajectories in expected directions: those with better social circumstances were more likely to experience good health trajectories and less likely to see their health decline over time, compared with their more disadvantaged counterparts. However, social position better distinguished individuals' initial health than it did their changes in health.
Third, differences in the socioeconomic gradient in dynamic health in the four countries raise potentially very fruitful questions about the ways in which a state's policies and practices affect population health. Within each country, there was a large discrepancy in health between disadvantaged and ‘average’ individuals, but only a small difference between those in advantaged and average circumstances. Banks et al33 also found that there was a threshold below which the relationship between (lack of) income and poor self-rated health was stronger. There is a need for further work to determine whether there is a critical cut-point in the accumulation of social risks or a particular constellation of risks that put individual health in jeopardy.
Fourth, even more striking was the finding that by far the strongest social gradients in health dynamics were seen in the USA and the weakest in Germany, whereas positions between these two extremes were occupied by Britain followed by Denmark. This result was consistent with our expectation that the social gradient in health dynamics would be strongest in the USA and Britain. Welfare regime theory did not lead us to expect that the gradient would be weaker in Germany than in Denmark, although existing empirical evidence is in keeping with our results.34 35 Danish findings of cohort differences at older ages and widening inequalities with age are consistent with monetarist changes that took place in that country over the period covered by our analysis. Beginning in 1993, the duration of unemployment benefits was reduced and eligibility was restricted, early retirement allowances were cut, and tax changes disproportionably affected those on lower incomes.36 Welfare typologies are useful organising paradigms but, like health, they are not static phenomena, and Denmark's ‘flexicurity’ labour market may mean that it is no longer a typical exemplar of a non-right hegemony.37 38 Indeed, Denmark's rising educational inequalities in health have been attributed to the marginalisation of those with less education following the change to flexicurity policies,39 40 and other work supports our finding of larger education effects in Denmark than Britain.41
Policy implications
It is difficult to isolate the specific policy instruments that might have contributed to the social structuring of health change. Nevertheless, in line with other studies,28 42 we find that social inequalities in health are already established at baseline in the less egalitarian countries and appear quite resistant to change. Research and policy need to target the early years of the life-course before these differences arise. Faster declines in health among women and minority citizens of the USA suggest that support for these groups is lacking in liberal welfare states. These declines were independent of baseline social conditions. It may thus be that women and minorities experience more detrimental changes in social circumstances following ill health in the USA, contributing to a spiral of disadvantage. It is less clear why formerly partnered individuals in Denmark appear to have poorer long-term health protection. The long-term health repercussions of family breakdown have been associated with the ensuing increased risk of poverty,43 44 and exploratory analyses support this explanation as poor health outcomes are confined to formerly partnered individuals with the least education. This may be another example of the unintended consequences of Denmark's flexicurity system. Finally, it is noteworthy that health declines are slowed for those who have left the labour force. Other research suggests that leaving work allows individuals to recover from poor health and increases their chances of being able to return to the work force at a later date.45
Limitations
The first limitation is that the health measure differed between countries. For this reason, we avoided making direct cross-national comparisons of the health trajectories. However, it should be noted that even with identical questions and response categories for self-rated health we cannot assume equivalence,46 because national characteristics and cultural influences on response styles may bias results. Second, although great care was taken to ensure comparability of the social covariates, the measure of minority status may not have the same meaning in each country, given different migration histories. Nevertheless, the general pattern of findings is unlikely to be influenced by such divergences as may remain. In light of this, we concentrate on comparative patterns rather than on specific effect sizes. Third, other measures of health might have offered different insights, especially in relation to the study of change. It cannot be assumed that a multidimensional measure of health will exhibit the same trajectory patterns as self-assessed health.47 Fourth, without assessing the health impact of specific risk management policies, our conclusions about the role of the state are only tentative in nature. Finally, other social profiles to represent our advantaged and disadvantaged social groups might have highlighted different contrasts between health trajectories in the four countries.
Conclusion
In closing, we return to the question of whether we see national differences in health dynamics that reflect the strengths (or weaknesses) of social safety nets and redistributive policies. The finding of less steep social gradients in long-term health in countries classified as having more equalising policies and higher social expenditures is consistent with this notion. In order to understand better these relationships between social circumstances and health, we need more research on the specific ways in which individual lives are linked to state policies and programmes. This is critical for developing public policies that effectively reduce the differential traces of unfavourable circumstances on people's bodies.
What is already known on this subject
People from disadvantaged households have poorer self-rated health than those who are better off.
International research has shown that socioeconomic inequalities in health differ across different types of welfare states (welfare state regimes).
Less well researched is the relationship of social disadvantage with health change in different welfare contexts.
What this study adds
The relationship between disadvantage and self-rated health varies by welfare state type. Health dynamics were most strongly patterned by social circumstances in the liberal welfare state and least in the conservative welfare state.
On average, self-rated health declined with age. Social position better predicted initial health state than the rate of decline.
There were large discrepancies between disadvantaged and ‘average’ individuals' health trajectories, but only small differences between the trajectories of those in average and advantaged circumstances.
Acknowledgments
The data used in this paper were made available through the UK Data Archive. The data were originally collected by the ESRC Research Centre on Micro-social Change within the Institute for Social and Economic Research at the University of Essex. Neither the original collectors of the data nor the archive bear any responsibility for the analyses or interpretations presented here.
References
Footnotes
Funding The research for this paper was supported by funding from the Social Sciences and Humanities Research Council of Canada (grant 410-07-0913) and the UK Economic and Social Research Council (grant RES-596-28-0001).
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.