Social inequalities in mortality: changes in the relative importance of income, education and household size over a 27-year period

Abstract

Background: The aim of this paper was to measure socio-economic inequalities in mortality over a 27-year period, and estimate the simultaneous effects of education and income adjusted for changing proportions and potential confounders. Methods: Census data in 1970, 1980 and 1990 with information about income, educational level and household size were linked to mortality records at the individual level and analysed with absolute mortality rates, Poisson regression (univariate and multivariate) and Relative Index of Inequality (RII). Results: Mortality differences increased between income quartiles and educational groups as well as between singles and non-singles. RII between income quartiles increased from 1.6 to 3.6 for men and 1.7 to 2.7 for women while RII between the educational groups increased from 1.6 to 2.8 for men and 1.5 to 2.1 for women. However, RII derived from the adjusted multivariate regression analysis was 1.8 (men) and 1.9 (women) between the income groups in 1990 and 2.2 (men) and 1.7 (women) between the educational groups. In the 1990s current income was more important than previous income, and the rate ratios (RR) increased for groups dropping into the lowest income quartile between two censuses. Conclusion: Low education and single status have become relatively more important risk factors for mortality over time. Confounder effects of education and household size could partly explain the seemingly large increase in inequalities between income groups. Results may also indicate a stronger reverse causation (poor health causing low income) over time.

Social inequalities in health have increased during the last decades in Norway as well as in other Western countries.^1–5 The Norwegian welfare system has been known for its universal coverage and high benefit levels. However, major welfare reforms have taken place during the last three decades, and there has been a general increase in prosperity as well as educational level. A higher proportion of the population is either single or single-headed, and married women are increasingly working outside the home. In order to study the time dimension more thoroughly, we wanted to study socio-economic inequalities in mortality from the 1970s to the 1990s and analyse the simultaneous effects of income and educational level on mortality over time, adjusting for changes in household size and geographical distribution of income.

We have excluded occupational class as a socio-economic indicator for several reasons: lack of data for parts of the observed period, classification problems of the economically inactive and large changes in the labour market. The use of income is hampered by lack of knowledge on how to measure material standard of living and the mechanisms by which relative economic deprivation has a direct or an indirect effect on health, and vice versa.^6–13 We have grouped income into quartiles of adjusted household income in each census.^7,8,14 The economic position of singles has also become less favourable and the mortality gap between singles and non-singles has increased.¹⁵ Household size is considered to be a potential confounder and therefore included in these analyses.

We have considered household income (as a measure of material living conditions) and education (qualifications) as largely independent variables.^6,9 The correlation coefficient between the two has changed from 0.3 (1970) to 0.2 (1990).¹ Given the considerable changes in the society mentioned, these socio-economic variables are likely to differ in their relative importance to health over time. We have studied this hypothesis by comparing the effects of income and education in univariate and multivariate analysis. In addition, we have tried to move beyond the static perspective of a cross-sectional analysis by studying the effects of income dynamics between the 1970 and the 1980 census and between the 1980 and 1990 census.

Methods

Material

We have linked information about all individuals aged 45–59 years in 1970, 1980 and 1990 to census data, official registers on education, income, family size and death at Statistics Norway. The individuals were followed up to death or censored after 7 years. The numbers of individuals in each category are given in table 1. Altogether, 73,826 deaths were observed over 13,911,687 observation years. Less than one per cent of the individuals were excluded due to incomplete information about education, income or household size. We have restricted the study to the age group 45–59 years for several reasons: Firstly, to avoid confounding effects of retirement and secondly, because a more narrow age group than the more commonly used 25–64 will limit the overlap of the same individuals from one observation period to the next, and they will also have more similar major causes of death (chronic diseases).

Households

The censuses in 1970 and 1980 covered 100% of the population while the 1990 census covered only 28%. The Family Registry, established in 1986, is therefore used as the source of information about family size in 1990. Households with more than 12 individuals, mainly institutions, were excluded in 1970 and 1980 (0.45%). Household size was grouped into two categories: one person households (singles) and households with 2–12 persons (non-singles).

Residential area

Municipality of residence was categorised into urban or rural. The municipalities were defined as rural if the normal travelling time to a city of 50,000 inhabitants or more exceeded 90 minutes. The number of cities with more than 50,000 inhabitants has grown during the observation period.

Education

Years of formal education were grouped into four levels: 0–9 years, 10–12 years, 13–16 years and 17 years or more.

Income

Gross income (including tax) was used as the income measure in the 1970 census while net income (gross income plus social benefits and child support minus tax) was used as the income measure in 1980 and 1990. Individuals with zero income in 1980 or 1990 were excluded from the analysis (<0.3%) because zero income was considered to be a misclassification at that time. In 1970, however, around 3% had zero income; the majority with zero taxable income. These individuals were not excluded from the analysis.

The income measure used, adjusted household income, is defined as total household income divided by the square root of the number of individuals in each household^7,10,14 and grouped into four quartiles: low income, 0–24%; low-middle income, 25–49%; high-middle income, 50–74%; high income, 75–100%. In the analysis of income dynamics, income is limited to two groups: income group 1 (low income, 0–24 percentile) and income group 2 (higher income, 25–100 percentile). Because adjusted household income initially was systematically lower in rural (where mortality is lower) than in urban areas,^1,15,16 we have used separate income quartiles for rural and urban municipalities to avoid underestimating the impact of income on mortality. We have also used separate income quartiles for men and women and for single and non-single households for the same reasons. The use of separate income quartiles for singles and non-singles yields a statistically better model (less deviance) and is easier to interpret than a model with several interaction terms.

Statistics

Age-adjusted mortality rates were calculated for men and women separately and stratified by income, educational level and household size using five-year age bands and the total sum in all three censuses as standard population (table 1). The data were analysed using Poisson regression models and the statistical software package EGRET.¹⁷

The Relative Index of Inequality (RII) was calculated based on age-adjusted mortality rates for the income and the educational groups (table 2). The RII is a statistical measure that takes into account both the population size and the relative position of the groups. It does so by regressing the mortality rate of the socio-economic group on a specific measure of their relative position: the proportion of the population that has a higher position in the social hierarchy. A large score on RII implies large differences in mortality between low and high positions in the social hierarchy.

In table 3 mortality rate ratios (RRs) between the income groups, the educational groups and between singles and non-singles were calculated adjusted for age only (Model 1). In Model 2 RRs were calculated adjusting simultaneously for all covariates.

In table 4 we have observed income dynamics between the 1970 and 1980 census and between the 1980 and 1990 census. Income levels were limited to two groups as previously described, and transitions between the two income groups were recorded. The different transitions and non-transitions were used as variables in a regression analysis (in addition to the variables in table 3).

Results

Trends over time in univariate analyses

Mortality rates were generally stable between 1970 and 1980, but fell dramatically between the 1980s and the 1990s (table 1). The differences in mortality rates between the income and educational groups increased most from the second to the third observation period, while the absolute differences between singles and non-singles increased most from the first to the second observation period (table 1). Singles have higher mortality rates than the low income and the low educated groups in all observed periods. It is interesting to note the relative stability in mortality rates among men and women in the lowest income quartile over the three observed periods in contrast to the steep decline in mortality rates among the higher income groups between the 1980s and 1990s. Table 2 shows that the relative inequalities (RII) became essentially larger between the income groups (RII was 3.6 for men and 2.7 for women) than between the educational groups (RII was 2.8 for men and 2.1 for women) in the 1990s.

Trends in multivariate analyses

A comparison of Model 1 (univariate) and Model 2 (multivariate) in table 3 shows that the increasing RRs for singles cannot be explained by the other covariates. Neither can much of the increasing RRs among the low educated over time, except for a small fraction of the RRs for men in 1990–1997. However, an increasing part of the differences in RRs between the income groups can be explained by the other covariates. Nevertheless, adjusted for the effects of education, household size and geography, RRs between the income groups still increase from the 1980s to the 1990s.

In the multivariate model the differences in RR were essentially higher between educational groups than between income groups in the 1990s. The differences in RR for the highest and lowest educational group of women were not statistically significant in the 1970s and the 1980s. Table 1 shows that only a small fraction of women had 17 years or more education in 1970 (0.4%) and 1980 (0.6%).

The effects of changes in income group from one census to the next

Men with income rising from Group 1 (0–24 percentile) in 1980 to Group 2 (25–100 percentile) in 1990 had RR = 0.99 in the 1990s (RR = 1.06 for women) compared to those staying in income Group 2 at both censuses. RR for those dropping from higher income in 1980 to lowest income in 1990 was 1.32 (RR = 1.39 for women). One decade earlier these two groups had a very similar RR. People dropping from Group 2 in 1980 to Group 1 in 1990 had just as high mortality ratios in the 1990s as those being in the lowest income quartile at both censuses. There were 75,892 women and 75,433 men in the lowest income quartile in 1990, whereas 45,179 women and 50,101 men had descended from the highest three income quartiles in 1980 (data not shown).

Discussion

All indicators and all methods show increased inequalities in mortality from the 1970s to the 1990s. The estimated trend is unlikely to be a transient cohort phenomenon due to the long observation period, and the fact that <30% of the 45–59-year-olds in the 1980s were recorded as 45–59 years in the 1970 census, and they only contributed to about 20% of the observation years.

Inequalities in mortality related to educational level

Tables 1 and 3 show increased inequalities between the educational groups from the 1970s to the 1990s. Measured as RII (table 2), and thereby adjusting for the changing sizes of the educational groups, the inequalities also increased considerably over time. For calculating RII the relationship between mortality and the explanatory variable is supposed to be linear. Since the relationship between income/education and mortality was not always linear, the Variation Coefficient Index of Inequality (VCII), defined as the weighted sum of the average distance to the mean for the different covariant groups divided by the mean, was also calculated (data not shown). These results, however, were essentially identical to the calculated RIIs.

Table 3 shows that very little of the differences in RR between the educational groups could be explained by confounding effects of income, single households or residential area. Therefore, it seems that educational level really has become more important to health during recent decades. Factors that may be of importance are the considerable reduction of unskilled jobs in this society and the increasing flow of information that may require a higher degree of literacy. Lifestyle risk factors such as daily cigarette smoking, diet and physical inactivity are closely linked to educational level in Norway.^18–21 Data on smoking patterns show increasing differences in daily consumption between educational groups over time.²¹ The latter may contribute to further increases in inequalities in years to come.

Inequalities in mortality related to income level

Inequalities in mortality between the income groups increased from the 1980s to the 1990s, both measured as absolute mortality rates, RII and RR. Table 2 shows larger inequalities measured as RII between income groups than between educational groups in the 1990s. RII is calculated from age-adjusted mortality rates, and since a considerable part of the increased differences in age-adjusted mortality rates between the income groups is explained by confounding effects of education and household size, RII should be interpreted as an unadjusted measure. If we calculate RII for men in the 1990s based on the estimated RR in table 3 (Model 2), RII would be 1.8 between the income groups for men (1.9 for women) and 2.2 between the educational groups for men (1.7 for women).

The use of common income quartiles for singles and non-singles, instead of separate, would select more singles (with higher mortality rates) into the low-income group over time,^1,15,22 producing even larger inequalities in the 1990s. Also, the use of adjusted household income and the method used to adjust for the number of individuals in the household^7,8,14 may also be a potential confounder with large changes in family size; particularly the increasing numbers of singles as well as double-income families and fewer children^22,23 over time. In our data the highest income quartile earned approximately 4.6 times the median income of the lowest income quartile in 1970, but only 3.4 times this income in 1990. We must, however, be aware of the changed income measure from 1970 to 1980. According to Statistics Norway,¹⁶ the inequalities in income have been quite stable and the absolute income has increased for all groups during this period. Nevertheless, we find increased differences in mortality between the highest and lowest income groups.

The effect of income dynamics

By observing income over three successive censuses, we can study the effects of income stability as well as changes in different directions between 1970 and 1980 and between 1980 and 1990. Table 4 shows that the direction of the transition of income is important; more important to RRs in the 1990s than in the 1980s. Current income also seems to be more important in the 1990s than prior income, which is not in line which the findings of Benzeval and Judge.¹¹ Seen in combination with the stable mortality rates among people in the lowest income quartile from the 1970s to the 1990s (table 1), these results may indicate increased selection: poor health causing low income. On the other hand, both income and mortality may be sensitive to other changes in society. Both the changing family patterns and the changing labour market may be factors of importance. The income measure (adjusted household income) and the reduced stability of families may be confounders in the relationship between previous and current income and mortality presented in table 4. Selection effects are likely to be stronger if individual income rather than household income were used in the analysis. Changes in the labour market are also likely to play a major role by the increasing exclusion of people with handicaps or health problems as well as those with low education.¹⁶ There has been an extensive use of disability pension among middle-aged Norwegians during the last decades for groups with problems related to health as well as to unemployment. In 1997 approximately one out of every sixth person 50–59 years of age was a disability pensioner.²⁴ The welfare state might select more people with health and/or work related problems into the lowest income quartile by having a low threshold for publicly funded disability pension. Health-related exclusion from the labour market and adverse financial conditions of people with limiting longstanding illness have been demonstrated in other Scandinavian countries.^13,25 However, the exclusion of people, mainly low educated people with mental health, muscle and skeletal problems, from the labour market may also increase their risk of mortality.^24,26 The lack of information about employment status and disability pension in our data prevents us from controlling for these factors.

Inequalities in mortality between single and non-single households

Single residency has become more common (table 1) while relative and absolute differences in mortality between singles and non-singles have increased. Both stronger ‘positive’ selection of healthy individuals into marriage and stronger ‘negative’ selection out of marriage are possible explanations. However, increased divorce rates and a relatively worsened economic situation for singles may also be relevant. We may have overestimated the number of singles in 1990 by using The Family Registry as the source of information. Some cohabitants with lower mortality rates than singles²⁷ are likely to be classified as singles in the 1990 data, producing an underestimation of the mortality differences between singles and non-singles in the 1990s. Swedish studies have demonstrated socio-economic problems and severe ill health among single parents.²⁸ Single parents are categorised as non-singles in our analysis, which may have resulted in higher mortality rates among non-singles and thereby an underestimation of mortality differences between singles and non-singles.

Conclusion

Inequalities in mortality have increased between groups with different income and educational levels as well as between single and non-single households from the 1970s to the 1990s in Norway. According to these analyses, income and education are independent but partly overlapping indicators of socio-economic status with changing relative importance over time. Low education and living in single households have become more important risk factors in recent years—more important even than income. Inequalities increased between income groups as well, but confounder effects of education and household size explained a major part of the increased mortality differences between the income groups. Results may also indicate a stronger reverse causation, poor health causing low income, over time.

Financial support was received from the Norwegian Ministry of Health.

References