Poverty or income inequality as predictor of mortality: longitudinal cohort study
BMJ 1997; 314 doi: https://doi.org/10.1136/bmj.314.7096.1724 (Published 14 June 1997) Cite this as: BMJ 1997;314:1724- a Department of Family Medicine, University of Rochester School of Medicine and Dentistry, Rochester, NY 14620-2399, USA
- Correspondence to: Professor Fiscella
- Accepted 8 April 1997
Abstract
Objective: To determine the effect of inequality in income between communities independent of household income on individual all cause mortality in the United States.
Design: Longitudinal cohort study.
Subjects: A nationally representative sample of 14 407 people aged 25-74 years in the United States from the first national health and nutrition examination survey.
Setting: Subjects were followed from initial interview in 1971-5 until 1987. Complete follow up information was available for 92.2% of the sample.
Main outcome measures: Relation between both household income and income inequality in community of residence and individual all cause mortality at follow up was examined with Cox proportional hazards survival analysis.
Results: Community income inequality showed a significant association with subsequent community mortality, and with individual mortality after adjustment for age, sex, and mean income in the community of residence. After adjustment for individual household income, however, the association with mortality was lost.
Conclusions: In this nationally representative American sample, family income, but not community income inequality, independently predicts mortality. Previously reported ecological associations between income inequality and mortality may reflect confounding between individual family income and mortality.
Key messages
Ecological studies have documented a powerful relation between national, state, or community indices of inequality and mortality at the population level
These studies have not adequately controlled for confounding by family income measured at the individual level
This study, using data from a nationally representative sample from the United States, showed no relation between community income inequality and mortality after adjustment for family income
Poverty, not community income inequality, determines subsequent mortality; further study is needed to determine whether these findings apply to national or state-wide income inequality
Introduction
Studies have documented the powerful association between a person's socioeconomic status and mortality.1 2 3 4 Recently, ecological studies have suggested that income inequality is also correlated with overall mortality.5 6 Wilkinson reported a correlation of -0.81 between national income inequality among 11 industrialised countries and national life expectancy after controlling for gross national product per head.7 Comparable findings were recently reported using data at state level from the United States. Kaplan et al noted that state income inequality adjusted for state median income was significantly correlated (r=0.62) with all cause mortality, age specific mortality, low birth rate, homicide, violent crime, work disability, expenditures on medical care, and police protection.8 Kennedy et al reported that state income inequality adjusted for levels of poverty was strongly correlated (r=0.54) with age adjusted total mortality, infant mortality, coronary heart disease, malignant neoplasms, and homicide.9 Ben-Shlomo et al showed significant effects for income inequality measured at the British ward level on area mortality.10
These ecological or population level studies suggest that the relation between income and mortality in developed countries is a relative phenomenon. In other words, income inequality between countries, states, or communities, is more strongly associated with health than is poverty or mean per capita income. Income equality may be associated with mortality in several ways. Firstly, income inequality may affect health via cognitive processes such as perceived deprivation that promote hopelessness, hostility, or risk taking behaviour.11 12 13 14 15 16 Secondly, income inequality may be a measure or cause, or both, of social forces such as reduced social cohesion that affect health.17 Thirdly, income inequality may be a marker of a government's underinvestment in human resources.18 Lastly, the association between income inequality and mortality may simply represent confounding by family income at the individual level.
Studies reporting an association between income inequality and mortality have provided limited insight into the nature of this relation. Conclusions from previous studies are limited by potential “ecological fallacy” because they cannot adequately control for confounding at the individual level.19 In other words, the observed relation between income inequality and mortality observed at a population level may simply represent inadequately measured rates of income differences at the individual level. No published studies have specifically examined the effect of income inequality on mortality after adjustment for income at the individual level.
We examined whether inequality in income between communities predicts future individual mortality independent of family income. We used data from a nationally representative prospective cohort from the United States to assess whether residence in communities with greater income inequality was independently associated with mortality at follow up.
Methods
Source of data
The first national health and nutrition examination survey (NHANES I), conducted between 1971 and 1975 in the United States, collected sociodemographic data from multiple national probability samples of the civilian non-institutionalised population of adults aged 25 to 74 years.20 21 The epidemiological follow up study (NHEFS) collected mortality data through follow up surveys conducted in 1982, 1984, 1986, and 1987.22 23 Follow up data were derived from interview surveys, medical records from healthcare institutions, and all death certificates. The age, race, and sex specific mortality of the follow up cohort is comparable to that experienced by the American population.23 In all, 14 407 people were in the original survey, 3.4% of whom had missing information on family income; mortality status at follow up was ascertained on 95.8% of the people for whom such information was available.
The first survey used multistage stratified probability samples of people from 105 areas or primary sampling units in the United States. The areas approximated to counties or combined county areas. People living in areas of poverty, women of childbearing age, and elderly people were oversampled (surveyed at disproportionately higher rates). A mean of 131 (range 48-323) people were surveyed in each primary sampling unit. The revised weights provided on the 1987 follow up study's “public use” tapes were used to adjust for survey oversampling and non-response to yield population estimates for each community surveyed.
Measure of family income
Household income was assessed through response to a single question in the first national health and nutrition examination survey. The question asked subjects to say which of 12 income groups represented their total family income for the previous 12 months, including all sources of income, such as wages, salaries, social security or retirement benefits, help from relatives, and rent from property. The income categories ranged from under $1000 to $25 000 and over. Subjects were assigned to the mean value within their income category. Subjects in the highest income category (4.6%) were assigned a mean value by extrapolation.
Measure of income inequality
Several indices of community income inequality have been described, though the relation of each index to mortality risk is similar.9 We used an index that estimates the proportion of total income earned by the poorer half of the population in the area. The denominator for this index is the total aggregate income in the community (primary sampling unit), and the numerator is the aggregate income in the community earned by the poorer half of the population. The income inequality within the communities ranged from 0.18 to 0.37.
Statistical analysis
Multilevel modelling is required to avoid the problems of clustering within a group.24 We used the statistical package sudaan, which uses a Taylor series approximation method to compute variances that allow adjustment for multistage probability sampling.25 A Cox proportional hazard survival analysis was performed that included the index of community income inequality, household income, family size, sex, and age as covariates in the predictive model for mortality. The assumptions of the model were tested and found valid.
Results
Older age, residence at the time of the interview in a community with greater income inequality, and lower mean community income were all associated with the proportion of people in the community dying during follow up (table 1). Survival analysis showed that survival adjusted for age, sex, and family size was associated with income inequality (hazard ratio=0.23, 95% confidence interval 0.06 to 0.86); additional adjustment for mean community income did not greatly affect this relation (0.31, 0.10 to 0.90). However, after adjustment for household income, no significant relation between income inequality and mortality was evident (table 2). An analysis that excluded the income inequality measure showed no change in the effect size of income on mortality (0.97, 0.96 to 0.98).
Discussion
Analyses of data from a nationally representative prospective American cohort study show that individually measured family income strongly confounds the relation between community income inequality and mortality. Although aggregated data at the individual level simulated the findings of previous ecological studies,7 8 9 10 community income inequality did not independently predict mortality after adjustment for family income. Conversely, exclusion of the variable of community income inequality did not affect the relation between family income and subsequent mortality. These findings suggest that the effect of income inequality reported in ecological studies may result from confounding by income at the individual level.
Our findings imply that income, as a measure of access to resources, and not relative inequality, better explains the relation between income and mortality. Psychological or social factors related to income inequality may nevertheless have important health effects, as Wilkinson suggests.17 We believe, however, that existing studies have not adequately tested this hypothesis. Future studies of the inequality hypothesis should control for income at the individual level and use direct measures of factors such as social cohesion and perceived socioeconomic deprivation to advance our understanding of the relative contribution of these factors to health.
These findings are subject to several important caveats. The appropriate unit of analysis for measuring income inequality is not known. Ecological studies have shown effects of inequality at the national level,7 American state level,8 9 and British ward level.10 We used community (mostly American counties) income inequality as our unit of analysis and showed significant effects for income inequality in a simulated ecological analysis. The validity of our analysis is supported by the significant correlation between community income inequality and community mortality rates after adjustment for mean community income. Although the magnitude of this correlation was smaller than that reported in previous ecological studies, this may represent less non-differential misclassification bias.26 Because our ecological analysis used prospective data that had been carefully collected at the individual level, there may have been less misclassification than in ecological studies that have used national or state cross sectional data. Brenner et al showed that non-differential exposure misclassification in ecological studies as opposed to individual level studies can lead to significant overestimation of effects.26 Further support for the validity of our findings is provided by preliminary data from the panel on income dynamics from the United States. In this study state income inequality showed no effect on mortality after individually measured income was controlled for (G Duncan, personal communication).
Limitations
Still larger units of analysis might yield different results. This hypothesis is particularly plausible if it is assumed that income inequality is simply a proxy for national policies that promote general social welfare and health.8 27 However, if income inequality is assumed to influence health directly via cognitive processes then the appropriate unit remains speculative. For example, if people judge their socioeconomic status relative to that of their neighbours, then the community may be the appropriate level of analysis, but if they use national media sources as a frame of reference then analyses should focus on the national level. A stronger conceptual framework is needed to guide future studies in this area and also to provide direct testing of the hypothesis that individual perceptions of relative socioeconomic standing influence health.
These findings are limited by the area sampling methodology used by the first national health and nutrition examination survey. Although the survey was designed as a representative sample of the American population, sampling within communities was not random. The cluster sampling strategy underestimates the true variability in each community, thus underestimating community income inequality. Despite this bias, we found significant effects for income inequality in the ecological but not individual level analysis.
Our analysis does not account for an individual's relocation from one community to another during the study. Nor does our analysis account for increasing levels of income inequality in the United States during the study.28 Such misclassification bias tends to overestimate the effects observed in an ecological study and to underestimate the effects observed in an individual level analysis. This bias operates similarly for family income. Both family size and income may change considerably over time, resulting in further misclassification bias. In addition, a selective bias resulting from greater loss to follow up among poorer people tends to understate the effect of poverty on mortality. Although each of these biases underestimates the effect of family income on mortality, we none the less found significant effects for family income.
Colinearity between family income and community income inequality may have masked the independent effect of community income inequality on mortality. The finding that the relation between family income and mortality was unaffected by adjustment for community income inequality suggests, at least, that family income is a far more powerful predictor of health status than community income inequality. Although our analysis does not exclude a modest effect of community income inequality on health, these findings militate against a large effect.
Conclusion
These findings suggest that community income inequality does not have large effects on mortality independent of the effects of family income. However, income inequality and family income are closely intertwined. Countries, states, or communities with large income inequalities are likely to have more poverty. Countries whose explicit goal has been eradication of poverty also have less income inequality. Thus, whether public policy focuses primarily on the elimination of poverty or on reduction in income disparity, neither goal is likely to be achieved in the absence of the other.
Acknowledgments
Funding source: None
Conflict of interest: None