Elsevier

Social Science & Medicine

Volume 62, Issue 8, April 2006, Pages 2024-2034
Social Science & Medicine

Shifting dollars, saving lives: What might happen to mortality rates, and socio-economic inequalities in mortality rates, if income was redistributed?

https://doi.org/10.1016/j.socscimed.2005.08.059Get rights and content

Abstract

Personal or household income predicts mortality risk, with each additional dollar of income conferring a slightly smaller decrease in the mortality risk. Regardless of whether levels of income inequality in a society impact on mortality rates over and above this individual-level association (i.e., the ‘income inequality hypothesis’), the current consensus is that narrowing income distributions will probably improve overall health status and reduce socio-economic inequalities in health. Our objective was to quantify this impact in a national population using 1.3 million 25–59-year-old respondents to the New Zealand 1996 census followed-up for mortality over 3 years.

We modelled 10–40% shifts of everyone's income to the mean income (equivalent to 10–40% reductions in the Gini coefficient). The strength of the income–mortality association was modelled using rate ratios from Poisson regression of mortality on the logarithm of equivalised household income, adjusted for confounders of age, marital status, education, car access, and neighbourhood socio-economic deprivation. Overall mortality reduced by 4–13% following 10–40% shifts in everyone's income, respectively. Inequalities in mortality reduced by 12–38% following 10–40% shifts in everyone's income. Sensitivity analyses suggested that halving the strength of the income–mortality association (i.e., assuming our multivariable estimate still overestimated the causal income–mortality association) would result in 2–6% reductions in overall mortality and 6–19% reductions in inequalities in mortality in this New Zealand setting.

Many commentators have noted the non-linear association of income with mortality predicts that narrowing the income distribution will both reduce overall mortality rates and reduce inequalities in mortality. Quantifying such reductions can only be done with considerable uncertainty. Nevertheless, we tentatively suggest that the gains in overall mortality will be modest (although still potentially worthwhile from a policy perspective) and the reductions in inequalities in mortality will be more substantial.

Introduction

There is strong international evidence for lower income being associated with poorer health status (Backlund, Sorlie, & Johnson, 1996; Blakely, Kawachi, Atkinson, & Fawcett, 2004; Bucher & Raglan, 1995; Ecob & Davey Smith, 1999; Lantz et al., 1998; Marmot, 2002; Martikainen, Makela, Koskinen, & Valkonen, 2001; McDonough, Duncan, Williams, & House, 1997; Sorlie, Backlund, & Keller, 1995). Furthermore, there is convincing evidence of a non-linear association of income with mortality such that each extra dollar of income buys a little less health gain (Backlund et al., 1996; Deaton, 2002; Ecob & Davey Smith, 1999; Gravelle, 1998; Subramanian & Kawachi, 2004; Wagstaff & van Doorslaer, 2000). Given this pattern it has been suggested that “raising the incomes of more disadvantaged people will improve the health of poor individuals” and “help reduce health inequalities” (Lynch et al., 2004). Moreover, as the health loss of the rich is expected to be less than the health gain of the poor for redistribution of incomes, the average health status of the population should increase (Gravelle, 1998). Our aim in this paper is to model changes in overall mortality rates and socio-economic inequalities in mortality that might arise from redistribution of income.

It is important to note that we are not addressing the ‘income inequality’ hypothesis, per se, in this paper. That is the hypothesis that a society with more equal income distributions will have better health outcomes for everyone, over and above that predicted by his or her personal (or household) income. This hypothesis posits positive ‘spill over’, contextual or ecologic effects, but is contested (Lynch et al., 2004; Subramanian & Kawachi, 2004). The strongest evidence is at the state-level in the United States, but there are non-confirmatory studies at regional levels in other countries—including New Zealand (Blakely, Atkinson, & O’Dea, 2003). In this current paper we address the health impacts of individuals moving up and down the income–mortality curve as predicted by the individual-level association of income with health, but we do not model a shift in the entire income–mortality association whereby a narrower income distribution confers an additional contextual advantage in lower mortality risks for all income groupings.

It is also important to be cognisant at the outset of the many limitations of the modelling exercise presented in this paper. Researchers of the association of income with health (usually) base their interpretations on the implicit assumption that at least some of the observed association of income with health is causal, and—by extension—that changes in an individual's income should result in some change in health. The implicit recommendation for policy-makers is that income redistribution will reduce inequalities in health. However, when challenged as researchers to quantify the impact of income redistribution on overall population health and inequalities in health, we are not aware of any research that has provided such explicit estimates. We believe it is a legitimate role of researchers to at least estimate the likely health impacts of income redistribution. Indeed, other social epidemiologists are also taking tentative steps on quantitative health impact assessments of various income-related policy options (Cole et al., 2005). Whilst these estimates will inevitably be uncertain, and must come with an ‘uncertainty warning’, in our view the provision of such quantitative estimates sharpen the policy analysis and debate.

What are the key limitations of any modelling exercise of the impact of income redistribution on population health and inequalities in health? Many, although we will address five in particular: asking the right counterfactual or policy-relevant question; life-course determination of health; confounding of the observed income–health association that we base our modelling on; time lags between income change and health change; and the possible deadweight costs to society of income redistribution.

Asking the right counterfactual or policy-relevant question: In this paper, we model the impacts of moving everyone's income some ‘X%’ to the mean income. Policies that are to some extent redistributive are the norm in most developed countries, and setting a counterfactual question about different levels of such redistribution is not unrealistic: measures of income inequality vary between countries or over time within countries (Atkinson, 2003), and government policies directly or indirectly influence income distributions (e.g., taxation and welfare benefit policies). But is this the most likely policy action? Much, but not all, of the increase in income inequality in developed countries since the 1970s is the consequence of increased returns to education (Atkinson, 2003), and the state may not be readily able to undo or off-set these changes by way of taxation or welfare policies. Further, it may be more efficient to address income inequalities not by taxation and income redistribution per se, but by targeted provision of free welfare services such as education and health care. Nevertheless, an estimate of one possible policy mechanism—income redistribution—provides more information for policymaking and debate than hitherto existed.

Life-course determination of health: It is increasingly recognised that adult health is a function of a lifetime of exposures—indeed intergenerational histories (Kuh & Ben-Shlomo, 1997; Kuh, Ben-Shlomo, Lynch, Hallqvist, & Power, 2003). Regarding findings on income and health from longitudinal studies: “long-term income is more important for health than current income; income levels are more significant than income change; persistent poverty is more harmful for health than occasional episodes; and income reductions appear to have a greater effect on health than income increases” (Benzeval & Judge, 2001). Research on the US Panel Study of Income Dynamics, for example, has shown that persistently low income was particularly important for mortality risk, although income instability was also important (McDonough et al., 1997). We can perhaps distil the issues to two key questions: does a change in income cause a change in health, and by how much?; and what is the time lag between any change in income and a change in health status? We address these issues as confounding and time lags in the next two paragraphs.

Confounding: Confounding in epidemiology is defined as the mixing of effects whereby an exposure of interest has an association with an outcome of interest that is (in part at least) due to some correlated variable that predicts the outcome. It plagues observational studies. The gold standard methodology to estimate the unconfounded association of income with health, therefore, would be a randomised trial of income supplementation—however, such a study has not been conducted (Connor, Rodgers, & Priest, 1999). One alternative approach is to control for those variables that may be confounders of the income–health association in observational studies. For example, we have previously found that about half of the age and ethnicity-adjusted association of income with mortality was attributable to confounding factors (Blakely et al., 2004). But such analyses that control for confounders are still prone to error from either measurement error of the confounders or simply not including all potential confounders (Davey Smith & Phillips, 1990; Phillips & Davey Smith, 1991; von Elm & Egger, 2004). Another alternative approach is to use longitudinal studies with repeated measures on individuals that allow an assessment of how much a change in income predicts a change in health. We are aware of one such study that meets this latter requirement (McDonough & Berglund, 2003). Using the US Panel Study of Income Dynamics, McDonough and Berglund estimated self-rated health as a function of persistent poverty, transient poverty and income to needs ratio. Whilst not a highlighted finding of their study, close inspection of their results in Table 6 demonstrates that the coefficient for changing income to needs (0.0101) was approximately 18% of the magnitude of the coefficient for income to needs at baseline (0.0559). That is, the impact of changing income to needs on contemporaneously changing self-rated health was about 20% of the magnitude of the baseline income estimate—controlling for transient and persistent poverty, education, race, marital status and age. Such a result is indicative only: the standard errors of the coefficients were approximately 10% of the coefficient magnitude, and no allowance has been made for time lags. But it does provide an indication that the fraction of the income–health association that is causal for contemporaneously measured self-rated health and in the US setting may be as low as 20%.

Time-lags: We are aware of no reliable quantitative studies of the time lag between income change and change in health status. However, for outcomes such as mortality there must be some elapsed time for an income change to alter one's mortality risk, be it by stress, dietary or other pathways. Some of these pathways (as demonstrated in life course epidemiology) will take decades. However, not all causal mechanisms will take that long. For example, the rapid response of life expectancy to economic and social upheaval in the eastern European countries post 1989 point to the possibility for rapid health responses to socio-economic change (Men, Brennan, Boffetta, & Zaridze, 2003; Notzon et al., 1998).

Deadweight costs to society of income redistribution: Redistributing income is not a cost-free policy in that there are welfare reducing deadweight losses associated with tax-collection and re-distribution systems (Deaton, 2002). That is, whether due to tax avoidance or other inefficiencies, redistribution of income may lower total income or welfare to society. However, there are many complex issues involved (as detailed further in the Discussion section) and so we have used the simplifying assumption that there is no overall change in total income.

In this paper, we were able to use estimates of the income–mortality association adjusted for (measured) confounders, but we were not able to quantitatively explore life-course determination, time lags, and dead weight costs.

Section snippets

Methods

This paper builds on work published elsewhere from the New Zealand Census-Mortality Study (NZCMS) (Blakely et al., 2004). Briefly, the shape and strength of the income–mortality association was estimated among four census-mortality cohorts formed by anonymous and probabilistic record linkage. In this paper, we just use the most recent census-mortality cohort (i.e., 1996–99). The age range was restricted to 25 – 59-year-olds, with the upper limit imposed to avoid problems with people retiring

Results

Table 1 shows the weighted person-years and deaths, and observed rate ratios for each of the ten categories of household income in the 1996–99 cohort. The coefficients for the logarithm of household income from Poisson regression models that adjusted for age and ethnicity were –0.488 and –0.457 for males and females, respectively. The coefficients from the multivariable models were –0.286 and –0.293. (These coefficients were reasonably similar for the earlier cohorts in the NZCMS (1981–84,

Discussion

An important role of social epidemiology is to inform policy debates on reducing inequalities in mortality with, where possible, quantified effects. Many researchers have pointed to the non-linear association of income with mortality as a win-win scenario—narrowing income distributions will both improve overall mortality, and reduce inequalities (Gravelle, 1998; Kawachi, 2000; Lynch et al., 2004; Subramanian & Kawachi, 2004). Our modelling supports this argument, but makes it clear that the

Acknowledgements

June Atkinson assisted with analyses. Anna Matheson, Brian Easton and Philippa Howden-Chapman commented on drafts of the paper. The NZCMS was supported by Health Research Council of New Zealand funding, and receives ongoing funding from the New Zealand Ministry of Health. Both authors are independent of the funding agencies with regard to academic outputs.

References (36)

  • B.L. Cole et al.

    Projected health impact of the Los Angeles City living wage ordinance

    Journal of Epidemiology and Community Health

    (2005)
  • J. Connor et al.

    Randomised studies of income supplementation: A lost opportunity to assess health outcomes

    Journal of Epidemiology and Community Health

    (1999)
  • G. Davey Smith et al.

    Declaring independence: Why we should be cautious

    Journal of Epidemiology and Community Health

    (1990)
  • A. Deaton

    Policy implications of the gradient of health and wealth. An economist asks, would redistributing income improve population health?

    Health Affairs

    (2002)
  • Förster, M., d’Ercole, M. (2005). Income distribution and poverty in OECD countries in the second half of the 1990's....
  • H. Gravelle

    How much of the relation between population mortality and unequal distribution of income is a statistical artefact?

    British Medical Journal

    (1998)
  • I. Kawachi

    Income inequality and health

  • D. Kuh et al.

    A Life Course Approach to Chronic Disease Epidemiology

    (1997)
  • Cited by (20)

    • The potential impact of an implemented income redistribution package on obesity prevalence in New Zealand

      2021, Social Science and Medicine
      Citation Excerpt :

      The consistency by which socioeconomic status is associated with wellbeing has led numerous researchers to call for political action to address the social determinants of health (Marmot et al., 2008). It is challenging to model the health effects of policies amending the broader social determinants of health, and few studies have attempted to do so (Blakely and Wilson, 2006; D'Souza et al., 2008). Disposable income is influenced through tax rates, minimum wage levels, and social welfare, making it the most immediately amenable component of socioeconomic status to policy intervention.

    • Social epidemiology for the 21st century

      2018, Social Science and Medicine
      Citation Excerpt :

      What these critics failed to notice, however, is that the same argument applies to the potential for income redistribution to raise the level of health in society. Some researchers have now attempted to quantify the implied gains in population health as a result of income redistribution (Blakely and Wilson, 2006; Brodish and Hakes, 2016). For example, based on a simulation using data from the National Longitudinal Mortality Study, Brodish & Hakes (2016) estimate that each 10% reduction in income inequality, as assessed by the Gini coefficient at the state??

    • Quantifying the individual-level association between income and mortality risk in the United States using the National Longitudinal Mortality Study

      2016, Social Science and Medicine
      Citation Excerpt :

      Nevertheless, individual income and the magnitude of the absolute effect of income on health remain important pieces of the puzzle. We know of one other paper that attempted to quantify the absolute income effect in a developed country (Blakely and Wilson, 2006). The authors simulated the impact on mortality risk of hypothetical shifts of household income toward the overall mean household income by proportions ranging from 10% to 40%, following all respondents to the New Zealand 1996 census (n = 1.3 million) forward for three years.2

    • Patterns of family doctor decision making in practice context. What are the implications for medical practice variation and social disparities?

      2013, Social Science and Medicine
      Citation Excerpt :

      Social disparities in clinical activity were not clearly evident which confirms other findings (Dixon, Le Grand, Henderson, Murray, & Poteliakhoff, 2007). On balance, therefore, this study suggests that, while ethnic (Blakely et al., 2005) and socio-economic (Blakely & Wilson, 2006) inequalities in health outcomes are clearly evident and marked in New Zealand, as elsewhere (Mackenbach et al., 2008), social selectivity in clinical decision making in everyday primary medical care does not seem to work consistently or strongly to reinforce those wider inequalities. Nevertheless, the greater variability of practitioner decision-making for socially disadvantaged patients, particularly in fee-for-service settings, may provide support for the role that practitioner behaviour and practice context may play in shaping aggregate patterns of practice variation and social disparity.

    View all citing articles on Scopus
    View full text