International Journal of Epidemiology 2003;32:1037-1040
© International Epidemiological Association 2003
Income Inequality and Health |
Response: In defence of the income inequality hypothesis
Lynch, Harper, and Davey Smith’s metaphor of the SS Income Inequality1 is amusing, but we think that a more accurate representation of the current debate in this area would be a kangaroo court, in which the defendant (viz. the hypothesis that income inequality is detrimental to population health) is in imminent danger of being summarily executed without the benefit of a fair hearing. Indeed, some jurors already seem to have decided that a relationship between income inequality and health does not exist.
One recent assertion, for instance, was that ‘statistical adjustment for ethnicity statistically accounts for all of the association between income inequality and health within the US’.2 Other assertions, based on an ecological analysis,3 were that ‘adjustment for education ... also accounted for all of the association between income inequality and mortality‘2 and that the ‘evidence for the income inequality hypothesis is weak, beyond its important mechanical effects on individual income’,1 also based on ecological evidence.4,5 Examples of other claims include: ‘the evidence favoring a negative correlation between income inequality and life expectancy has disappeared’6 and that ‘we can muster little evidence to show that the extent of income inequality, per se, affects population health’.7
These are strong claims which, taken at face value, imply that income inequality is not a public health concern and the public health community has no cause to be alarmed about the sharp increase in income inequality that has occurred in the last two decades both within and between countries. However, we are not so confident that the income inequality story can be so hastily dismissed. In particular, several key accusations levelled by the prosecutors in this case can be tested with new evidence and better-designed studies. As witnesses for the defence, we would like to draw the attention of the jurors to evidence based on the more appropriate multilevel methods.
 |
On not adjusting for average state income |
|---|
 Top On not adjusting for... On the charge that...On not adjusting for...On testing the income...On not adjusting for...On testing the income...On the choice of...On national time trends...References |
|---|
Lynch and colleagues point out that our analysis8 failed to control for average state income, and therefore:
all we can conclude is that health effects of income inequality remain after adjustment for ‘per cent black’, but that this income inequality effect was unadjusted for mean income.1
Since it is a straightforward matter to add state-level income in our multilevel regression model, we now present a comparison of Model 4 estimates that appeared in our paper8 with and without adjustment for 1990 US Census median income for the states (Table 1
).
|
As can be seen, the odds ratios (OR) of reporting poor health was still statistically significant (1.23, 95% CI: 1.07, 1.41) for a 5% change in the state-level Gini. While the coefficient associated with state median income was statistically significant (OR = 0.97, 95% CI: 0.95, 0.99 for $1000 increase in state median income), the results for ‘per cent black’ remained statistically non-significant (OR = 1.03; 95% CI: 1.00, 1.06 for a 5% change in per cent black in a state).
|
On the charge that individual income explains the contextual effect of income inequality |
|---|
|
TopOn not adjusting for...On the charge that...On not adjusting for...On testing the income...On not adjusting for...On testing the income...On the choice of...On national time trends...References |
|---|
Along with Lynch and colleagues who posit that income inequality is not important ‘beyond its important mechanical effects on individual income’,1 others have argued that the apparent association between income inequality and health could be due to misspecification of individual income, and residual confounding.9 We present the results of alternative specifications of individual income in the relationship between state-level income inequality and poor health in the US (Table 2
).
|
The OR of reporting poor health increases by 1.37 for every 5% increase in the state Gini coefficient when a linear effect of income is assumed (Model 1). Considering income in terms of transformed log yields an OR of 1.37 (Model 2), while a non-linear specification in the form of a second order polynomial yields an OR of 1.36 (Model 3). When income is specified as deciles and as quintiles, the estimated OR is 1.34 (Model 4) and 1.35 (Model 5), respectively. Using categories of income (Model 6) yields an OR of 1.36. Additionally, adjusting Model 6 for state median income yields an OR of 1.32 (Model 7). Across the six different specifications of individual level income, therefore, the differences in OR for poor health associated with a 5% increase in the Gini were not substantial, suggesting that the relationship between state income inequality and individual health is independent of the income–health relationship at the individual level.
|
On not adjusting for educational attainment |
|---|
|
TopOn not adjusting for...On the charge that...On not adjusting for...On testing the income...On not adjusting for...On testing the income...On the choice of...On national time trends...References |
|---|
The results reported in our paper,8 as well as the evidence presented in this rejoinder (Tables 1
and 2), clearly demonstrate that individual educational attainment does not explain the observed multilevel relationship between income inequality and health. Yet, it is unfortunate that the prosecutors repeatedly cite, here1 and elsewhere,2 an ecological analysis, which found that adjusting for state level of educational attainment removes the apparent association between income inequality and aggregate mortality in the US.3 In this instance, we hope that the jurors would agree that evidence from a multilevel model, with adjustment for individual educational attainment, trumps ecological evidence. The prosecution also fail to take note of previous evidence10 published in response to the cited ecological analysis,3 in which individual educational attainment was added as a covariate in a multilevel model examining the association between US state income inequality and poor self-rated health. We reported that:
‘controlling for education attenuated but did not completely explain the relation between levels of state income inequality and self-rated health. Our results do not support the contention that education at the individual level fully confounds or mediates the association of income inequality with health.’10
We remain mystified by the apparent weight that prosecutors continue to give to ecological evidence when everyone in the courtroom agrees that this is severely problematic and that multilevel evidence is undoubtedly more reliable.
|
On testing the income inequality hypothesis within the US |
|---|
|
TopOn not adjusting for...On the charge that...On not adjusting for...On testing the income...On not adjusting for...On testing the income...On the choice of...On national time trends...References |
|---|
Reviewing the accumulated multilevel evidence on income inequality and health,11 we observe that the effect of state income inequality in the US is almost universal. Crucially, the statistically significant effect of state income inequality has been shown across a range of outcomes (self-rated health;8,11–14 hypertension, smoking, body mass index, and sedentarism;15 depressive symptoms;14 and mortality16). Meanwhile, the null evidence within the US has largely come from analyses that consider income inequality at smaller levels of geography (such as metropolitan areas or counties). We believe that there is compelling evidence implicating income inequality at the state level (as against other levels) on health outcomes.
|
On not adjusting for regional fixed effects in US data |
|---|
|
TopOn not adjusting for...On the charge that...On not adjusting for...On testing the income...On not adjusting for...On testing the income...On the choice of...On national time trends...References |
|---|
Lynch and colleagues refer in their commentary to our recent exchange with two economists, Mellor and Milyo.11,17 Using US Current Population Survey data, Mellor and Milyo concluded that there was no association between state income inequality and poor health after controlling for US regions as fixed effects.17 The purpose of our commentary11 was to show that, using the same dataset, but different approaches to modelling the data (US regions as random effects versus fixed effects), we came to a different conclusion, suggesting an adverse impact of income inequality on self-rated health. Whilst not conclusive either way, we were simply pointing out that Mellor and Milyo’s confident dismissal of income inequality as a public health concern was not robust to alternative model specifications.
|
On testing the income inequality hypothesis elsewhere |
|---|
|
TopOn not adjusting for...On the charge that...On not adjusting for...On testing the income...On not adjusting for...On testing the income...On the choice of...On national time trends...References |
|---|
We previously noted that the null studies of income inequality had been carried out in countries that were more egalitarian than the US—such as Sweden, Denmark, Japan, and New Zealand.11 Indeed, we also acknowledged in our paper,8 that income inequality in some settings may not (and perhaps need not) be associated with health outcomes given differences in political economy and value systems for tolerating particular levels of inequalities. Lest the court condemn the defendant on the basis of insufficient evidence, we made a plea for additional studies to be carried out in societies that are more unequal than the US, such as Chile.18 Lynch and colleagues apparently object to our plea for a stay in execution, noting that this amounts to a ‘shifting of the goalposts’ (or shall we say, the gallows). They imply that evidence from countries outside the OECD is inadmissible, since the defendant is being tried on the basis of posing as a theory that applies only to rich countries. We respectfully disagree.
The ability of income inequality to explain health variations between rich countries was just one of the original predictions of the theory. In ecological data, it may seem that income inequality does not explain variations in life expectancy between rich countries.19 However, such a conclusion may be erroneous since it fails to account for the within-country variations (both at the individual as well as at the area level within each country), not to mention that the cited evidence is again based on ecological cross-sectional data. Failure to corroborate one prediction should not result in a death penalty for the whole theory. Nor does it mean that income inequality does not exist in non-OECD countries, or that it is unimportant for the health of people in less economically developed nations.
|
On the choice of the outcome |
|---|
|
TopOn not adjusting for...On the charge that...On not adjusting for...On testing the income...On not adjusting for...On testing the income...On the choice of...On national time trends...References |
|---|
On the one hand, Lynch and colleagues state that both morbidity and mortality are equally important public health outcomes1 and on the other, they seem to assign different weights for the different outcomes. As it stands, they seem to favour the evidence that utilizes mortality as the outcome. Admittedly, self-rated health is not the same as mortality. The issue on the table, however, is not about the specific social epidemiology of mortality or self-rated health; both can be viewed as important markers that capture some underlying aspect of health. Instead, the debate is about the empirical validity of a particular ‘context’ (the context of income inequality) that drives population health, in general, and its accompanying variations. While the prosecutors’ call for future research in this area to be more attentive to specific outcomes is very welcome, as we noted before, the multilevel evidence of state income inequality in the US seems to be present for a range of outcomes, including mortality.11
|
On national time trends in income inequality and population health |
|---|
|
TopOn not adjusting for...On the charge that...On not adjusting for...On testing the income...On not adjusting for...On testing the income...On the choice of...On national time trends...References |
|---|
Potentially the most damning evidence against the income inequality story is the contradictory national trends over time in rising income inequality versus improving population health. Such univariate time trends have been previously presented for New Zealand,2 for the US,20 and now for South Korea.1 If health is getting better all the time, why should be fret about widening income inequality?
First, we must recognize that it is often very difficult to tease out causality from ecological time trend data, especially in the absence of knowledge about relevant induction times and lag periods. We have previously pointed out that a naïve analysis of smoking trends and lung cancer rates among US women risks the erroneous interpretation that quitting smoking (as US women have been doing since 1964) increases the risk of lung cancer.21 Such an interpretation is erroneous, because the rising rates of lung cancer in US women actually reflect their increased uptake of smoking 20–30 years earlier. If income inequality takes 10–15 years to affect population health,11,22 then it may still be too soon to detect an adverse impact of the surge in income inequality that occurred in the US in the 1980s and especially the 1990s. We concur with Lynch and colleagues that a life-course perspective would ideally take into consideration such lag effects, and as such, less reliance ought to be placed on crude visual inspection of time trend data to dismiss a theory.
Secondly, although much of the multilevel evidence on income inequality and health has been cross-sectional to date, this does not mean that the investigators were assuming an instantaneous effect of income inequality on health outcomes. The same US states that were unequal in 2000 were also unequal in the 1980s and 1990s. The state rankings of income inequality do not change a whole lot. Armed with this observation, it is still possible that cross-sectional data yields the correct answer, and as such capture the cumulative damage to health wrought by decades of living under conditions of inequality.
Lastly, ecological time trend data of the sort that Lynch and colleagues produce says little about what is happening to the health of sub-groups, such as the poor. Average life expectancy can improve for a nation, even as the health of disadvantaged groups stagnates or even deteriorates. Repeatedly adducing observations of questionable quality—whether from New Zealand, South Korea, or the US—does not amount to convincing epidemiological evidence to refute the income inequality theory. What would be helpful first steps are detailed examinations of the time trends in health of different (and potentially vulnerable) sub-groups. Indeed, no such evidence has been produced to date (and in the US would be very difficult to produce, given the lack of socioeconomic information on official vital records).
To summarize, we hope to have clarified our stand on the defence of the thesis that income inequality is a public health concern. No doubt our arguments will not be the last word on each of the issues we have raised. But further debate and discussion, based upon fresh multilevel empirical evidence (incorporating time as well as place dimensions), would be far preferable to hasty judgements formed on the basis of less than complete information and analysis.
|
References |
|---|
|
TopOn not adjusting for...On the charge that...On not adjusting for...On testing the income...On not adjusting for...On testing the income...On the choice of...On national time trends...References |
|---|
1 Lynch J, Harper S, Davey Smith G. Commentary: Plugging leaks and repelling boarders—where to next for the SS Income Inequality? Int J Epidemiol 2003;32:1029–36.
2 Pearce N, Davey Smith G. Is social capital the key to inequalities in health? Am J Public Health 2003;93:122–29.
3 Muller A. Education, income inequality, and mortality: a multiple regression analysis. BMJ 2002;324:23–25.
4 Rodgers GB. Income and inequality as determinants of mortality: an international cross-section analysis. Popul Stud 1979;33: 343–51.[CrossRef]
5 Gravelle H. How much of the relationship between population mortality and unequal distribution of income is a statistical artefact? BMJ 1998;316:382–85.
6 Mackenbach JP. Income inequality and population health. BMJ 2002; 324:1–2.
7 Lynch J, Davey Smith G. Commentary: income inequality and health: the end of the story? Int J Epidemiol 2002;31:549–51.
8 Subramanian SV, Kawachi I. The association between state income inequality and worse health is not confounded by race. Int J Epidemiol 2003;32:1022–28.
9 Milyo J. Income distribution, socioeconomic status, and self-rated health in US—Letter Author ignored data in their study. BMJ 1999; 318:1417.
10 Blakely T, Kawachi I. Education does not explain association between income inequality and health. BMJ 2002;324:1336.
11 Subramanian SV, Blakely T, Kawachi I. Income inequality as a public health concern: where do we stand? Commentary on Mellor J, Milyo J. Is exposure to income inequality a public health concern? Health Serv Res 2003;38:153–67.[CrossRef][ISI][Medline]
12 Kennedy BP, Kawachi I, Glass R, Prothrow-Stith D. Income distribution, socioeconomic status and self-rated health: a US multilevel analysis. BMJ 1998;317:917–21.
13 Blakely T, Kennedy BP, Kawachi I. Socioeconomic inequality in voting participation and self rated health. Am J Public Health 2001;91:99–104.[Abstract]
14 Kahn RS, Wise PH, Kennedy BP, Kawachi I. State income inequality, household income, and maternal mental and physical health: cross sectional national survey. BMJ 2000;321:1311–15.
15 Diez-Roux AV, Link BG, Northridge ME. A multilevel analysis of income inequality and cardiovascular disease risk factors. Soc Sci Med 2000;50:673–87.[CrossRef][ISI][Medline]
16 Lochner K, Pamuk ER, Makuc D, Kennedy BP, Kawachi I. State-level income inequality and individual mortality risk: a prospective multilevel study. Am J Public Health 2001;91:385–91.
17 Mellor J, Milyo J. Is exposure to income inequality a public health concern? Lagged effects of income inequality on individual and population health. Health Serv Res 2003;38:137–51.[CrossRef][ISI][Medline]
18 Subramanian SV, Degaldo I, Jadue L, Vega J, Kawachi I. Income inequality and health: multilevel analysis of Chilean communities. J Epidemiol Community Health 2003(In press).
19 Lynch J, Davey Smith G, Hillemeier M, Shaw M, Raghunathan T, Kaplan G. Income inequality, the pyschosocial environment, and health: comparisons of wealthy nations. Lancet 2001;358:194–200.[CrossRef][ISI][Medline]
20 Lynch J, Davey Smith G. Rates and states: reflections in the health of nations. Int J Epidemiol 2003;32:663–70.
21 Kawachi I, Blakely T. When epidemiologists and economists disagree. J Health Politics 2001;26:533–41.
22 Blakely TA, Kennedy BP, Glass R, Kawachi I. What is the lag time between income inequality and health status? J Epidemiol Community Health 2000;54:318–19.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  S. Subramanian and I. Kawachi Commentary: Chasing the elusive null--the story of income inequality and health Int. J. Epidemiol., June 18, 2007; (2007) dym102v2. [Full Text] [PDF]
|
|
|
|
 |
|
 F. J Zimmerman and J. F Bell Income inequality and physical and mental health: testing associations consistent with proposed causal pathways. J. Epidemiol. Community Health, June 1, 2006; 60(6): 513 - 521. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. V. Subramanian and I. Kawachi Income Inequality and Health: What Have We Learned So Far? Epidemiol. Rev., July 1, 2004; 26(1): 78 - 91. [Full Text] [PDF]
|
|
|
|
|
|
A. V. Diez Roux The Study of Group-Level Factors in Epidemiology: Rethinking Variables, Study Designs, and Analytical Approaches Epidemiol. Rev., July 1, 2004; 26(1): 104 - 111. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||