IJE Advance Access first published online on June 8, 2007
This version published online on June 18, 2007
International Journal of Epidemiology, doi:10.1093/ije/dym102
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Commentary: Chasing the elusive null—the story of income inequality and health
School of Public Health, Harvard University, 677 Huntington Avenue, KRESGE 7th Floor, Boston, MA 02115-6096 USA.
* Corresponding author. Department of Society, Human Development and Health, Harvard School of Public Health, 677 Huntington Avenue, KRESGE 7th Floor, Boston, MA 02115-6096, USA. E-mail: svsubram{at}hsph.harvard.edu
Accepted 17 April 2007
 |
Background |
|---|
 Top Background DiscussionConclusionAcknowledgementsReferences |
|---|
Backlund and colleagues,1 provide new multilevel evidence of a strong and robust association between US state income inequality and individual mortality in the <65-year-old adult population (relative risks of 1.39 for men and 1.13 for women, Table 2,1), even after conditioning this association on a range of covariates. Yet, this statistically and substantively significant finding is not a part of the study's conclusion. Instead, the null association observed in the elderly population (
65 years) is emphasized to conclude that ‘this explains why income inequality is not a major driver of mortality trends in the United States because most deaths occur at ages 65 and over.’ In this comment, we evaluate the substantive and empirical aspects of the study, which we believe helps to settle some disagreements in the field. The study, however, is also characteristic, somewhat unfortunately, of the way in which some of the debate on income inequality and health has been portrayed. Specifically, the conclusions have been at variance with the very empirical evidence presented by the researchers who are sceptical of this association. |
Discussion |
|---|
|
TopBackgroundDiscussionConclusionAcknowledgementsReferences |
|---|
The source of ‘mixed’ findings
The fundamental premise of the study by Backlund and colleagues is to ‘clarify’ why ‘2 out of 5 studies’ found a positive association between income inequality and mortality. Before commenting on what Backlund and colleagues present as points of clarification, a brief description of these five studies is necessary, since, as they say, ‘the devil is in the details’.
The five studies are highly heterogeneous and not always comparable with each other. To start with, the ecological study by Deaton and Lubotsky2 is at best hypothesis-generating, rather than an empirical test of whether there is an association between a contextual effect of income inequality and individual mortality. The multilevel nature of the income inequality hypothesis inhibits the usefulness of this ecological study as evidence either in support or refuting the hypothesis.3 Similarly, the study by Wolfson and colleagues,4 as important as it was in providing a basis for the claim that the ecological association between state income inequality and mortality is not an ‘artefact’,5 was nonetheless based on a simulation exercise and did not directly test the contextual effect of income distribution on individual health.
Of the remaining three studies, the two null studies cited by Backlund and colleagues have the following characteristics. The study by Fiscella and Franks,6 was based on a sample of only 14 407 adults (with number of deaths not reported), where income inequality was measured by using income from the sampled data, at the level of ‘communities’, with community being the primary sampling unit in the National Health and Nutrition Examination Survey. The study does not report what these primary sampling units represent. We and others have pointed to the importance of level of aggregation for studying the impacts of income inequality,3,7 with the key point being at smaller levels of spatial aggregation, the range of income inequality is likely to be constricted because of the presence of economic residential segregation. The other null study by Daly and colleagues,8 meanwhile, was based on an even smaller sample of 6500 adults with just 341 deaths in the first period and 375 deaths in the second period, with income inequality measured at the state-level (though here too the authors did not report on the number of US states included in their sample).
Finally, we have the prospective multilevel study by Lochner and colleagues,9 which had a sample of 546 888 persons in 48 US states, with 19 379 deaths over a 8-year follow-up period and which included a variety of individual covariates. This study found a 12% increased mortality risk for adult individuals residing in high-income inequality states as compared with low-income inequality states. To date, along with the study by Backlund and colleagues, the study by Lochner and colleagues remains, statistically speaking, the most powerful test of the association between state income inequality and mortality.
We draw readers’ attention to the details of these studies in order to exemplify how even without the benefit of the new study by Backlund and colleagues, there is slim justification for characterizing the empirical evidence between state income inequality and mortality in the US as ‘mixed’. Of course, the evidence can be considered ‘mixed’ if researchers wish to place an equal or greater weight on aggregate,2 and under-powered studies,6,8 whilst discounting the largest prospective multilevel study.9
Repeating the mantra of ‘mixed’ evidence doesn’t make it so
The study by Backlund and colleagues would appear to provide a strong counter ballast to the claim that the association between US state income inequality and mortality is ‘mixed’/‘debatable’/‘not present’. However, we are perplexed by why Backlund and colleagues decided not to report anywhere in their study the main effect of state income inequality on individual mortality risk. This by itself, based on the rationale presented in the article, would have been sufficient to ‘clarify’ the apparent ‘mixed’ findings in the existing literature. Instead, the authors only present findings stratified.
Based on crudely averaging the results that Backlund and colleagues present for the two age strata (<65, and 65), the relative mortality risk associated with increased state income inequality should be 
1.20 for men, and 1.06 for women. At the population level, we should expect to find that conditional on individual compositional variables, there is ‘at least’ 10% increase in mortality risk (averaged across sexes) due to residing in high-income inequality states; an estimate that is in the same ballpark as what was reported by Lochner and colleagues,9 even though the study by Backlund and colleagues have the advantage of including a wider age range that includes non-elderly as well as elderly adults.
The new study, while failing to report results based on the overall pooled analysis, does ask a potentially important question, i.e. could the positive association between state income inequality and individual mortality be different for different population groups? Indeed, there is no a priori justification to restrict this second-order question to age, and sex; one can extend this to other socio-economic and demographic groupings such as individual income, education, race and so on. Such investigation has been extensively conducted before in the context of self-rated health, with little support for the hypothesis that state income inequality affects different population groups differently.10 The evidence presented in the context of mortality by Backlund and colleagues is, therefore, interesting in the sense that for the elderly population the deleterious effects of income inequality do not appear to be present. Further research may be necessary to explain the apparently ‘null’ finding in the elderly age-group.
On the lack of state income inequality effect on elderly mortality
The substantive conclusion of the study by Backlund and colleagues is that the null association between state income inequality and mortality in the 65 age-strata ‘explains why income inequality is not a major driver of mortality trends in the United States because most deaths occur at ages 65 and over’. Before accepting this conclusion, it is worth pointing out that the association of any risk factor with mortality is substantially weaker at older ages than in middle age. This is because of the well-known problem of high background mortality rates which impose ceiling effects on the estimation of relative mortality as populations age. Put simply, we eventually have to die as we grow older. For example, in the Cardiovascular Health Study (a cohort of individuals 65 years and older), the relative risk of mortality even after 26–50 pack-years of smoking was 1.13 (95% CI 0.90–1.43), compared with never smoking—an association that is considerably attenuated and statistically insignificant at conventional levels of precision in comparison to what is typically reported in middle-aged populations.11 Indeed, mortality differentials, in general, even by socioeconomic status or race tend to be notably attenuated at older ages. Studies even suggest a "cross-over" such that elderly minority racial groups (e.g., elderly blacks) tend to have a lower risk of mortality compared to the elderly white majority.12 Taken at face value, the conclusion by Backlund and colleagues implies a level of nihilism whereby we should only focus on the drivers of mortality after age 65, and ignore risk factors, including the socio-economic and race/ethnic ones, for premature adult mortality, since that is when most people die.
On the importance of the study by Backlund and colleagues
The study by Backlund and colleagues does help settle a few lingering and emerging issues that have pre-occupied sceptics. The first relates to whether the income inequality and mortality association is confounded by compositional effects of race/ethnicity or more specifically, ‘fraction black’ at the state level,2 as well as uncontrolled regional effects.13 While neither has been shown to matter in studies of self-rated health,14–16 Backlund and colleagues have now shown that the association between state income inequality and individual mortality risk is robust to these adjustments.
Second, by examining state income inequality measured in 1980 and 1990, the study by Backlund and colleagues counters scepticism related to whether the state income inequality and health association in 1990 was a fluke. Specifically, it has been claimed that the strong association observed using 1990 estimates of state income inequality is not present when state income inequality from other periods, i.e. before-1990 or after-1990, is considered.17 Backlund and colleagues used both 1980 and 1990 measures of state income inequality, and report relative risks of 1.33 and 1.39 for men, and 1.14 and 1.13 for women for the two time periods, respectively; a finding that refutes the claim that the association reported in 1990 was an exception. Furthermore, the claim that the strong association is not present after 1990 is also untrue.18 Interestingly, a closer inspection of the data from the very study that made this claim, (Table 117) reveals that the aggregate correlation between state income inequality and all-cause mortality rate was 0.58 in 1990 and 0.44 in 2000, with the difference between the two correlation coefficients being statistically non-significant (P = 0.356). The corresponding correlation between state income inequality and self-rated health for the two time periods were identical.(Table 117) Yet, a conclusion was made that was at odds with the evidence presented—an instance of trolling for the null?17
Finally, questions have been raised regarding the sensitivity of the income inequality-health association to the source of income data that is used to measure income inequality.17 We commend the study by Backlund and colleagues on their use of census-based data to construct their measure of income inequality for the following reason. Most US studies rely on census data, and census-based income inequality measures fail to capture the top end of the income distribution because of top-coding of income in the census forms. When income-tax return data were used, thereby capturing the top end of the income distribution, the association between income inequality and health apparently weakens in aggregate models.17 Before accepting this conclusion, it is worth noting that the problem with the income tax return data is that they miss the bottom end of the distribution, i.e. those who do not file tax returns owing to poverty and very low incomes. Interestingly, this fundamental problem is borne out by the very data presented to critique the use of census data (Figure 2,17 reproduced with permission for convenience as Figure 1). The state-variation in census-based Gini coefficient varies roughly between 0.38 and 0.48 (with 0 representing complete equality and 1 representing complete inequality); while the corresponding range based on the income tax return data is 0.45 and 0.55. Thus, income tax return data allows us to capture the top end of the income distribution, but only at the expense of ignoring the bottom end. Instead of privileging one source of data over another as was done by researchers (in this instance, favouring the tax return data, which happened to yield the weaker result?),17 a more appropriate inference should have been to perhaps recommend the need to derive a weighted estimate of state income inequality, one that draws on the advantages of both data sources and thereby potentially capturing the entire spectrum of the income distribution.
|
|
Conclusion |
|---|
|
TopBackgroundDiscussionConclusionAcknowledgementsReferences |
|---|
In summary, the study by Backlund and colleagues, with its clear methodological and empirical strengths, provides important confirmation of the association between state income inequality and mortality in the United States. However, the thrust of the conclusion (emphasizing the null association found for the older subgroup), as we have argued, does not accurately reflect the contribution of the study to the ongoing ‘debate’ about income inequality.
|
Acknowledgements |
|---|
|
TopBackgroundDiscussionConclusionAcknowledgementsReferences |
|---|
S.V.S. is supported by the National Institutes of Health Career Development Award (NHLBI 1 K25 HL081275).
|
Notes |
|---|
The orginally published version of this paper was incorrect. On Page 2, Column 2, Line 11: the reference number (12) should be AFTER the period or fullstop, Not before.
Invited Commentary on Backlund E, Rowe G, Lynch J, Wolfson MC, Kaplan GA, Sorlie PD. Income inequality and mortality: a multilevel prospective study of 521 248 individuals in 50 U.S. states. Int J Epidemiol 2007. 
|
References |
|---|
|
TopBackgroundDiscussionConclusionAcknowledgementsReferences |
|---|
1 Backlund E, Rowe G, Lynch J, Wolfson MC, Kaplan GA, Sorlie PD. Income inequality and mortality: a multilevel prospective study of 521 248 individuals in 50 US states. Int J Epidemiol (2007 Mar 15); doi:10.1093/ije/dym012.
2 Deaton A, Lubotsky D. Mortality, inequality and race in American cities and states. Soc Sci Med (2003) 56:1139–53.[CrossRef][ISI][Medline]
3 Subramanian SV, Kawachi I. Income inequality and health: what have we learned so far? Epidemiol Rev (2004) 26:78–91.
4 Wolfson M, Kaplan G, Lynch J, Ross N, Backlund E. Relation between income inequality and mortality: empirical demonstration. Br Med J (1999) 319:953–55.
5 Gravelle H. How much of the relationship between population mortality and unequal distribution of income is a statistical artefact? Br Med J (1998) 316:382–85.
6 Fiscella K, Franks P. Poverty or income inequality as predictors of mortality: longitudinal cohort study. Br Med J (1997) 314:1724–28.
7 Wilkinson RG, Pickett KE. Income inequality and population health: a review and explanation of the evidence. Soc Sci Med (2006) 62:1768–84.[CrossRef][ISI][Medline]
8 Daly M, Duncan GJ, Kaplan GA, Lynch JW. Macro-to-micro links in the relation between income inequality and mortality. Milbank Mem Fund Quart (1998) 76:315–39.
9 Lochner K, Pamuk E, 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.
10 Subramanian SV, Kawachi I. Whose health is affected by income inequality? A multilevel interaction analysis of contemporaneous and lagged effects of state income inequality on individual self-rated health in the United States. Health &Place (2006) 12:141–56.[CrossRef][ISI][Medline]
11 Fried LP, Kronmal RA, Newman AB, et al. Risk factors for 5-year mortality in older adults: the Cardiovascular Health Study. JAMA (1998) 279:585–92.
12 Johnson Nan E. The Racial Crossover in Comorbidity, Disability, and Mortality. Demography (2000) 37:267–83.[ISI][Medline]
13 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]
14 Subramanian SV, Blakely T, Kawachi I. Income inequality as a public health concern: where do we stand? Commentary on "Is exposure to income inequality a public health concern?" Health Serv Res (2003) 38:153.[CrossRef][ISI][Medline]
15 Subramanian SV, Kawachi I. Response: In defence of the income inequality hypothesis. Int J Epidemiol (2003) 32:1037–40.
16 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.
17 Lynch J, Harper S, Kaplan GA, Davey Smith G. Associations between income inequality and mortality among US states: the importance of time period and source of income data. Am J Public Health (2005) 95:1424–30.
18 Ram R. Income inequality, poverty, and population health: evidence from recent data for the United States. Soc Sci Med (2005) 61:2568–76.[CrossRef][ISI][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||