Article Text

Download PDFPDF

Income inequality and health: the role of population size, inequality threshold, period effects and lag effects
  1. Naoki Kondo1,
  2. Rob M van Dam2,
  3. Grace Sembajwe3,
  4. S V Subramanian4,
  5. Ichiro Kawachi3,
  6. Zentaro Yamagata1
  1. 1Department of Health Sciences, Interdisciplinary Graduate School of Medicine and Engineering, University of Yamanashi, Chuo-shi, Yamanashi, Japan
  2. 2Department of Epidemiology and Public Health and Medicine, Yong Loo Lin School of Medicine, National University of Singapore, Singapore
  3. 3Department of Environmental Health, Harvard School of Public Health, Boston, Massachusetts, USA
  4. 4Department of Society, Human Development, and Health, Harvard School of Public Health, Boston, Massachusetts, USA
  1. Correspondence to Naoki Kondo, Department of Health Sciences, Interdisciplinary Graduate School of Medicine and Engineering, University of Yamanashi, 1559-1 Shimokato, Chuo-shi, Yamanashi 409-3898, Japan; nkondo{at}yamanashi.ac.jp

Abstract

Background Income inequality has been associated with worse health outcomes in several but not all studies. The heterogeneity across studies may be explained by the variations in the size of area or population over which income inequality was evaluated. Moreover, the studies above a certain inequality threshold, conducted more recently, and incorporating a time lag may have stronger associations between income inequality and health. The authors investigated if the strength of the association between income inequality and health was altered by these factors.

Methods The authors conducted a multivariate meta-regression analysis using nine multilevel cohort studies on income inequality and mortality and 14 multilevel cross-sectional studies on income inequality and self-rated health.

Results Among cross-sectional studies, studies evaluating country-level inequality (average population>24 million) were more likely to show a stronger association between income inequality and poor health compared with those evaluating income inequality within small average populations (<820 000). There were no significant differences in the effect size of inequality–health association relating to the differences in the population size within a country across which income inequality was evaluated in both cross-sectional and cohort studies. The authors found that the threshold effects, period effects and lag effects were independent of the population size.

Conclusions Income inequality at the country level may have stronger adverse contextual effects on health than inequality in smaller areas, perhaps by best reflecting social stratification in a society. Furthermore, we found that threshold, period and lag effects were independent of area unit for evaluating inequality, which may have important policy implications.

  • Socioeconomic factors
  • meta-analysis
  • income inequality
  • self-rated health
  • mortality
  • deprivation
  • elderly
  • social inequalities
  • multilevel modelling
  • social epidemiology

Statistics from Altmetric.com

Introduction

More than three-quarters of the countries belonging to the Organization for Economic Co-operation and Development have experienced a growing gap between rich and poor during the past two decades. In a highly unequal society, a substantial segment of the population is impoverished and unhealthy. Income inequality may also affect the health of not just the poor but also the better off in society by increasing psychosocial stress resulting from invidious social comparisons and the erosion of social cohesion.1–3 Using a meta-analysis approach, we previously identified a statistically significant association between higher income inequality and health in terms of premature mortality and poor self-rated health (SRH).4 We also demonstrated that the income inequality–health association may be stronger among societies or areas in which average income inequality was higher (ie, threshold effects) in recent years, and in particular after around 1990 (ie, period effects) and among studies incorporating time lag between income inequality and health (ie, lag effects).

A potential criticism to these ‘effects’ we suggest is that they may be statistical artefacts or just the markers of the variations across studies in terms of the size of area or population over which income inequality measures were evaluated since inequality in small areas/populations may not be able to reflect the scale of social class differences or social stratifications in a society.5 To evaluate if the population size for evaluating income inequality alters or nullifies the inequality–health association and whether or not threshold, period and lag effects were independent of the selection of the area unit, we reanalysed our previous data set with additional information on the population size for evaluating inequality in each study.

Methods

Data

For the present meta-regression analysis, we used the same data set as previously used in our meta-analysis.4 A complete account of our publication search, paper inclusion criteria, data abstraction, methods for the standardisation of estimates and a list of papers meta-analysed was published elsewhere.4 In brief, a researcher trained in online article searches (NK) searched papers using PubMed, ISI Web of Science and the National Bureau of Economic Research database. We have also reviewed those papers cited in systematic reviews published on this topic.5–8 To be included, studies had to use multilevel data (at least two levels including one or more regional variable(s)); address sample clustering caused by multilevel data structure; adjusted for age, sex and individual socioeconomic status; and be peer reviewed. Only multilevel data can properly distinguish the contextual health effects of income inequality from the effect of individual socioeconomic status.6 9 We followed published guidelines for meta-analyses of observational studies.10 Two investigators (NK and GS) independently extracted information on study design, data sources, country of data origin, sample size, number of cases, age, sex, estimations, response rate, follow-up rate, follow-up duration, measure of income inequality, outcome, outcome specifications (binary or ordinal/number of SRH items), area unit over which income inequality was evaluated, adjustment variables, statistical modelling strategies and methods for addressing data clustering. We resolved discrepancies between the data abstracted by the two investigators. As the result of our publication search, we selected 19 cross-sectional studies with outcomes of poor SRH and nine cohort studies with an endpoint of mortality. For the present analysis, we used all nine cohort studies11–20 and 14 cross-sectional studies,21–35 excluding those using ordinal regression models. If necessary, we contacted authors to obtain missing information. Table 1 shows the basic information on the studies that were meta-analysed.

Table 1

Characteristics of studies used for meta-regression

Measurements

In the present study, we evaluated average population size of the administrative units over which income inequality was measured. We obtained the information on the population size from a research article, if the authors provided the information, otherwise we downloaded the population data of the administrative units (eg, US states, Japanese prefectures) from official online websites and databases published by national and international governments. Then, we calculated the average population across the administrative units. If we did not find such information of the administrative units, we divided the total population of the study area by the total number of the administrative units in the area. Using Ichida et al's23 data, which we have access to, we confirmed that values of average population size calculated using multiple alternative methods remain very similar.

We evaluated income inequality by the Gini coefficient (GINI), the most commonly used measure of income inequality. GINI is formally defined as half of the arithmetic average of the absolute differences between all pairs of incomes within the sample, with the total then being normalised on mean income. If incomes are distributed completely equally, the value of the GINI will be zero. If one person has all the income (complete inequality), the GINI will assume a value of 1. Some studies used other measures of income inequality; because alternative measures are all highly correlated (Pearson's r>0.94), according to Kawachi and Kennedy36, we transformed all measures to GINI. In our meta-regression, explanatory variables included average population size over which inequality was evaluated: median GINI (as a continuous variable) or the threshold of income inequality (whether median GINI in each study was 0.3 or higher vs lower than 0.3—the value previously suggested as a potential threshold over which the impact of income inequality on health may become significantly higher), the study period (whether baseline data were taken later than 1990 vs earlier), whether the study incorporated time lag between income inequality and health (yes vs no, only in cross-sectional studies), the follow-up duration (7 years or longer vs <7 years, only in cohort studies) and adjustment for area income (yes vs no). These were the factors previously suggested that they significantly altered the effect size on inequality–health association.4

Statistical analysis

We conducted a meta-regression analysis with random effects models. First, the bivariate meta-regression model evaluated the factors potentially explaining study heterogeneity (ie, average population size, median GINI, time lag, adjustment for area income, following up duration and baseline data year) on the overall associations between 0.05 unit increase in GINI and mortality (cohort studies) and poor SRH (dichotomised, in cross-sectional studies). Then, we created multivariate meta-regression models; each of them included two explanatory variables, that is, average size of the population in which income inequality was evaluated and one each from the remaining variables. Because the effects of average population size could be non-linear, we created dummy variables representing three different sizes, in addition to the size variable with a continuous specification. In cross-sectional studies, because between-country studies (ie, studies evaluating income inequality at country level) had a far larger average population size (see table 1), we first separated out between-country studies from within-country studies and then divided the remaining into two groups at the median of the population size. I2 statistics and the Cochran Q test were used to evaluate heterogeneity.37 38 All p values were two tailed. All analyses were done using STATA V11 (StataCorp LP).

Results

Bivariate meta-regression using continuous variable of average population size showed that cross-sectional studies evaluating income inequality in larger population were likely to show stronger association between income inequality and poor SRH (regression coefficient (β)=0.002, p<0.001, per 500 000 unit population increase). Likewise, between-country studies (average population >23 700 000) were more likely to have strong association between income inequality and poor SRH compared with those evaluating inequality in a small population (average population was between 8 and 820 thousand) (β=0.088, p<0.001) (table 2). Among within-country studies, average population size did not significantly explain the study variations in effect size. On the other hand, as we previously reported, higher average inequality (GINI of 0.3 or higher) (β=0.038, p=0.01) and incorporation of time lag between inequality and poor SRH (β=0.019, p<0.001) were statistically significant factors associated with stronger associations between inequality and poor SRH, whereas studies adjusting for area income tended to show weaker associations between inequality and poor health (β=−0.021, p=0.02). These findings did not change even in the multivariate meta-regression models simultaneously considering average population size, except for the model for area income adjustment in which the regression coefficient was attenuated to −0.016 (p=0.12).

Table 2

Meta-regression models explaining study heterogeneity on the association between income inequality and poor SRH

These results were mostly consistent with the results of our meta-regression for cohort studies of income inequality and mortality (table 3). In contrast to the cross-sectional studies, there were no between-country cohort studies. Differences according to the average population sizes of areas in which income inequality was evaluated within a country did not explain variation in the effect sizes across studies (β=−0.033, p=0.19). The effects of threshold, period and area income adjustment remained statistically significant even adjusting for the average population size. That is, higher average income inequality (p=0.002), using more recent data (ie, later than 1990) (p=0.04), and adjusting for area income (p=0.03) were associated with stronger associations between income inequality and premature mortality.

Table 3

Meta-regression models explaining study heterogeneity on the association between income inequality and mortality

Discussion

Our meta-regression analysis found that between-country studies were likely to show stronger association between income inequality and health compared with within-country studies. Although the adjustment for average population size for evaluating income inequality indeed attenuated the effects of previously suggested factors explaining the heterogeneity in the effect sizes of studies on income inequality hypothesis, the threshold, period and lag effects remained statistically significant.

Does area size matter?

On the one hand, Wilkinson and Pickett5 have argued that income inequality at larger scales (eg, the whole society) is more strongly associated with adverse health outcomes than inequality at smaller scales (eg, neighbourhoods) because the former captures the full extent of social stratification in society (ie, there is less variation in inequality at smaller scales, especially if the neighbourhoods are highly segregated by socioeconomic status). Wilkinson and Pickett's interpretation is also consistent with some literature on subjective social status and health. When researchers have compared two versions of the subjective social status ‘ladder’ (asking individuals to place themselves on an imaginary ladder of social hierarchy), studies have indicated consistently stronger associations between lower position on the ladder and poor health outcomes when the rater is invited to imagine the ladder as representing the whole society as opposed to their local community.39 One interpretation of this result is that invidious social comparisons are more ‘toxic’ when they are made with the rest of society (eg, a rural person comparing themselves with a Wall Street banker) than with their more proximate neighbours. The present study in part supports this; although the effect size of income inequality did not vary among within-country studies, it was significantly larger in the studies evaluating inequality in the largest average population size: country. National social hierarchies may matter to people more than local social class structures.

On the other hand, caution is warranted in interpreting our findings because in general, the larger the population unit studied, the greater the degree of unobserved heterogeneity. For example, cross-national studies of income inequality involve many unobserved sources of heterogeneity (eg, cultural factors, historical contexts, geography), and there is a strong likelihood that these confound the association between income inequality and health outcomes. A partial solution to this problem is to use fixed effects analysis to control for time-invariant unobserved heterogeneity. When cross-national studies have employed fixed effects, the association between country-level inequality and health has been null.40

Threshold, period and lag effects

Subramanian and Kawachi6 have introduced the threshold hypothesis in their findings, which show that the statistically significant association between inequality and health is more likely to be observed in countries with higher income inequality—typically in the USA and other nations having a high inequality such as Chile—and less likely to be found among less unequal countries. Lynch and colleagues7 have alternatively extrapolated similar findings showing that the USA is an exception, having some unique characteristics including the distribution of health-relevant resources and risk factors such as diet and smoking. However, as we have previously revealed, some recent large-scale cohort studies, mostly from Nordic countries, also showed a positive association between income inequality and mortality.4 On the other hand, our study revealed that the threshold effects were independent of the area size. Taken together, we conclude that the adverse impacts of income inequality on health may exist even in less unequal societies but the impacts become stronger when income inequality exceeds a certain threshold value, and this may be applicable to any society or community regardless of its size.

Nevertheless, although in this study we used the median GINI (=0.3) among the cohort studies included in our meta-analysis as a potential threshold, our sensitivity analysis using an alternative cut-off value (GINI=0.37) in the meta-regression of cross-sectional studies did not show a stronger health impact of income inequality among studies with higher GINIs. This may be because the value (0.37) was a wrong ‘threshold’, not capturing the correct point where the effect size changed. Due to the limited sample size, however, we should also consider other possibilities that the threshold effects identified is by chance or that the suggested threshold value (0.3) is too rough of an estimate. Further studies are needed to identify a more precise threshold value that can be generally applicable.

Period effects may be another factor in explaining study heterogeneity independent of the area size. Given the threshold effects hypothesis, this might be the reflection of the effect due to increasing income inequality in recent years, particularly after the 1990s. Alternatively, unobserved social changes other than increasing income inequality may explain the period effects. For example, adverse health impacts by the erosion of social capital23 41 and degraded work environment may have happened in recent years along with the globalisation in economies and the labour market. There is empirical evidence that a series of regional and global economic crises after the late 1990s may have expanded health disparities across social classes and deteriorated health among some specific subpopulations.42–45

In addition, among cross-sectional studies, the lag time between income inequality and SRH best explained study heterogeneity (with the least residual heterogeneity), supporting the existence of temporal relationships between income inequality and individual health. Income inequality may affect population health, by increasing psychosocial stress resulting from invidious social comparisons and the erosion of social cohesion,1–3 and these pathways may require a certain time period. This finding coincides with a study by Blakely and colleagues that scanned lag time between inequality and poor SRH. They found possible lag effects of up to 15 years among people aged 45 year or older, although analytic approaches on its evaluation have been under discussion.46–49

Adjustment for area income

In our cross-sectional meta-regression, the effect of the adjustment of area income was nullified when we adjusted for population size, whereas this was not shown in our meta-regression of cohort studies. Although this could give useful insights into the study of income inequality hypothesis, further discussion is difficult at this moment, given the multiple potential interpretations of the models adjusting for area income. For example, area income can represent area-level material affluence and area relative income compared with overall absolute average income.50

Study limitations

Our study has several limitations. First and foremost, all meta-analyses of observational studies are prone to biases in the original studies.51 For example, although we evaluated multiple models using alternative sets of covariates, the estimates from the original studies might have been prone to residual confounding. Second, the small number of studies included in the present analysis limited the ability to conduct further multivariate adjustments in meta-regression analysis. Third, five cross-sectional analyses did not report the necessary information to permit us to include them in the meta-analysis.47 52–55 Their omission might have influenced our conclusions. However, the findings of our meta-regression for cross-sectional studies were consistent with that for the cohort studies that involved larger samples and, as we previously revealed, had no evidence of a publication bias.4 Finally, the results of our study cannot be generally applicable to some parts of the world, in particular among developing countries in Asia and Africa, where both population health statuses before and after the epidemiologic transition coexist. For example, Subramanian et al56 demonstrated that income inequality across Indian states was associated with the double burden of under- and overnutrition. Large developing nations like India and China are very interesting study fields for this topic because they have huge population sizes even at the state or province levels, which can be as large as an ordinary country as a whole. With data from these countries, the modifying effects of area size on the income inequality–health association can be formally evaluated.

Conclusions

Income inequality in a large population may have stronger adverse effects on health. It may be particularly applicable to a very large population, typically, in the whole country. These findings suggest the existence of differential pathways linking unequal social contexts to health, when we evaluate them in different area units. Further studies investigating those pathways should be valuable for a better understanding of the association between income distribution and health. In addition, our demonstration of the threshold effects that was independent of the population size for inequality evaluation would also have an important policy implication. If income inequality actually has a threshold over which its adverse impacts become remarkable or unignorable, such a threshold value could be used as a numeric goal for public health and economic policies. If the threshold effects could truly be applicable regardless of area size for evaluating inequality, it may be possible to find a threshold value that can be universally applicable to any level of society/communities in the global, national and local settings. To this end, further investigation is awaited because of the lack of empirical evidence from many parts of the world, and in particular, from developing countries.

What is already known on this subject

  • The effects of income inequality on individual health can largely vary across different sizes of area or population over which income inequality is measured.

  • Studies having higher average income inequality, using more recent data, and incorporating time lag between income inequality and health outcomes are likely to show a more consistent association between income inequality and poor health. However, these findings may be strongly affected by the size of the area or population unit across which income inequality is evaluated.

What this study adds

  • Income inequality measured at the scale of larger population units (eg, countries) appears to have a stronger association with health outcomes than inequality measured in smaller population scales.

  • The threshold, period and lag effects may exist independent of the population unit. With further studies developing more generally applicable and precise evidence on the value of income inequality threshold, the evidence can be used as a public health goal for any level of government.

Acknowledgments

We thank Jimpei Misawa, Shintaro Fukushima and Yukinobu Ichida for their support on data acquisition.

References

Footnotes

  • Funding This work was supported by Grant-in-Aid for Scientific Research on Innovative Areas (No. 22119504) provided by the Ministry of Ministry of Education, Culture, Sports, Science and Technology, Japan. SVS is supported by the National Institute of Health Career Development Award (NHLBI 1k25 HL081275). These sponsors were not involved in the study design; the collection, analysis and interpretation of data; the writing of the article or the decision to submit it for publication.

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.