Income inequality and population health: Correlation and causality☆
Introduction
The past 15 years have witnessed an explosion of research on the relationship between ecological measures of income inequality and aggregate measures of population health. Initial studies suggested a strong correlation between inequality and life expectancy (Wilkinson, 1992) and infant mortality (Pampel & Pillai, 1986) among developed countries. Follow-up studies disputed these findings (Judge, 1995, Judge et al., 1998). Later studies of the relationship between inequality and health broadened the panel of countries to include developing countries (Ellison, 2002, Hales et al., 1999) and other indicators of health, such as murder rates (Lee & Bankston, 1999). Recent analytical reviews of the literature on inequality and health have ranged in tone from critical (Deaton, 2003) through skeptical (Lynch et al., 2004), to enthusiastic (Wilkinson & Pickett, 2006). Interestingly, though they disagree in their conclusions regarding the relationship between inequality and health, all three reviews find strong evidence for a relationship between inequality and murder.
Population health indicators like average life expectancy, the infant mortality rate, and the murder rate are aggregates of individual-level variables. Income inequality, on the other hand, is an ecological-level variable: it is an attribute of the social unit that cannot be disaggregated to individual-level variables. Thus, when investigating the impact of inequality, it is important to identify the proper ecological unit within which inequality in the distribution of income is hypothesized to impact health. If, as cohort (Fuhrer et al., 2002, Marmot and Smith, 1997) and hormonal response (Dickerson & Kemeny, 2005) studies suggest, the effect of inequality on health is mediated by individuals' positions in hierarchies of social stratification, it is important to operationalize inequality using the primary ecological unit within which people are stratified. Although there is much debate about just what is the most appropriate ecological unit for studying the relationship between inequality and health (Subramanian & Kawachi, 2004), the general practice in sociology has historically been to treat the country as social system in which people are stratified (Wright, 1979). While the study of other units may yield important insights, the study of national income inequality is certainly an important part of the puzzle.
While at least 45 studies have examined the ecological correlation between income inequality and population health at the country level (Wilkinson & Pickett, 2006, p. 1770), few recent studies have been based on broad panels (N > 100) including both developed and developing countries (Beckfield, 2004, Ellison, 2002), and none have examined the stability of the correlation between inequality and health for such a large panel of countries over an extended time period. A difficulty in interpreting the plethora of results based on panels of limited geographical and temporal scope is that the effects of inequality may be inconsistent over samples and periods. Another is that there are certainly strong selection biases embedded in the happenstance of data availability, and these biases are likely to be related to all three key variables in the literature: income inequality, population health, and national income per capita. For example, countries experiencing government turmoil are likely to have poor population health outcomes, as well as to be missing data. Less dramatically, countries with lower levels of government administrative capacity are less likely to report income inequality data at frequent intervals, and thus are underrepresented in most studies. Exacerbating these difficulties, country-level income inequality measures are very unreliable, exhibiting wide variability year-on-year, and not consistently operationalized across geography and time.
These challenges highlight the importance of assembling the widest possible panel of countries for use in analysis. Though there are many ways to measure income inequality (see Cowell, 1995 for a review), in practice only Gini coefficients are sufficiently widely available to be used in a broadly cross-national study of the correlates of inequality. The Gini coefficient possesses the four basic qualities of a good inequality measure – anonymity, scale independence, population independence, and satisfaction of the transfer principle (Babones & Turner, 2003) – but it is in many ways technically inferior to other measures, such as the Theil entropy index, which in addition to other advantages allows subgroup decomposability. Moreover, methodological inconsistencies in the construction of the Gini coefficient are a particular problem: there is no one standard set of methods used to measure income inequality. The required methodological adjustments are not small; they are typically of the order of 6–8 Gini points (the Gini coefficient runs from 0 to 100, with 90% of all observed data falling approximately between 30 and 60) (Babones & Alvarez-Rivadulla, 2007). This implies that typical error levels due to the inappropriate operationalization of income inequality are of the order of 20% of the observed range of national inequality levels. Many existing studies have simply adopted published income inequality series unquestioningly, seemingly unaware of (Mellor & Milyo, 2001) or unconcerned with (Gravelle, Wildman, & Sutton, 2002, p. 578) the problems in their construction. Though Gini coefficients may not be ideal measures of inequality, researchers must, nonetheless, be careful to use them as appropriately as possible.
More sophisticated recent studies of inequality and health at the country level have attempted to adjust for the methodological inconsistencies in the raw inequality data, but have committed grave errors in execution (Beckfield, 2004, Ellison, 2002). In these studies, population health is regressed on income inequality plus a set of indicator variables that are meant to control for the methodology of operationalization of the Gini coefficient. This approach results in a regression equation that models health as a function of inequality and the method of operationalizing inequality. There is no reason to believe that the method of operationalizing inequality has any effect on health, except spuriously (countries with weaker statistical services tend to have expenditure-based, rather than income-based, inequality data), and in fact the methodological indicator variables in such models tend to be non-significant (Beckfield, 2004, p. 239) despite the highly significant effects method of operationalization is known to have on measured inequality (Babones & Alvarez-Rivadulla, 2007). Since any error in the explanatory variable tends to attenuate the strength of the resulting regression estimates, these oversights and errors have created a potentially serious downward bias in the reported relationship between income inequality and population health.
The present study uses data and methodological notes from the broadest existing database of Gini coefficients, published by the World Institute for Development Economics Research (2000), to estimate continuous, internationally and intertemporally comparable country-level income inequality series for the widest possible panel of countries, then uses these data to investigate the relationship between inequality and three population health outcomes (life expectancy, infant mortality, murder rate) over the period 1970–1995. The maximum number of countries represented here in cross-sectional analyses (up to 137 countries for 1995 and 126 countries for 1970) is larger than that of any previous study, as is the number of countries studied longitudinally (125 for life expectancy and 126 for infant mortality, about half of which exhibit significant trends in inequality). In addition, this study reports for the first time upper bound estimates for the potential size of the so-called “income artifact” (Gravelle, 1998) through which the curvilinear relationship between income and health is hypothesized to account for the observed correlation between inequality and health. The results of these analyses indicate that income inequality is non-artifactually correlated with population health, with suggestive (but not conclusive) evidence that this relationship is causal.
Section snippets
The relationship between inequality and health
The most comprehensive study to date of the relationship between income inequality and population health at the country level reports a correlation of r = −0.31 between inequality and life expectancy and a correlation of r = 0.33 between inequality and infant mortality for an unbalanced panel of 115 countries (Beckfield, 2004, p. 236), while a separate study pegs the correlation between inequality and homicide at r = 0.40 for a panel of 50 countries (Lee & Bankston, 1999, p. 39). These correlations
Non-spuriousness of the relationship
It has become widely accepted that at least a portion of the observed relationship between income inequality and population health is attributable to the curvilinear relationship between income and health (Gravelle, 1998). The argument, in a nutshell, is that a rise in income at the low end of the income distribution has a greater effect on health than does an equivalent rise in income at the high end of the income distribution: $1000 buys more health for an individual with an income of $10,000
Testing the causality of the relationship
The mere existence of a non-artifactual correlation between income inequality and population health does not, in itself, imply a causal relationship between the two variables. Customary criteria for causality include a correlation between two variables, temporal precedence (with the causal variable manifesting in advance of the dependent variable), and the elimination of plausible alternative explanations for the correlation between the two variables. Correlation has been established, and the
Conclusions
The results reported in this study clearly point to two firm conclusions. The first conclusion is that there is an unambiguous link between income inequality and population health at the country level. Prior research has debated this point, but the results reported here are sufficiently robust, are sufficiently stable over time, and cover a sufficiently large proportion of the countries of the world to be considered definitive. The correlations of income inequality with life expectancy and
References (34)
- et al.
Socioeconomic factors, material inequalities, and perceived control in self-rated health: cross-sectional data from seven post-communist countries
Social Science & Medicine
(2000) Letting the Gini out of the bottle? Challenges facing the relative income hypothesis
Social Science & Medicine
(2002)- et al.
Income, income inequality, and health: what can we learn from aggregate data
Social Science & Medicine
(2002) - et al.
National infant mortality rates in relation to gross national product and distribution of income
Lancet
(1999) - et al.
Income inequality and population health
Social Science & Medicine
(1998) - et al.
Income inequality and population health: a review and explanation of the evidence
Social Science & Medicine
(2006) - Babones, S. J., & Alvarez-Rivadulla, M. J. Patterns of inequality in the world-economy: Introducing the Standardized...
- et al.
Standardized income inequality data for use in cross-national research
Sociological Inquiry
(2007) - et al.
Global inequality
Does income inequality harm health? New cross-national evidence
Journal of Health and Social Behaviour
(2004)
Measuring inequality
Dynamics of income distribution
Health, inequality, and economic development
Journal of Economic Literature
Acute stressors and cortisol responses
Psychological Bulletin
Does economic growth benefit the masses? Growth, dependence, and welfare in the third world
American Sociological Review
Socioeconomic position, health, and possible explanations: a tale of two cohorts
American Journal of Public Health
How much of the relation between population mortality and unequal distribution of income is a statistical artefact?
BMJ
Cited by (190)
The effect of housing price inequality on mental health
2023, Labour EconomicsStudying social determinants of health using fuzzy-set Qualitative Comparative Analysis: A worked example
2022, Social Science and MedicineInequality and Life Expectancy in Africa and Asia, 1820‐2000
2022, Journal of Economic Behavior and Organization
- ☆
This paper has benefited greatly from close readings by Richard G. Wilkinson and Siddharth Chandra.