Few would take exception to the idea that an improvement in the material well-being of the poor would enhance not only their living standard but their health levels as well. A number of influential recent studies, however, purport to show that inequality in income--not poverty per se--has detrimental health consequences. This “inequality hypothesis” is meant to apply to everyone, regardless of wealth or social standing, and predicts that the risk of illness depends upon whether one lives in a society that is stratified or egalitarian. Thus, according to this hypothesis, while the poor may suffer the most from inequality, the better off and even the rich suffer as well.
This is a dramatic claim--and one with potentially far-reaching implications. It extends far beyond the current paradigms upon which contemporary Western social welfare policy is premised. Current welfare policy, after all, posits that overall national health can be improved by transferring resources from society’s more affluent members to its poorest and most vulnerable groups. The inequality thesis, by contrast, would seem to suggest that simply taking wealth away from the rich--and thereby reducing measured economic inequality--should in itself produce an improvement in national health. Indeed, the inequality thesis suggests that, all other things being equal, a cutback in the income of the well-to-do could be expected to improve the health status of the poor, and possibly the rich themselves--even if society were left with a lower average income level as a result of those cutbacks.
It is hard to overstate how quickly and thoroughly the inequality hypothesis has become conventional wisdom among many medical sociologists and public health scholars. It is not confined to the radical left. In Unhealthy Societies, for example, Richard G. Wilkinson of Nottingham University Medical School argues that income inequality is “one of the most powerful determinants of health” and “the most important limitation of the quality of life in modern societies.” The academic world is by no means immune to fads, but the sudden popularity of the inequality hypothesis--in schools of public health, scholarly journals like the American Journal of Public Health, institutions such as the American Public Health Association, and health philanthropies like the Robert Wood Johnson Foundation--is quite extraordinary.
Additionally, the notion that income inequality is bad for health has recently surfaced in political discussions, taxpayer-funded policy research forums, and the popular media. British Prime Minister Tony Blair has stated, “There is no doubt that the published statistics show a link between inequality and health.” The World Bank has dedicated an entire web page to the inequality hypothesis. In the United States, the National Institutes of Health, Centers for Disease Control and Prevention, and reporters for the New York Times and Washington Post, among others, have given favorable coverage to the subject. The hypothesis has become accepted wisdom among public health researchers and epidemiologists.
But the enthusiasm of many researchers and observers goes well beyond what might be warranted by the weight of the evidence alone. A very persuasive, if less publicly heralded, body of scholarship that challenges the inequality hypothesis is currently emerging. To get a better sense of this important debate, it is useful to examine both sides of it--the evidence adduced to support, refute, and qualify the inequality hypothesis. In addition, it is necessary to evaluate the study methodologies and data interpretation, as well as policy recommendations, of both sides. Against this background, it appears that the evidence and arguments for the inequality hypothesis are wanting in many respects, and that a number of influential scholars have jumped to policy conclusions on the basis of ideologically appealing, but technically dubious, findings.
Origins of the Hypothesis
The income inequality hypothesis originated as an ad hoc explanation for the repeated observation that income inequality (the extent to which wealth is concentrated or dispersed over a population) is associated with mortality levels: The greater the degree of inequality, the higher the mortality levels in that population. From correlation was assumed causation. Antecedents of this line of academic thinking can be traced back at least to the 1970s, when income inequality emerged on the margins of the public health literature in the form of neo-Marxist jeremiads.
This line of argumentation was particularly associated with Johns Hopkins Professor Vicente Navarro and the publication he edited, the International Journal of Health Services. Navarro and his colleagues maintained that the capitalist system necessarily generated both economic inequality and ill health, and that it did so in rich and poor countries alike. This initial neo-Marxist thesis linking capitalism, inequality, and disease, however, was fundamentally nonquantitative, and it relied more on assertion and Marxist scripture than careful data analysis to bolster its case. It would take almost two decades for some of Navarro’s tenets to be tested in a more quantifiable manner by more mainstream scholars.
One of the earliest studies arguing for a causal link between health and income inequality appeared in the British Medical Journal in 1992. In “Income Distribution and Life Expectancy,” Richard Wilkinson of the University of Nottingham Medical School compared nine Western industrialized countries and reported that those with less income inequality had populations with longer life expectancies. Wilkinson has been the most vocal proponent of the so-called social-production theory of health. Since the appearance of his seminal article, well over two dozen research studies and commentaries confirming the income inequality hypothesis have been published.
Wilkinson and other supporters of the hypothesis argue that health is one of the most sensitive indicators of the social costs of income inequality. Beyond a relatively modest level of economic development, they argue, further advances in standard of living do not seem to matter much, and the linear relationship between life expectancy and income breaks down. This observation prompted Wilkinson to ask the following question: How can one country “be more than twice as rich as another without being any healthier,” particularly when it comes to life expectancy?
The many lines of exploration that have been pursued to answer this question assume two main types: aggregate data studies and individual-level analyses. (A third type of study exists--observational examinations of human and animal social hierarchies--but this is not the place to discuss these.) These studies, which help establish patterns of disease and health status in relation to social position, form the basis for speculation about the mechanisms by which environmental factors create stress or lead to behaviors that produce adverse health consequences.
Aggregate Data Studies
This first type of study examines correlations between aggregate levels of health (that is, the mortality of a specified population) and income inequality. These aggregate data studies purport to offer evidence that inequality affects all members of a group, not just the poorest ones. One of the earliest studies of this phenomenon was “Income and Inequality and Determinants of Mortality,” conducted in 1979 by G.B. Rodgers of the International Labour Organization. Rodgers examined income dispersion applying the Gini coefficient, a measure of income distribution, to data from 56 rich and poor countries in the context of three health measures: life expectancy at birth, life expectancy at age five, and infant mortality rate (deaths in the first year of life per 1,000 live births). He concluded that the “difference in average life expectancy between a relatively egalitarian and a relatively inegalitarian country is likely to be as much as five to ten years.” In the 1980s, a handful of studies used the Gini coefficient in analyzing the relationship between health measures and found similar results.
It is important to note that Rodgers posited at first only a correlation, not a causal relationship, between health and income. But in his 1992 study, Wilkinson suggested a causal relationship. He examined Organization for Economic Cooperation and Development member countries using data from the Luxembourg Income Study and found a high correlation between life expectancy and the proportion of income earned by the bottom 70 percent of the population.
Wilkinson concluded that differences in per capita gross national product (GNP) alone could not explain more than 10 percent of the variance in life expectancy. He noted further that mortality rates are not related to per-capita economic output but rather to the scale of economic inequality in each society. The association was unaffected by adjustments made for average absolute income level and remained evident across a range of decile shares of income distribution. Subsequent to Wilkinson’s report, a series of crossnational studies have demonstrated that the more even the distribution of income, the higher the life expectancy.
Robert J. Waldmann of Columbia University complemented these findings by using another measure of inequality. In a 1992 article for the Quarterly Journal of Economics he examined pairs of countries in which the poor (defined as the lower 20 percent of household income distribution) had equal real incomes but where the rich (defined as the top 5 percent of the household income distribution) in one country were much wealthier than in the other. He found that the infant mortality rate was higher in the half of the pair in which the rich households were wealthier. He tested for explanations other than income, including the degree of urbanization of the households, the literacy of the mothers, and access to medical services. Yet he found that none adequately accounted for the positive association, likely causal in his view, between the incomes of the rich and infant mortality.
In their book Is Inequality Bad for Our Health? Harvard researchers Norman Daniels, Bruce Kennedy, and Ichiro Kawachi provided further evidence in support of the inequality hypothesis. The authors highlight seeming paradoxes such as the fact that equally poor countries such as Cuba and Iraq do not have similar life expectancies--Cuba’s reportedly exceeded that of Iraq by about 17 years. Conversely, low GDP-per-capita Costa Rica and the high GDP-per-capita United States were said to have similar life expectancies. And comparably wealthy countries with more equal income distributions, such as Sweden and Japan, had higher life expectancies (by two to five years) than the United States. The authors conclude that “the health of a population depends not just on the size of the economic pie but on how the pie is shared,” adding that “the degree of relative deprivation within a society also matters.”
Numerous studies of the U.S. population have examined the association between income inequality and aggregate health measures at the state level. Daniels, Kennedy, and Kawachi found that, in the United States between 1980 and 1990, states with the highest income inequality showed slower rates of improvement in average life expectancy than did states with more equitable income distributions. They concluded that “the more unequal a society is in economic terms, the more unequal it is in health terms.” George Kaplan of the University of Michigan and his colleagues in a recent study found a strong correlation between inequality and death rates. In particular, the authors discovered that income inequality was significantly associated with a higher incidence of age-specific mortality, low birth weight, homicide, violent crime, work disability, welfare receipt, smoking, expenditures on medical care, unemployment, and low educational attainment. What is more, all these measures worsened with increased income dispersion.
Individual-Level Analyses and Questions of Causation
The second type of analysis measures the effect of income inequality on health after controlling for the effects of individual income. These individual-level analyses ask the following question: Can the observed correlation between inequality and health be explained by the intervention of other variables, or is there truly a causal relationship between the two? When this type of analysis is considered, the association between income inequality and health outcomes does not appear as secure as its proponents suggest.
Questions remain about the extent to which statistical artifact has been mistaken for real effect. In his 1998 article in the British Medical Journal, Hugh Gravelle of the University of York asserts that there may be a very simple explanation for some, or all, of the reported associations between inequality of income and population health used to support the relative income hypothesis. “A positive correlation between population mortality and income inequality can arise at the aggregate level even if inequality has no effect on the individual risk of mortality,” he stated. “Thus, we do not need the relative income hypothesis to explain the observed associations between population health and income inequality--the absolute income hypothesis will serve.”
International comparisons show that health improvements become smaller and smaller with increasing wealth. Thus the relationship between per-capita income and national health--however it is measured--should not be expected to be linear. To the contrary, as Jennifer Mellor of the College of William and Mary and Jeffrey Milyo of the University of Chicago argue, the function is one in which we would expect to see “diminishing returns” to average income. Furthermore, they continue, health levels should depend not only on average income levels but also on income distribution. This is because information on income distribution serves as a proxy for the number of persons at lower levels of income. Consequently, Mellor and Milyo conclude, aggregated studies do not offer convincing evidence on this matter. Harold Pollack of the University of Michigan puts it another way:
Money matters near the bottom of the distribution and may not matter at all for many outcomes when one exceeds the median. Controlling for the median income, then, any income dispersion measure is highly correlated with the percentage of the population that is under the poverty line.
The influence of particular variables is significant as well. For example, when individual characteristics replace aggregate-level mortality in the analyses, and when different years are examined, the relationship between health and income inequality often disappears. When the strong regional patterns in health outcomes that exist across the United States are ignored, spurious associations between inequality and health may result. Some, like Pollack, question the validity of one of the aggregated econometric measures used in most analyses: “Cross-sectional regressions that use inequality measures such as Gini are virtually uninterpretable.”
There are some glaring exceptions to the health and income inequality pattern. In Denmark, for example, where per-capita income is similar to that of the United States, but where income dispersion is lower, life expectancy is also slightly lower than in the United States. Thus the important but unanswered question remains: If an underlying relationship between deprivation and poor health does indeed exist, is reported annual dispersion of a society’s income the most appropriate index for describing inequality in that population?
Milyo and Mellor have questioned whether the correlation between inequality and health is in fact not causal but spurious. There are three possible interpretations of a correlation between variables A and B: either A causes B, B causes A, or A and B are independent of one another but both related to a third variable. Taking into account the well-established relationship between health and material well-being and social status, Milyo and Mellor point out obvious advantages that come with wealth: Well-off people can afford better health insurance and higher-quality care; they can demand better work environments, afford safer cars, and live in less polluted and less crime-ridden neighborhoods. In this way, being richer can make one healthier.
Yet consider the reverse dynamic: Being healthy can also make one better off. Poor physical or mental health can influence an individual’s ability to work for long hours or at all, thus limiting his income. This is known as the “healthy worker effect.” What follows could lead to further health impairment because the worker has less money with which to purchase health-enhancing goods and protections. The cumulative wear and tear on such individuals, coupled with whatever psychic stress they experience as a result of deprivation of social status, may be considerable.
Additionally, so-called third factors can account for the habits and limited opportunities that often lead to poorer health. Sedentary lifestyle, obesity, high-fat diets, aversion to medical care, and risky behavior, which typically underlie many of the differences in health status between the less wealthy and the better off, may well be the product of educational level. Better-informed people know about the importance of exercise, screening tests for cancer, and a healthy diet. They are more confident when interacting with physicians and better at negotiating bureaucracies (for example, HMOs). Personal characteristics that tend to be associated with greater life success, such as prudence, perseverance, and an ability to delay gratification, are also likely predictors of good health or competence in managing illness. Indeed, the substantial association of health with certain measures of human agency raises the question of whether income inequality itself has any appreciable direct effect on mortality. The association may instead reflect the effects of other factors--education in particular--that are also related to mortality.
Indeed, in a recent study in the British Medical Journal Andreas Muller of the University of Arkansas tested whether the relationship between income inequality and mortality in the United States is a consequence of different levels of formal education. He conducted state-by-state analyses of age-adjusted mortality from all causes and three independent variables: the Gini coefficient on income in 1989 and 1990, per-capita income from those years, and the percentage of people older than 18 who did not complete high school. An income inequality effect was found, but it disappeared when the percentage of people without a high school diploma was added to the regression analysis. Muller concluded that the lack of a high school education accounts for the income inequality effect and is a powerful predictor of mortality variation across states. He writes, “The physical and social conditions associated with low levels of education may be sufficient for interpretation of the relationship between income inequality and mortality.” These conditions likely include the risk of occupational injury, the inability to attain protective goods and services, and cigarette smoking.
Ecological Bias (in Theory)
From a methodological standpoint, most quantitative research purporting to support the inequality thesis is potentially compromised by a problem statisticians designate as “ecological bias.” Ecological bias arises when “ecological correlations”--that is to say, correlations witnessed in aggregated data--differ from the underlying correlations that would be observed if one were examining individual data.
Ecological bias is a particular risk in studies of the inequality thesis for a very simple reason: The relationship between income and mortality is highly unlikely to be linear. In general, an individual’s health will not be doubled by a doubling of income--and multiplying his or her income by a factor of 10 will not correspondingly reduce mortality odds or health risks by an order of magnitude. To the contrary, the relationship between income and mortality in almost all populations seems to be curvilinear. That is to say, additional increments in income correspond to further improvements, albeit steadily diminishing improvements. Recent international data, indeed, suggest that a doubling of a country’s per capita income is associated with an absolute increase in life expectancy at birth of just about six years.
Consider what this curvilinear correspondence between income and mortality means for aggregated data--and for measurements to test the inequality hypothesis. Even if increased income dispersion has no negative impact whatever upon health, the society with the higher level of income inequality will, all things being equal, appear to have an “unexpectedly” short life expectancy.
A simple hypothetical example illustrates the problem. Suppose we invent a country called “Equalia,” with a population of 100,000. Every person in Equalia earns the national median income of $50,000 a year, and every person lives exactly 75 years. In this country, then, life expectancy is 75, average per capita income is $50,000, and the Gini coefficient of income inequality is exactly zero (on a possible scale of 0 to 100 percent).
Now suppose an invented individual--call him “Bill Gates”--suddenly moves to Equalia. Bill’s income is $5 billion a year, and his life expectancy (thanks in part to the superb medical treatment he is able to afford) is exactly 100 years. Suppose further that Bill’s immigration leaves everyone else’s income and health totally unaffected.
What will the aggregated data show? Before Bill came to Equalia, life expectancy was 75 years; after he moved in, it was ever so slightly higher. Before he moved in, average income per household was $50,000; after he arrived, average income was virtually twice as high--essentially, $100,000 per household. And whereas the Gini coefficient for Equalia’s income distribution was zero before Bill’s arrival, post-Bill Equalia would have a Gini coefficient of nearly 50. In other words, half of the country’s income would be in Bill’s hands, and the rest would be evenly distributed among everyone else.
By the sort of analysis the inequality-thesis school favors, post-Bill Equalia would be placed on a scatter plot along with other populations and compared in terms of mortality (or life expectancy), income, and income distribution measures. Of course, life expectancy in the country would be almost identical before and after Bill’s move. But post-Bill Equalia’s income level would be far higher than that of pre-Bill Equalia, and the country’s income distribution would be more uneven. The correspondence between life expectancy and income in post-Bill Equalia would look much less favorable than in pre-Bill Equalia. Indeed, once the statistics are crunched, post-Bill Equalia might be seen as “suffering” from a national life expectancy fully six years lower than might have been expected on the basis of its income level alone. Further use of regression analysis could produce results that would demonstrate that income inequality in post-Bill Equalia had cost the nation years of lost life expectancy (against levels otherwise predicted). Yet in our hypothetical example, not a single household in Equalia had its health or income affected by Bill’s entry into the country. The adverse relationship between inequality and health suggested by nation-level data in post-Bill Equalia is entirely spurious--a consequence of ecological bias, pure and simple.
Ecological Bias (in Practice)
It is not only in the hypothetical case of Equalia that “ecological correlation” may misrepresent the true correspondence between health risks and economic stratification. Ecological fallacy may also undermine conclusions from previous studies supporting the income inequality hypothesis. It is therefore important to control for confounding at the individual level.
To that end, in a 1997 study appearing in the British Medical Journal, Kevin Fiscella and Peter Franks of the University of Rochester examined whether the relationship between income inequality and mortality observed at the population level may simply represent inadequately measured rates of income differences at the individual level. Fiscella and Franks used demographic and mortality data from 1971 and 1987 from the first National Health and Nutrition Examination Survey and related follow-up surveys. To measure income inequality, the researchers used an index that estimates the proportion of total income earned by the poorer half of the population in an area. The authors found that aggregated data replicated earlier findings that supported the income inequality hypothesis. After they adjusted for individual household income, however, no significant relation between income inequality and mortality was evident. They concluded that “income, as a measure of access to resources, and not relative inequality, better explains the relation between income and mortality.”
The fact is, few studies explicitly test whether inequality has a more pronounced effect on the health of the poor. In those that do, the results are mixed at best. Ellen Meara of Harvard University examined the relationship between various measures of household income inequality on infant mortality and low birth weight. She estimated the effects of inequality with and without state-specific effects--that is to say, taking into account the possibility that particular states might have especially good, or poor, health outcomes due to some special circumstance. After controlling for household income and other maternal characteristics, Meara found no significant correlation between income inequality and adverse birth outcomes among poorer women.
Mellor and Milyo also explicitly examined whether inequality has a particularly strong effect on poorer individuals. They controlled for the possibility that particular regions of the country might have characteristically better or worse health than other regions--possibly due to such factors as local dietary or behavioral habits that could be spuriously correlated with income inequality-- and explored whether the relationship between health and income inequality is robust across geographical units. They found that statistical association between income inequality and health outcomes is greatly attenuated once controls are added for individual income. In fact, when Mellor and Milyo controlled for variables such as education and race, they found a weak inverse relationship--that is, the more dispersed a state’s income, the less healthy the individuals.
Little research is explicitly devoted to testing the theory that perception of relative deprivation leads to illness or foreshortened lifespan. Angus Deaton and Christina Paxson of Princeton University attempted to do this by measuring inequality within birth cohorts rather than across geographical regions. The authors reasoned that people may be more likely to appraise their social status differently if they compare themselves to others at the same stage of life rather than with their neighbors. They found no robust association between inequality and mortality when inequality is measured within birth cohort. In fact, in some specifications, the association was the opposite of what the income inequality hypothesis would predict.
Social Capital As a Causal Mechanism?
Because a clear causal relationship between economic inequality and health cannot be identified, much of the inequality thesis literature to date is devoted to determining a possible intermediary factor that links the two. The main intermediate variable is surmised to be “social capital,” a concept that describes the pattern and intensity of networks among people with shared values. It gauges civil cohesion through a consideration of citizenship, neighborliness, trust, community involvement, social networks, and political participation (among other factors). Proponents of the hypothesis claim that inequality causes people to perceive their neighbors as more alien or less trustworthy than would be the case in an egalitarian society. As a result, citizens are less concerned about the welfare of their neighbors, and a decline in public health results.
Measuring social capital, however, is exceedingly difficult. Alternative indexes can produce contrasting or even contradictory readings for a given society. For this reason, the claim that social capital is the mechanism by which economic inequality adversely affects human health is problematic.
It is true that some data suggest a relationship on the individual level between health and civic cohesion. Numerous epidemiological studies have shown that people who are socially isolated die at two to three times the rate of well-connected individuals. But what about entire, socially isolated communities? If populations are not well integrated socially, as reflected in a hierarchical structure that highlights real or perceived differences in interests across individuals, what effect might this have on the health of the group?
Daniels, Kennedy, and Kawachi were among the first to address this question. They conducted a cross-sectional study of 39 states in which they examined the relationship between social capital and mortality. The aim was to estimate state variations in group membership and levels of trust.
They quantified social capital by considering the per-capita number of groups and associations to which residents of each state belonged. They also weighted responses to several items on the General Social Survey conducted by the National Opinion Research Center. The first survey item was a measure of “perceived lack of fairness,” which was measured by responses to the following question: “Do you think most people would try to take advantage of you if they had the chance, or would they try to be fair?” The second item concerned “social mistrust” and was measured by responses to the question, “Generally speaking, would you say that most people can be trusted or that you can’t be too careful in dealing with people?” So was the third: “Would you say that most of the time people try to be helpful, or are they mostly looking out for themselves?”
For each state, the authors calculated the percentage of respondents who agreed with the first part of each statement and found an association between social capital and mortality. (The authors acknowledge that their model did not consider the full range of factors that might influence income inequality and social capital, and also recognize the inability to discern the direction of causality.)
Based on their findings, Kawachi and his colleagues propose that income inequality affects health by inhibiting the formation of social capital, which in turn undermines civil society. It erodes social cohesion, as indicated by higher levels of measured social mistrust, and reduced participation in civic organizations. Lack of social cohesion, they argue, leads to a decline in engagement in activities and institutions such as voting, serving in local government, or volunteering for political campaigns. Low levels of engagement, in turn, undermine the responsiveness of government when addressing the needs of the worse-off. The authors conclude: “States with the highest income inequality, and thus the lowest levels of social capital and political participation, are less likely to invest in human capital and provide far less generous safety nets.”
But even if there proves to be a correspondence between health and social capital, it does not necessarily follow that economic inequality per se has a direct bearing on health prospects. If that were true, a causal relationship would have to exist between inequality and social capital formation. Available evidence, however, suggests that the statistical association between economic inequality and social capital is quite tenuous.
This problem is indicated in Francis Fukuyama’s 1995 study Trust: The Social Virtues and the Creation of Prosperity. In the course of that work, Fukuyama describes and contrasts countries and areas that he designates as “high trust” and “low trust” societies. In Fukuyama’s estimate, Germany, Japan, and the United States are countries with high levels of social capital, while France, Italy, Hong Kong, and Taiwan are settings where the endowment of social capital is noticeably lower.
As official economic statistics illustrate, however, no obvious correspondence exists between income inequality and social trust among Fukuyama’s exemplars. It is true that economic inequality (as measured by the Gini coefficient) is on average somewhat higher in the “low trust” than the “high trust” societies identified. But income inequality would also seem to be greater in “high trust” Germany than in “low trust” Italy. Income distribution would appear to be distinctly more skewed in “high trust” America than in either “low trust” France or “low trust” Taiwan. And within the “low trust” group, Italy’s measured level of income inequality is barely half that of Hong Kong. Indeed, according to World Bank data, outside of Scandinavia and the former Soviet bloc, no country in the world today reports a more even distribution of national income than does “low trust” Italy. Given the weak empirical foundations for any argument linking inequality and social capital, the hypothetical mechanism by which inequality would have an impact on health would seem to remain just that--merely hypothetical.
The inequality hypothesis has become an increasingly influential school of thought within public health literature. But when exploring the claims its proponents advance, the evidence they cite in their favor, and the methodologies underlying their arguments, it becomes clear that this is a theory built on stilts. How has a notion with such questionable empirical documentation--research relying far too often on limited or unrepresentative data sets, hazily expounded causality, and elementary econometric fallacies--acquired so much respect within the academy and so much authority in policy circles?
This troubling question becomes even starker when one considers that some of the important studies adduced in support of the inequality hypothesis appear to be difficult to replicate with different but analogous sets of data. The essence of the scientific method is to frame and operationalize a hypothesis whose predictions comport with observable results in a consistent manner. If the hypothesis is valid and testable, its result should be generally reproducible, rather than unique to a particular experiment. But key facets of the evidentiary foundation for the inequality hypothesis fail this basic test.
Robert Waldmann’s influential 1992 study on international infant mortality rates and economic inequality is a case in point. Although Waldmann himself has not been an exponent of the inequality thesis, his econometric analysis has become staple fare for those who argue that inequality has adverse effects on public health. Using World Bank per-capita income and income distribution data from the 1960s and 1970s for a sample of 57 countries, 41 of which he categorized as “developing,” Waldmann concluded that “infant mortality appears to be positively related to the incomes of the rich (the upper 5 percent of the income distribution) when the incomes of the poor (the lowest 20 percent) are equalized among countries.” He took this result to be so robust that he described it as a “striking empirical regularity.”
If this is indeed as striking an empirical regularity as he claims, it would be reasonable to expect a similar result using larger and more recent World Bank data. Because such data are presently available, we thought it would be interesting to repeat Waldmann’s analysis using these new data and the same regression equations that generated his striking and thought-provoking conclusions. As it turns out, the “striking empirical regularity” that Waldmann found in 1992 is by no means evident in international data from the mid 1990s.
For the sample of all countries, rich and poor, recent data affirm Waldmann’s finding of a strong and statistically significant relationship between infant mortality and the per-capita income levels of the middle-income grouping. This is hardly a surprising result, insofar as this middle group accounts for the overwhelming majority of each country’s population, and also presumably the greatest share of every country’s births and infant deaths. But where Waldmann uncovered a powerful and significant positive relationship between infant mortality and the share of income accruing to the top income grouping, our analysis finds a negligible association--less than one hundredth the scale of Waldman’s. This association, moreover, could be either positive or negative, given the weak statistical relationship itself and the large margins of error involved.
Some of our other results appear even more incongruent with Waldmann’s. When just the income level of the poor and the income share of the rich are used to predict infant mortality, Waldmann found a strong and positive association between inequality (so measured) and the infant mortality rate. But for the mid 1990s, using a substantially larger data set, that relationship is weak and the coefficient is negative. This is true for a sample including both rich and poor countries, and for the developing countries by themselves. What does that mean? Plainly put, this purportedly “striking empirical regularity” of the supposedly perverse relationship between income concentration and infant mortality is rather less “striking”--or “regular”--than proponents of the inequality hypothesis themselves apparently understand. Based on these larger, more recent data samples, the notion that a country’s infant mortality rate is directly influenced by its economic elite’s share of total income is a proposition that plainly cannot be generally substantiated.
Considering the empirically questionable nature of the data and methodology used to support the inequality hypothesis, its widespread popularity must therefore be explained in other, nonempirical terms. One reason the hypothesis may have proved so compelling in academia, as well as in the policy world, is that it points toward a social reform long favored by radical egalitarians: the redistribution of wealth. In view of the new and now well-documented activist mission of many schools of public health--that is, to promote “social justice”--it is not surprising that public health circles might react enthusiastically to a thesis that would seem to support their own preferences for an expansive social and economic policy agenda.
It seems there is a profound, almost elemental appeal to the basic premises of the hypothesis, and, most especially, to its implicit practical corollary--namely, that by restructuring society and reducing inequality, the health chances and life prospects can be improved for all. This elemental theme, however, is hardly new to social or political discourse. In an important sense, the theme is as old as the concept of modernity itself. In one form or another, versions of the same romantic, utopian call have been heard ever since writers first began to imagine that we could improve humanity by purposely refashioning the sort of society that human beings inhabited. Like the Marxist and neo-Marxist ideologies to which it is related, the inequality hypothesis is best understood as a creed or faith. To describe it as a scientific hypothesis is a misnomer; it is more accurately a doctrine in search of data.
Nicholas Eberstadt is the Henry Wendt Scholar and Sally Satel is a resident scholar at AEI.