The Inequality and Health Debate: What do we learn from the twentieth-century in the developed world?
An important debate in the social health literature is whether more inequality causes worse health. At some later date I’ll post a bibliography, or maybe commenters can help. In any case the list of publications is long, the contributors illustrious, and the findings varied and at odds with each other. Some of the most important papers representing a range of findings include those by Deaton, Deaton and Lubotsky, Mellor and Milyo, Lynch, et al., Kawachi, Subramanian, et al., Navarro, et al., Wilkinson, et al., and Marmot, et al.
Note that the debate is about the effect of inequality, per se, on health. Everybody knows that being rich reduces mortality and being poor increases it. The relationship between income and health (mortality, infant mortality, life expectancy, morbidity) is so well known in the literature that it is simply known as “the gradient.” It obtains at the macro and micro levels in dozens of studies. For example, let me quote Angus Deaton, who is BTW an inequality-mortality skeptic, “Men in the United States with family incomes in the top 5 percent of the distribution in 1980 had about 25 percent longer to live than did those in the bottom 5 percent. Proportional increases in income are associated with equal proportional decreases in mortality throughout the income distribution” (Angus Deaton “Policy Implications Of The Gradient Of Health And Wealth”). But I digress.
There are three basic channels through which an association between inequality and health could occur. The first two are causal in that social inequality affects individual health.
- Direct. Inequality creates stress, which is bad for health.
- Indirect. Inequality disrupts the production of health-supporting public goods or causes the production of health-reducing public bads, which is bad for health.
- Artifactual. More income improves the health of the poor more than it improves the health of the rich. (The health-income relationship is concave.) A more unequal society will have worse average health than a more equal society with the same mean income because the health gain to the rich from being much richer is not as great as the health loss to the poor from being much poorer. Note that individual income only affects individual health, but the distribution of income affects average health.
A fairly recent entry in the field is Leigh and Jencks, “Inequality and mortality: Long-run evidence from a panel of countries” (Journal of Health Economics 26 (2007) 1-24). Here is a link to a working paper version which is very similar to the published version. In a nutshell, the income share of the richest 10 percent of the population is the measure of inequality, and life expectancy at birth and infant mortality are the two main measures of health outcome.
First, Leigh and Jencks presents strong time series evidence in favor of the inequality-mortality hypothesis. Here is the history of twentieth-century inequality in a set of highly-developed countries (share of income accruing to the richest 10 percent on the vertical axis:
And here is the twentieth-centry history of improving health in these same countries. Life expectancy and infant mortality, respectively:
The story so far looks pretty clear: Most of the twentieth century was marked by increasing equality and democratization within the industrialized countries. I think that’s a pretty acceptable basic narrative even if there are some bad counterpoints (fascist periods in some of the countries, wealth-concentrating tendencies in the US through 1929 and starting again in the early 1980’s). Again the basic narrative is that the twentieth century was marked by increasing enfranchisement and economic equality, envisioned and realized by labor movements. And the twentieth century was marked by much improved health, growing life expectancy and shrinking infant mortality. These improvements came in large part from public health demanded and created by labor movements and their allies: sewerage, clean piped water, food safety regimes, vaccination, universal health insurance.
Then, Leigh and Jencks presents mild evidence (suggestive but statistically insignificant–still, not bad for 22 data points) from a current cross-section of countries in favor of the inequality-mortality hypothesis. Again, life expectancy and infant mortality:
Look how the social democracies have less concentrated income and better health outcomes.
Then, Leigh and Jencks carefully applies accepted econometric methods to see if the relationships obtain controlling for average income (they do) and then adding country fixed effects to see if changes in inequality within a country are associated with changes in health within the country (the results still obtain) and finally adding year fixed effects to control for changes in health and equality shared across all countries in an epoch (the relationship disappears completely). The abstract summarize:
In the absence of country and year fixed effects, the income share of the top decile is negatively related to life expectancy and positively related to infant mortality. However, in our preferred fixed-effects specification these relationships are weak, statistically insignificant, and likely to change their sign.
and the conclusion concludes:
Without country and year fixed effects, we also find that more inequality is associated with higher mortality. But once we include country and year fixed effects the relationship between inequality and health becomes small and statistically insignificant. The confidence intervals around our estimates are sufficiently tight to make substantively important detrimental effects of inequality on population health unlikely.
The correct interpretation of the final econometric result is: Years in which a country had relatively more equal income distribution than did other highly developed economies were not years in which the country had relatively better health than did other highly developed economies. I want to be fair to the paper because the work is careful, plausible, and consistent with standard econometric practice, but I think that the approach suffers from cannot-see-the-forest (look at the twentieth-century time series) through-the-trees (doesn’t manifest itself in year-by-year, country-by-country changes).
Here are several potential technical critiques, and I would encourage budding applied econometricians, in particular those enrolled in UMass Economics 753, to pursue these in a replication paper.
- The inclusion of fixed effects amplifies the importance of measurement error, and measurement error (in an explanatory variable) attenuates the estimate of true relationships. This is probably a valid critique of Kristin Forbes work on inequality and growth, and it may apply here as well. Measurement error in inequality seems very likely.
- The paper does not subject the time series to tests for unit root, even though the health and economic series are all clearly
trending over the century. In general, unit root problems cause spurious correlation (finding a relationship where there is
none as opposed to missing a relationship where there is one), but I’ll think more about whether either problem can occur.
- The relationship between inequality and health may be marked by long and variable lags. Jencks and Leigh experiments with lags, but the lags run from one to five years. A more reasonable scale might be that a generation-long (century-long?) struggle for equality might yield generational health benefits. Social inequality is not a stock-market ticker with accurate up-to-the-minute readings of the state variable. Neither is health.
You can see my more-or-less futile attempt to request revision of a significant misinterpretation of the results in the comments of the New Economist Blog. You can see description of the study by one of the authors and an interesting discussion at Andrew Leigh’s own blog.