3 More Ways Inequality Poisons Societies
by Benjamin Studebaker
Before I started writing this blog, I read an interesting book called The Spirit Level: Why Greater Equality Makes Societies Stronger by Kate Pickett and Richard Wilkinson. It occurred to me the other day, when I found myself making use of some of the book’s statistics in an argument, that I have not yet shared its findings with blog readers. This struck me as something of a massive oversight–the book establishes statistical relationships linking numerous social pathologies to income inequality, and it even shows that inequality has a much stronger influence on these pathologies than raw wealth in absolute terms.
Before we begin in earnest, a brief note–there are places in the book where the authors make their argument a bit stridently (David Runciman has an excellent review highlighting the primary issue–the authors sometimes claim that inequality is bad for everyone when their research is only strong enough to conclusively claim that it is bad for societies on average). That said, the statistics they use are quite sound and meet academic standards–if you find yourself skeptical of any of the claims in this post, I urge you to read their online FAQ, where they respond to many of the common counterarguments anti-egalitarians have tried to make in response to their work.
To be clear, the authors measure inequality by comparing the income of the top quintile (20%) with that of the bottom quintile. The other statistics are sourced from peer-reviewed academic research, and the original sources can be found at their website.
With that out of the way, let’s get started. Our three ways are:
- Healthcare–inequality results in worse health outcomes on average across a variety of metrics.
- Education–inequality results in worse child and educational outcomes on average across a variety of metrics.
- Crime–inequality results in higher homicide and incarceration rates.
Let’s discuss each in turn.
Healthcare
When inequality is higher, more infants die on average:
More unequal societies also suffer higher rates of mental illness:
The same goes for drug abuse:
And obesity:
There is disagreement over how inequality leads to health problems–the most common explanation offered is that relative inequality causes people to feel marginalized and alienated from society. It also makes social interactions more contentious, as people are made more aware of the social hierarchy and their place in it. Together, these things increase stress and anxiety, worsening mental, cardiovascular, and immune health directly and pushing people to seek potentially harmful coping mechanisms.
Education
Inequality has negative effects on children’s well-being (a composite statistic devised by UNICEF which takes into a variety of things–you can read about how it is calculated here):
Students do worse on standardized tests in unequal societies:
Unequal societies also see higher rates of teenage pregnancy:
And social mobility is significantly lower in unequal societies:
Here the causal mechanism is more straightforward–affluent parents often use their resources to give their children a variety of advantages that poorer children don’t have (time and attention, private schooling, summer experiences, etc.). This often causes poor students to perform poorly, and that means they often end up in the same economic position as their parents. Their children go on to have the same disadvantages, perpetuation a cycle of poverty.
Crime
Unequal societies have higher homicide rates:
And unequal societies imprison more of their citizens:
Here both the health and education causes come into play–because more people are mentally ill or feel alienated or marginalized, they are more likely to commit crimes, because more people have been denied education and work opportunities, they have fewer alternative legitimate courses of action available to them. These factors cause people to lash out more often, sometimes violently (see the recent Baltimore riots, for instance).
Inequality persists in part because so many people presume that only absolute poverty matters–that poverty is only a serious social concern when people are without food, shelter, or other such things. In political theory terms, we often call such people “sufficientarians”–they believe that justice is achieved if everyone achieves some minimal standard of sufficiency, regardless of how unequal the society may be as a whole. The Spirit Level helps us see why sufficientarianism is not sufficient–even if our society’s poor have the basics of life, inequality will still lead to feelings of inferiority, alienation, marginalization, and hostility. It will still give some young people opportunities that are denied to others. These things will produce negative social pathologies even if poor families are at the minimal sufficiency standard, and in many rich countries some people are not even that well off–even today, there are significant numbers hungry and homeless people living in highly developed countries.
We continue to see no change because people are unaware of The Spirit Level statistics and because people are too quick to blame the poor for the problems. Right wingers argue that poor people can overcome their sense of inferiority, alienation, or lack of opportunity by working harder. They insist that it’s not about the cards you’re dealt, but what you do with them that counts. This emphasis on the individual is naive and unrealistic. People are not healthier, better educated, and less likely to go to prison in Norway or Japan because they are harder working or more virtuous but because of systems of social policies designed to promote well-being and social cohesion. More equal societies recognize that when individuals under-perform, it is not because those individuals are defective–it’s because the societies are. They move resources, opportunities, and incentives around to produce better behavior. They don’t stand around waiting for individuals to transcend their own socioeconomic backgrounds–they make better backgrounds in the first place.
That’s what we should all be trying to do.
Interesting. I’ve heard some correlations between US infant mortality and the number of hospital interventions performed on laboring mothers. Not saying this is the cause but I wondered how the two were related, and now, if that ties back to inequality.
I believe the correlation holds even if you take the United States out completely.
[…] claim that relative poverty is not important at all. This is not true–I recently wrote a post detailing all kinds of evidence that relative poverty and relative inequality contribute to a […]
Reblogged this on chonniu .
i must disagree with you here Benjamin. I should note that the vast majority of health economists (if they even know about the book) tend to see it as an exaggeration. There are 4 issues that most 2nd or 3rd year econometric students will tell you imply these results are biased:
I have written about them here http://www.iea.org.uk/blog/is-the-spirit-level-right-about-inequality-and-obesity
But in general the analyses found in the book:
1) Do not control for possible (and obvious) other variables (the most obvious is poverty or low SES)
2) Do not account for possible reverse causation (simultaneity bias)
3) Use aggregate measures whereas cohort studies of individuals are widespread (in my blog you will see I mention an analysis I did myself on cohort data)
4) They rarely use measures of statistical significance which is the norm in peer reviewed medical statistics etc.
A number of issues with this:
1) There’s lots of debate over whether Gini coefficient is the best way to measure inequality. The Spirit Level guys compared the ratio of the 80th and 20th percentiles, Piketty likes to use the income of top end percentiles, and so on–there are many ways to do it, depending on what precisely one is looking for. This is a particularly important issue when you’re looking at small areas (i.e. regions and districts) because Gini coefficients tend to skew low relative to other measures in these areas. Check out this stuff:
http://www.mitpressjournals.org/doi/abs/10.1162/rest.2003.85.1.226#.VUvRw45VhHw
Click to access 200801_convergence.pdf
2) I don’t think cohort studies are particularly relevant–the claim is not that exposure to inequality directly causes inequality for specific individuals, but that inequality produces higher rates of obesity on average. There are a number of ways it can cause this (emotional eating is one, another is stratified access to healthy food as opposed to junk food–now that I think about it, this is probably more likely, and it would not show up in a cohort study–consider the differences in diet between someone who eats at Wal-Mart and someone eating at Whole Foods. This difference is caused by inequality, but it leads to obesity disproportionately in people at the bottom of the distribution).
3) I’m not clear on how you measured poverty–this could certainly have an impact on your result.
4) The research you’re presenting focuses only on the obesity relationship (which has R=0.57). This is not the strongest relationship in the data (nor do I think it is the most important to the authors’ larger argument). Social mobility has R=0.93, imprisonment has R=0.75, teen pregnancy and mental health have R=0.73, etc.
4) Many of the social problems travel together–the same countries tend to do well on all problems, and no research has proposed a variable other than inequality that yields similar results on this wide array of issues. So I have to ask–do you see a better explanation for performance not just on obesity, but on all of the listed issues?
1) My point about other variables is that a third (or variety of other variables are driving the relationship). The easiest idea: the book is plotting inequality against X, but if it turns out on average that inequality is higher only because there a more poor people in a community then POVERTY (absolute SES not relative).
This is not a shocking idea: the hundreds of econometric analyses in this area have controls (other SES variables like education etc and perosnal variables like age)…….this is obvious (google “determinants of x”).
2) Cohort studies are the gold standard in this kind of area. Its not about regional effect on one individual…models can predict AVERAGE changes in BMI or as in my paper average changes in probability of becoming obese. So data points are individuals not aggregates where you cant measure the granularity within these aggregates.
As a health economist we prefer trials and if not cohort studies…….no one does the type of analysis the spirit level does. It’s incredibly biased just google some cohort studies.
These are conventional examples of studies in these areas (they are multilevel so basically cohort, use a wide variety of controls in their regressions and use statistical significance otherwise no one would publish them):
Cantarero, D. and M. Pascual (2007) ‘Obesity and Socio-economic Inequalities in Spain: Evidence from
the ECHP’, Economics Bulletin, 9(3), 1–9.
Carson, S. A. (2010) A Quantile Approach to the Relationship between Body Mass,Wealth, and
Inequality. CESifoWorking Paper No. 3288, Munich: CESifo Group.
Chang, V.W. and N. A. Christakis (2005) ‘Income Inequality andWeight Status in US Metropolitan
Areas’, Social Science & Medicine, 61(1), 83–96.
Chen, Z. and D. Meltzer (2008) ‘Beefing up with the Chans: Evidence for the Effects of Relative Income
and Income Inequality on Health from the China Health and Nutrition Survey’, Social Science &
Medicine, 66(11), 2206–17.
Diez-Roux., A.V. and Link., B. G et al. (2000) ‘A Multilevel Analysis of Income Inequality and
Cardiovascular Disease Risk Factors’, Social Science & Medicine, 50(5), 673–87.
.
Seriously though see the other literature in all these areas (depression, mortaility, happiness, obesity and crime). There is no implication inequality (the DIFFERENCES between levels) makes a difference. At least in obesity decades of research has shown ABSOLUTE measures of status (income and education) are determinants.
And aren’t absolute explanations more intuitive? The spirit level analysis implies if we have 2 societies identical in every way (same poverty and education levels among poor) and we chuck 50% of the income of the top 20% of income earners into the sea after say 10 years all these social problems will become worse!!!!! Very strange….
The above comment does not address the specific concerns I raised about Gini or about cohort studies and gives insufficient information about how poverty was defined. No attempt was made to offer an alternative explanation of the results. Happy to discuss this further, but I need to see that my reply is being engaged with. Got the impression my reply was not even read.
Ben your response below was silly. You did not for example respond to:
– my concern about controls (a common method for reducing bias among econometricians and medical statisticians)
-You did not respond to my issue about the possibility of reverse causality in these correlations: e.g. does crime instead cause inequality to increase
– You did not respond to my issue about statistical significance.
Some studies use a gini, some ratio measures (some both), read some of the ones below please.
Please read the vast array of literature on the determinants of health factors which use statistical methods that deal with the 4 issues I mentioned above.
This is serious – I don’t mean to sound offensive – because otherwise policy makers will naively make policy to reduce relative SES levels when the evidence seems to be that absolute SES levels matter (education, income, family mobility etc).
Do you have an email or something or facebook – I would love to discuss this in private – peoples egos get in the way in public and so I prefer that!
Have a nice day 🙂
Let me be clear–I am calling into question the methodology being used in this work.
Controls: There’s a potential issue where if you control for something that is itself caused by inequality, you will eliminate a potential causal mechanism by which inequality affects the initial variable. For instance, if you considered anyone with an income under the US poverty line to be in poverty and then you controlled for poverty, you might be eliminating the possibility that in our contemporary context, inequality contributes to obesity by causing that sort of poverty–by pushing more people under that threshold, thereby forcing them to buy inferior food products. America may have more people in poverty than Norway because America has large inequalities while Norway does not–America allows people to fall into poverty, while Norway intervenes to preserve a more equal distribution. This is the case even though in absolute terms, America and Norway have similar per capita incomes and few people in either country would meet very strict poverty criteria (such as the ones used in the developing world). In this situation, if an effort were made to reduce inequality through redistribution (or, in the specific case, reduce inequality of access to healthy food), we could conceivably see a considerable reduction in obesity as a result. Research that controlled for poverty and came to the conclusion that inequality doesn’t matter could totally miss this and be used by opponents of redistribution to obstruct good policies. Similarly, educational performance is a variable The Spirit Level claims is in part caused by inequality, so if you control for education you may be diminishing your ability to measure the effects of inequality (perhaps people of poor education are less likely to make healthy food choices). Controls sound nice in theory, but in social science we have to be careful that our controls are not themselves mechanisms by which the independent variable acts upon the dependent variable. This means we have to think very carefully about the many possible ways social phenomena can affect one another. Econometric analysis sounds great, but if little consideration has been given to the theory of what is actually going on in these societies, it can lead us astray.
Reverse Causation: If reverse causation were operable, we would need an alternative explanation of why all these social problems travel together (the same countries performing well and the same ones performing badly). No such alternative explanation has been offered or established in the research.
Cohorts: no one is suggesting that if you take a person in the middle or upper end of the distribution and relocate that person from an equal society to an unequal one (where the person remains in the same place in the distribution), this will somehow cause that person to become obese. What is being suggested is that people at the low end who have inferior access to healthy food are more likely to be obese, and there are likely to be more people at the low end in unequal societies, such that more unequal societies see higher obesity rates on average. A cohort study is not going to see this, particularly a cohort study that controls for poverty and dismisses the various ways that inequality can lead to unnecessary poverty in modern societies.
Gini/regional inequality measures: even if we throw out the Gini data, another major issue is that if you measure inequality in a region that is, as a whole, wealthier than the other regions in the society (but quite stratified internally), you’ll get a situation where it looks like high inequality doesn’t affect obesity. But what is really happening is that on the national scale, most of the people living in the region are in the upper part of the distribution and thereby immune to the specific causal mechanism (inequality pushes people down the income distribution which diminishes their access to healthy food). On the regional level, people can appear to be on the low end of the distribution when on the national level they are in the middle or on the high end. And the same thing in poor regions–people can appear to be in middle or on the high end when nationally they are at the low end. The smaller the regions, the bigger this error–a study on a local area is counting people as rich or poor who don’t count as rich or poor at the national level and thereby may not have the same access or lack of access that is associated with wealth or poorness in that society.
The link between inequality and some kind of poverty or deprivation does not originate with me–in his review, Runciman writes:
“What these graphs tell us is that overall there is a better chance of getting fat or dying young if you live in an unequal society. But it doesn’t follow that almost everyone is going to benefit from increased equality. That depends on whether the disadvantages of inequality are distributed across the social scale, or whether they cluster at the bottom. One possible explanation for the poor showing of unequal societies like the US might be that the bottom 20 per cent are hopelessly cut adrift from the benefits of prosperity, and this group does so badly in quality-of-life terms that it brings the average down for the society as a whole. If a significant minority of people are dying very young, or growing very fat, or learning very little, then the average scores will be worse, but it doesn’t follow that almost everyone is worse off.”
It seems to me that these studies specifically exclude the possibility that inequality causes obesity via this mechanism, and that this is a huge mistake. We must account for the possibility that inequality creates a 20% (or some other %) of the population that is hopelessly cut adrift.
Ben I cant reply to your comment at the bottom of our discussion for some reason (??) – but I will provide a quick response.
You seem to be arguing controlling for other (absolute determinants like education and income/poverty) of these bad things (e.g. crime, obesity) is wrong because it could be inequality causing these bad things.
First I will reiterate: regression analysis is the norm in all health economics and medical statistics.
But in particular you are saying that what if there is a strong correlation between 2 X variables (the absolute determinant and inequality in the regression:
(bad thing) = a(absolute determinant) + b(inequality) +….u.
This is called multicollinearity and it is not usually a problem if the sample is big enough. It is NEVER an option in statistics to not control for a variable because of it – not controlling for a variable causes a much bigger issue. It means that the effect size of inequality in our case (and even sometimes direction of effect!!!!) is polluted by this “absolute determinant” (in our example) – this is called “ommited variable bias”. THIS IS THE VERY REASON FOR REGRESSION ANALYSIS.
Second I think we disgree somewhere else. There is an incredible difference between inequality and poverty. Technially it is completely plausible for inequality to go down while poverty goes up OR for inequality to go up while poverty goes down.
In fact key policies sometimes help the poor but also help the rich proportionally more (e.g. some case of free trade). so for example inequality globally is going down (almost everyone agrees) because of globalisation and the rise of some countries, but inequality within some countries has gone up (e.g. China).
Another great example. hong kong is one of the most unequal places on earth but also has one of the highest human development indexes.
It is certainly theoretically possible for poverty and inequality to become de-linked, but in the countries we’re talking about (affluent western democracies), policies that cause inequality create social problems in part because the inequality often causes some people to be denied access to certain resources and opportunities that other people in the society may have. Education is one of these resources, so is healthy food. Research that does not account for this is going to run into trouble.
I think you are overconfident in the ability of regression analysis to deal with this sort of problem. In what I’ve read about regression analysis, collinearity among two variables can only be dealt with either by getting a data set where the variables are no longer collinear or dropping one of the collinear variables from the data set. A basic assumption of regression analysis is that the variables are not dependent on one another. Here are examples of others making a similar point:
http://blog.minitab.com/blog/understanding-statistics/handling-multicollinearity-in-regression-analysis
Click to access lecture-17.pdf
http://psychologicalstatistics.blogspot.com/2013/11/multicollinearity-and-collinearity-in.html
[…] bullying to child abuse. These things may be interrelated. We know, for instance, that inequality contributes to […]
[…] we are bad at both of these things does not necessarily mean that one causes the other. This chart compares the rates of mental illness to economic […]
[…] They are also much more economically equal than the US and outperform the US on lots of other metrics. They win on drug use: […]