After finding this post about income inequality and social problems, I decided to check out the ‘book version’, The Spirit Level: Why More Equal Societies Almost Always do Better, by epidemiologists Richard Wilkinson and Kate Pickett. It has a lot to recommend it (if you want the short pdf version, it’s available here). First, the evidence for a correlation across countries and within countries between income inequality and social misery is pretty overwhelming. Second, in the introduction, as opposed to an appendix nobody will read, Wilkinson and Pickett explain the basics of the statistical methods they use in language that non-mathematically inclined people can understand.
The authors construct an “index of health and social problems” which equally weights (in the U.S. cases) the following*: trust, mental illness, life expectancy, infant mortality, obesity, educational performance, teenage births, homicides, and imprisonment rates. They then asked if this index is correlated with the Gini coefficient, which measures income inequality**. Here’s what they found:
I realize that the authors are trying to convince people that income inequality is a bad thing–and it is, for every economic group. The authors do that rather well. But what I found interesting isn’t just the relationship (higher income inequality means more social dysfunction), although sadly, we still have to ‘debate’ this. What’s interesting is the residual: the variation around the regression line not explained by the relationship. In the above figure, the residual of each point is the distance a point is from the line if you were to draw a line parallel to the y-axis from the regression line that ran to the data point (a bit of an oversimplification, but it will do; red and blue lines are for MA and CT):
Some states do much better than expected (e.g., MA, NJ, CT, ND, MN, NY) than would be predicted by their income inequality, while others do much worse (e.g., AL, MS, LA, AK, AR). Why these below or above par performances exist is rarely touched on, except in one case: homicides. In terms of homicide rates, Massachusetts, New York, and Rhode Island all much better than would be predicted; while there is a significant correlation, it is rather weak (i.e., large residuals in both directions). However, when the homicide rate is adjusted by gun ownership rates, the correlation becomes very tight. In other words, NY, MA, and a few other states overcome income inequality through lower gun ownership rates; likewise, the various ‘murder alleys’ don’t do so bad once you realize you can’t walk ten feet without tripping over a firearm (this does make sense, since two-thirds of U.S. homicides are committed with firearms).
On a subject less depressing than murder and poor health, I’ve made the same point about educational test scores: some states seem to do much better than their childhood poverty rates would suggest. Whatever those states are doing, other states should learn from those cases and implement those policies (and we definitely should not overturn policies in states that are educating children well).
I wish Wilkinson and Pickett had spent more time discussing the residuals (even if it were only to propose testable hypotheses), since that’s one way we can dramatically improve public policy, and thereby, quality of life.
*The various values are normalized for each metric, and the z-score is used; the separate scores are then summed (I think).
**The authors demonstrate that other measures of inequality, such as the ratio of incomes of the top tenth and the bottom tenth yield similar results. What I would find interesting is what would happen if the percentage of people living in poverty (adjusted for each state’s cost of living) were used.