The Spirit Level revisited
10 June 2015
SUGGESTED
Government and Institutions
22 May 2015
Regulation
26 May 2015
Trade, Development, and Immigration
4 November 2024
Five years ago, in May 2010, I published The Spirit Level Delusion in response to Richard Wilkinson and Kate Pickett’s book The Spirit Level which was being widely read at the time. Using similar scatter-plots to those used in The Spirit Level, I showed that most of the statistical associations between inequality and social problems that were reported by the authors were the result of selection bias. This was certainly true of the key assertion that income inequality causes poor health and shorter life expectancies (leading to the claim that “inequality kills”).
As shown in The Spirit Level, this is the relationship between income inequality and life expectancy. There is a modest (r2=0.20) correlation, with the less equal countries appearing to have shorter life expectancies.
But, as I said in my rebuttal, the data used by Wilkinson and Pickett for this graph were peculiarly old. Indeed, I got my first hint that all was not as it seemed in The Spirit Level when I turned to the references and noticed that the authors used life expectancy figures from a 2006 report for one graph but used figures from a 2004 report for the graph shown above. Using the more recent figures weakens an already weak relationship, but a more fundamental problem was the exclusion of several countries from their analysis. Wilkinson and Pickett provide a justification for only studying rich countries in The Spirit Level, but there are a number of rich countries that are needlessly excluded. When those countries are added and the data from the 2006 report used, a rather different picture emerges, as the graph below shows:
Since it’s been five years since my rebuttal was published, I decided to look at the most recent life expectancy stats and see how The Spirit Level was holding up. The results are interesting. Even if you limit the analysis to The Spirit Level‘s questionable group of countries, the association with inequality has completely disappeared (r2=0.02). This remains true if you include the countries added above and if you use different measures of inequality.
It seems that the relationship between inequality and life expectancy only holds when we use data from early in the last decade and arbitrarily exclude a number of countries. It fails the basic scientific test of reproducibility. A law that only works under certain circumstances and in certain years is no law at all.
The fact that a correlation no longer exists even when we confine the analysis to the countries that were specially selected in The Spirit Level does not leave much wriggle room for the book’s authors. It is difficult to exaggerate to importance of the supposed link between life expectancy and inequality to The Spirit Level‘s argument. Richard Wilkinson made the health-inequality hypothesis his own in the 1990s, with the other connections made in The Spirit Level following on from this ‘discovery’. More than ever, that discovery looks like fool’s gold.
Christopher Snowdon is the IEA’s Director of Lifestyle Economics. He is the author of The Spirit Level Delusion.
As shown in The Spirit Level, this is the relationship between income inequality and life expectancy. There is a modest (r2=0.20) correlation, with the less equal countries appearing to have shorter life expectancies.
But, as I said in my rebuttal, the data used by Wilkinson and Pickett for this graph were peculiarly old. Indeed, I got my first hint that all was not as it seemed in The Spirit Level when I turned to the references and noticed that the authors used life expectancy figures from a 2006 report for one graph but used figures from a 2004 report for the graph shown above. Using the more recent figures weakens an already weak relationship, but a more fundamental problem was the exclusion of several countries from their analysis. Wilkinson and Pickett provide a justification for only studying rich countries in The Spirit Level, but there are a number of rich countries that are needlessly excluded. When those countries are added and the data from the 2006 report used, a rather different picture emerges, as the graph below shows:
Since it’s been five years since my rebuttal was published, I decided to look at the most recent life expectancy stats and see how The Spirit Level was holding up. The results are interesting. Even if you limit the analysis to The Spirit Level‘s questionable group of countries, the association with inequality has completely disappeared (r2=0.02). This remains true if you include the countries added above and if you use different measures of inequality.
It seems that the relationship between inequality and life expectancy only holds when we use data from early in the last decade and arbitrarily exclude a number of countries. It fails the basic scientific test of reproducibility. A law that only works under certain circumstances and in certain years is no law at all.
The fact that a correlation no longer exists even when we confine the analysis to the countries that were specially selected in The Spirit Level does not leave much wriggle room for the book’s authors. It is difficult to exaggerate to importance of the supposed link between life expectancy and inequality to The Spirit Level‘s argument. Richard Wilkinson made the health-inequality hypothesis his own in the 1990s, with the other connections made in The Spirit Level following on from this ‘discovery’. More than ever, that discovery looks like fool’s gold.
Christopher Snowdon is the IEA’s Director of Lifestyle Economics. He is the author of The Spirit Level Delusion.
6 thoughts on “The Spirit Level revisited”
Comments are closed.
SHARE
There is a modest (r2=0.20) correlation, with the less equal countries appearing to have shorter life expectancies
An r^2 of 0.2 corresponds to a correlation of 0.45, which most people would call moderate, not modest.
Looking at those graphs, I’d say racial composition is a better explanation of life expectancy differences than inequality.
these results suggest to me there are a number of omitted variables. The whole idea of regressing (linearly) life expectancy on a measure of inequality seems crazy to me.
Not sure what you are looking at Pat – that is a 0.02 r2, not 0.20. And of course the r2 is the measure of how much of the variance is being explained by this rather than other factors, so is a good measure of the importance of a factor to the explanation.
The basic point of the analysis confirming the failure of the Spirit Level hypothesis in the way that Christopher predicted is pretty much game, set and match in any honest academic debate.
Interesting to see what happens when 2014 life expectancy numbers are used. Do you have any data for changes on the x-axis over time, Christopher, or have “inequality” numbers remained stable for these countries since Richard Wilkinson and Kate Pickett published their work? (I’m wondering why you used updated figures on one axis only in your fresh scatter diagram). Would you please explain how the income inequality scale works. Its not based on GINI coefficients is it. Also, how confident can we be about the quality of the data used in these studies? Shouldn’t we be putting error bars on all the points? If so, how is this taken into account by the statisticians when calculating r-squared values?
After I originally left a comment I appear to have
clicked on the -Notify me when new comments are added- checkbox and from now on whenever a comment is added I recieve 4 emails with the same comment.
Is there a means you can remove me from that service?
Kudos!
Teddy – I can’t do that from here, but there should be a link at the bottom of these mails, which says “click here if you don’t want to receive further notifications.”