Spirit Level Criticisms

Recently, Wilkinson and Pickett's book The Spirit Level has been subject to strong criticism from researchers on the Right of the political spectrum (see for example Saunders 2010, Snowdon 2010 and Taxpayers Alliance 2010). The criticisms fall into three main categories:

In many cases the correlations are not robust to the removal of 'outlying' observations. For example Saunders finds that the relationship between inequality and the homicide rate disappears when the USA (which has high inequality and a very high rate of homicides) is excluded from the sample.

The findings are cross-sectional correlations which do not take account of other causal factors (apart from inequality) which might affect the health and social outcome measures. For example, Wilkinson and Pickett find correlations between a range of social problems and inequality at state level in the USA but Saunders argues that once the proportion of black people in each state is added as an explanatory variable, the relationship between inequality and social outcomes disappears. Snowdon cites a review of the economics literature on the relationship between income inequality and health by Deaton (2003) which argues that there is little if any causal impact of income inequality on health once the effects of individual incomes on health are controlled for.

The cross-sectional findings of The Spirit Level cannot be replicated across time for individual countries. For example the increase in inequality in the UK between the late 1970s and the early 1990s was not accompanied by any downward trend in life expectancy.

Wilkinson and Pickett respond to these criticisms in a paper published on the Equality Trust website in July 2010 (Wilkinson and Pickett 2010b). In terms of the removal of outlying observations, Wilkinson and Pickett point out that their analysis picks a set of countries and runs each analysis of correlation on the same sample (subject to good quality data being available), rather than 'cherry-picking' countries to include or exclude. In most cases the critics of The Spirit Level give little justification for excluding countries from specific regressions, other than that it is those countries which are driving the results most strongly - a technique which is almost guaranteed to result in finding no relationship between any two variables in a scatterplot!

The failure to take account of causal factors in the analysis is a more serious criticism. The causal mechanism through which income inequality affects health and social outcomes is not always clear or straightforward. However, as I argue in the next section, The Spirit Level's basic message - that higher inequality is bad for a range of health and social outcomes - holds up under several different causal interpretations.

Finally, on the time-series evidence, a big problem here is that when the level of inequality in a country undergoes a big change (as happened in the UK in the 1980s) it's not clear how long the effects of inequality on health or social outcomes should take to pan out. For example, life expectancy is by its very nature a lagged 'performance measure'. To take an example where the causal mechanisms are much clearer, there is typically a 20-year lag between a person's decision to take up tobacco smoking and increased incidence of lung cancer. Given that the effects of inequality on life expectancy are likely to be more subtle and nuanced than the effects of tobacco, and given that the increase in inequality in the UK occurred during the 1980s, it is perhaps not surprising that no strong time-series trend has been found, at least so far.

In general, the criticisms of The Spirit Level identify some areas where further clarification of the argument is useful (and has been forthcoming from the authors). But they do not overturn the basic message - that greater inequality is linked to worse outcomes on a range of health and social measures.

Notes

The response from the political right has not been wholly negative. For example David Cameron welcomed The Spirit Level's findings in his Hugo Young Lecture in London in November 2009 - although the Equality Trust (the non-profit organisation set up by Wilkinson and Pickett to promote further research on the issues raised by the book) questioned his interpretation of the conclusions. See 'Has David Cameron missed the point?', http://www.equalitytrust.org.uk/node/234

Some confusion seems to have been caused by the fact that Wilkinson and Pickett used the world's fifty wealthiest countries (identified using the World Bank's database) rather than the fifty countries with highest income per head. This explains why, for example, Portugal is in The Spirit Level whereas the Czech Republic isn't: Portugal has lower income per head but higher wealth per head than the Czech Republic. Also, Wilkinson and Pickett exclude countries with a population of less than three million because a lot of small-population, high-wealth countries are tax havens. Saunders (2010) argues in particular that countries with between one and three million population should be included, and the list of included countries should be chosen on the basis of income rather than wealth. However, the inclusion of these extra countries doesn't appear to change the general pattern of correlations for the most part from those outlined in The Spirit Level. Rather, it is the exclusion of outlying countries (based on arbitrary criteria) which drives most of Saunders's results.