Friday, 23 July 2010

Fact-checking The Spirit Level debate

Thanks to all who organised and attended the debate at the Royal Society of Arts on 22nd July. Due to the structure of the event, Peter Saunders and myself did not get the chance to reply to Wilkinson and Pickett's presentation but, although I haven't yet listened back to the debate (I'll post the video link when it is available), there are a few basic factual errors that need clearing up...


1. Trust

In response to my claim that the correlation between trust and inequality depends entirely on the Scandinavian countries, Pickett presented a graph which showed the same data (from The Spirit Level) but with the Scandinavian countries excluded. A correlation remained, albeit weaker.

This is true, but the two critical problems with their graph on trust remain: (1) As with all Spirit Level graphs, it excludes several wealthy countries; (2) it relies on data from the 1990s which has been superseded by the 2000s data (which is used in The Spirit Level Delusion). When the most recent data is used there is clearly no correlation between trust and inequality.



Source: World Values Survey


2. Happiness

Richard Wilkinson dismissed the evidence showing that happiness is not correlated with income inequality—but is (positively) correlated with income—by saying that happiness does not have a social gradient.

This is not true. Happiness certainly does have a social gradient. One of the best known demonstrations of this can be found in a paper by Robert Easterlin from 1974. It clearly shows happiness rising in line with income.


This particular article gave rise to the so-called 'Easterlin Paradox' and is one of the most famous papers in economics. It is certainly the most famous study in the field of 'happiness studies', and as such it is hard to believe that Wilkinson can be unaware of it.


Source: Does Economic Growth Improve the Human Lot? Some Empirical Evidence, Robert Easterlin, 1974



3. Health

In response to our evidence showing no relationship between inequality and life expectancy, Pickett referred to a 2009 study from the British Medical Journal. Wilkinson and Pickett also cited this study in their response to Peter Saunders and in their response to a recent critique I co-authored in the Wall Street Journal. In the latter, they said that the BMJ study "shows unequivocally that inequality is related to significantly higher mortality rates."

In fact, the BMJ study concludes that:

The results suggest a modest adverse effect of income inequality on health, although the population impact might be larger if the association is truly causal... The findings need to be interpreted with caution given the heterogeneity between studies.

Unequivocal?



4. The Spirit Level was written in 2007?

Kate Pickett referred to one of my 20 Questions, which reads:

Why do you say that the USA’s decline in homicide ended in 2005 when 2008 saw the lowest number of homicides since 1965? As you must know, America's murder rate has halved in the last two decades despite rising inequality.

All of which is true. Wilkinson and Pickett claim that the US homicide rate "started to rise again" in 2005 (p. 142). In fact, the murder rate fell in 2007 and 2008 and is now at its lowest rate since 1965.

During the debate, Pickett explained that there was a simple reason for them ignoring the ongoing decline in the US homicide rate—their book was written in 2007! That got a good laugh, but it is not true. As can be seen from the references at the end of The Spirit Level, they were still writing—and finding new sources—well into 2008.

For example:

(95) S. Bezruchkra et al., 'Income economic equality and health: the case of postwar Japan', American Journal of Public Health, (February 2008)

(298) K. Pickett & R. Wilkinson, 'People like us: ethnic group density on health', Ethnicity and Health, (September 2008)

(379) W. Hutton, 'Let's get rid of our silly fears of public ownership', The Observer, (April 2008)

Clearly there was time to acknowledge the US homicide rate in 2007, if not 2008.


5. Inequality in Japan

During the Q & A session, I mentioned that there are questions over how equal the distribution of wealth in Japan really is. I pointed out that Gini figures from the OECD show Japan to be on a par with Spain and Portugal (this was a mistake on my part—I meant to say Spain and Greece). Pickett responded by saying that these figures are for Gini before tax.

This, again, is not true. The OECD provides inequality data both before and after tax. The most recent data, from the mid-2000s, show:

Japan (before tax): 0.44

Japan (after tax): 0.32

Spain (before tax): 0.41

Spain (after tax): 0.32

Greece (before tax): 0.43

Greece (after tax): 0.32

Sweden (before tax): 0.43

Sweden (after tax): 0.23

As the OECD states in a 2006 article:

The Gini coefficient measure has risen significantly since the mid 1980s from well below to slightly above the OECD average and the rate of relative poverty in Japan is now one of the highest in the OECD area.

I have no firm view on which of the two sources (UN or OECD) have the most realistic figure for Japan, but as a point of fact the OECD figures I mentioned were after tax.

Source: OECD

The Spirit Level, the Policy Exchange and the race card

It was enough to suggest a vast right-wing conspiracy. Last week, three debunkings of the left’s new favourite text The Spirit Level appeared in as many days—first from the TaxPayers’ Alliance, then from Policy Exchange and then in the Wall Street Journal. This coincidence (I assure you it was a coincidence) was enough to rouse The Spirit Level’s authors—social epidemiologists Richard Wilkinson and Kate Pickett—into responding, to the Policy Exchange’s rebuttal at least. Disappointingly, this response was heavy on name calling and light on substantive arguments.

Wilkinson and Pickett insisted that their book was the result of “decades of research”. True enough. Richard Wilkinson has doggedly pursued the theory that “unequal societies are almost always unhealthy societies” for the last 35 years. During that time, life expectancy has risen rapidly, despite growing inequality, leaving ultra-egalitarian Denmark with the lowest life expectancy of any country studied in The Spirit Level. That none of this has swayed Wilkinson from his hypothesis is a tribute to his indefatigability, but stamina alone is not enough. People have spent their lives on more quixotic endeavours than Wilkinson, but that should not necessarily recommend them to us.

Wilkinson and Pickett stress that many peer-reviewed articles have offered at least partial support the relative income hypothesis. This is also true—albeit only in the area of health—even if Wilkinson has written a large number of them himself, but there are also plenty of peer-reviewed articles that beg to differ. Hence the long-running academic debate about inequality which The Spirit Level has done much to popularise but little to resolve.

Amongst that large body of scientific literature, there have been several suggestions of selection bias on Wilkinson’s part (that’s ‘cherry-picking’ to you and me) which have been echoed recently with regards to the The Spirit Level. Wilkinson and Pickett throw the same accusation at Peter Saunders, the author of the Policy Exchange critique, saying that he selectively removed a number of countries from his analysis. This would be a potent criticism had Saunders picked the most useful subset of countries and used them throughout. In fact, he only occasionally excludes a handful of outliers to show that Wilkinson and Pickett’s regression lines are being dragged this way and that by a few special cases, thereby creating the illusion of a sloping gradient where none exists.

It is valid, indeed crucial, to demonstrate this point. Take their graph on obesity, for example. Even the casual reader can see that the line goes upwards as a result of fat Americans and skinny Japanese. There is a conspicuous lack of a dose-response relationship when it comes to the other countries, but note the absence of Singapore and Hong Kong in this graph, both wealthy societies which marry extreme inequality with low rates of obesity (and low rates of most other health and social problems). Wilkinson and Pickett are no strangers to “arbitrarily cutting out certain countries” themselves.

Conversely, Wilkinson and Pickett accuse Saunders of including too many countries (he shows 44 to Wilkinson and Pickett’s 23). This is a view I have a little more sympathy with. Places like Russia and Chile clearly have absolute poverty in a way that France and Norway do not. Since Wilkinson and Pickett accept that wealth improves countries up to a point, it could certainly be said that several of the nations included in Saunder’s analysis have not reached that point. However, as both The Spirit Level Delusion and the TPA report show, one need only to include countries which are as wealthy or wealthier than Portugal to show that there is no relationship between inequality and most health and social problems. Wilkinson and Pickett have yet to justify their decision to exclude places like South Korea, the Czech Republic, Slovenia and Hong Kong.

Wilkinson and Pickett then aim a blow below the midriff when they accuse the Policy Exchange of being “from the political far right” and accuse Saunders of a “seriously racist slur” for showing the tendency of US states with large African-American populations to have the most social and health problems. The correlation with ethnicity is strong—much stronger than the correlation with inequality—but Wilkinson and Pickett dismiss it as “racist because it implies the problem is inherently the people themselves rather than their socioeconomic position”.

If Wilkinson and Pickett think it is racist to say that there are a host of cultural and historical reasons why blacks tend to do worse than whites in the USA, then there are plenty of black community leaders and black politicians who are racist. No serious discussion of modern-day America can ignore the legacy of slavery and segregation, as well as the more subtle forms of ongoing discrimination which continue to hold African-Americans back. There is no doubt that these factors contribute to income inequality, but to say they are caused by inequality is highly questionable.

The reasons why the black homicide rate is much higher, and black life expectancy much lower, than the corresponding rates for white Americans are many and varied but in The Spirit Level, they are—as ever—reduced to symptoms of inequality. This will not do. Income, inequality and ethnicity are so closely intertwined in the United States that it is difficult to see where one stops and the other starts, but Saunders argues persuasively that inequality is not the main driver.

A significant clue lies in the pages of The Spirit Level itself. Wilkinson and Pickett note with surprise that they can find no association between inequality and poor mental health (p68-69) and then mention, almost in passing, that rates of mental illness are evenly distributed between different races. This should have been a Eureka moment but, as Saunders writes, “they fail to draw the obvious conclusion from their failure to find a relationship with inequality, which is that they only get state-level correlations with income inequality when there are underlying correlations with race to generate them” [emphasis in original]. Inequality is a symptom, not the cause.

This highlights one of the main problems with The Spirit Level. The myopic obsession with income inequality blinds the eye to the countless cultural, political, historical and demographic reasons why countries are as they are. Wilkinson and Pickett’s hypothesis requires one to believe that these are all rooted in inequality, but taking each in turn we can see how implausible that is. Their ‘theory of everything’, like all grand unifying theories, makes an extraordinary claim and requires extraordinary evidence. As was repeatedly demonstrated last week, the evidence provided in The Spirit Level withers under the light of serious scrutiny. Crying ‘racist’ will not make it more robust.

[This article was first published by the TPA on 16.07.10]


[Peter Saunders has responded to Wilkinson and Pickett's attack on his website]


[Tino Sanandaji—one of the authors of the WSJ article—makes some additional comments on his blog]

Thursday, 22 July 2010

Correlation coefficients and p-values

The original edition of The Spirit Level did not show correlation coefficients or p-values for any of the graphs, but when I wrote The Spirit Level Delusion, I included r-squared coefficients to show the strength of the associations.

Later editions of The Spirit Level include an appendix which shows Pearson Correlation Coefficients and two-sided p-values. As the Pearson coefficient shows a different figure than the r-squared, I have included the Pearson coefficients and the two-sided p-values for The Spirit Level Delusion graphs below for anybody who should want to compare them.

The first number is the correlation coefficient (the strength of the correlation). The figure in brackets is the p-value (the confidence that the correlation has not been caused by chance). Both numbers go from 0.00 to 1.00. For the first figure, the higher the number, the stronger the correlation. For the latter, lower numbers represent stronger associations. Statistical significance is usually represented by a p-value of at least 0.05, with 0.01 or less being preferred.

Life expectancy (UN 2004): -0.4515 (0.03)

Life expectancy (UN 2006): 0.16 (0.42)

Life expectancy (UN 2009): 0.22 (0.27)

Obesity: 0.01 (0.96)

Smoking: -0.31 (0.11)

Alcohol: -0.25 (0.21)

Infant mortality: 0.13 (0.51)

Trust: -0.29 (0.13)

Teen births: 0.32 (0.10)

Happiness: 0.05 (0.81)

Homicide: 0.30 (0.12)

Recycling/suicide: 0.64 (0.04)

Prison: 0.59 (<0.01)

Crime: -0.15 (0.45)

Victim of crime: -0.344 (0.21)

Recycling: -0.739 (<0.01)

Charity: 0.40 (0.25)

Foreign aid: 0.527 (0.01)

Single parents: -0.06 (0.285)

Suicide: -0.475 (0.01)

Unemployment: 0.033 (0.88)

Community life: 0.54 (0.03)

Alcohol/divorce: 0.35 (0.12)

Education: -0.154 (0.44)

Quality of life: 0.13 (0.51)

Homicide/suicide: -0.03 (0.86)

% GDP in tax/inequality: -0.65 (<0.01)

Happiness/income: 0.62 (<0.01)

Trust/income: 0.56 (<0.01)