from Merijn Knibbe
Inequality considered: what do agent based macro data on wealth and health tell us about the consequences of inequality?
Inequality is back into the limelight. Of course, we already knew that, especially in Anglo Saxon countries, neo liberal policies had caused an increase in inequality (just consult consecutive versions of the UN World Development reports). Piketty and Saez (2003, 2007, 2010 ) showed that, in a long term perspective, the increase in the USA was extraordinary and also confined to the very top. Also, Pickett and Wilkinson (2009) showed that the consequences of inequality (in fact: the consequences of the processes that lead to high inequality) are detrimental: unequal countries as well as unequal states of the USA tend to have high scores on indicators of social as well as individual pathology. Both findings have been challenged. Among others, Gordon pleads that the findings of Pikkety and Saez are not as bad as they look, though he does not challenge the findings themselves. Saunders (2010) pleads, however, that the findings of Picket and Wilkinson are compromised to the core. In another blog on this site, I’ve given some short comments on Gordon, basically stating that, in a National Accounting perspective and when you want to know what money will buy, you have to deflate income and consumption with the consumption deflator and not with the GDP deflator. Here, I will consider the plea of Saunders: are the findings of Picket and Wilkinson compromised to the core?
2. Initially, Saunders does a good job
Saunders political aims are clear. He tries to construct and publish evidence based oninformation which enhances the conservative cause, which sees high incomes as a reward for hard work and taking risk. I like that: his affiliations are clear (below, you will find my affiliations). As far as this means that he has to trash simple minded ‘progressive’ statements and data, it’s also a good thing. And he does, when it comes to the evidence, a good job at this: he updates the data set of Picket and Wilkinson, adds 22 new countries to the data set and throws in new variables (in the USA especially the ‘race’ thing). His dataset is better and more up to date than the dataset of Pickett and Wilkinson. Eleven middle income ‘transition economy’ countries as well as ten South American, Asian and African countries are added. The modus income of the old data set was about 35.000 PPP dollars, the modus of the countries which he ads is about 16.000,– (his figure 3). This does give a different perspective
3. I’m not completely at ease with the way Saunders uses data – but he does score some points
Saunders goes on by convincingly showing that if you remove the ‘outliers’ USA (high inequality, highest social pathology) as well as the Scandinavian countries (lowest inequality, low social pathology) from the initial 22 countries data set, the remaining differences are smaller (somehow, however, that does not strike me as a complete surprise). He also shows that differences between these (all quite rich) outliers are indeed huge – you definitely need not to have high inequality to become a rich nation! He uses boxplots to identify outliers, but if there is a positive relationship between variables this is not an ideal method. Boxplots measure differences between variables and the mean – and the most ‘leftward’ and ‘rightward’ variables are always bound to be rather far from the mean! Outliers should be defined as ‘far from the regression line excluding the outliers’ and not ‘far from the median’. Also, as regression lines are, by definition, straight and as real relations may be convex or concave as well as ‘fuzzy’ (a lot of ‘white noise’ or ‘real life’), we should (but that’s surely not restricted to this study) not be to too eager to dismiss so called outliers. Sideline: in fact outliers are often the most interesting cases! An additional point: dismissing the USA from the dataset as an outlier does give a somewhat awkward feeling: its 300 million people does not compare badly with the number of inhabitants of all other 21 countries combined.
Dismissing some countries from the data set and adding other countries with lower incomes and another culture results in correlation coefficients between inequality and social pathology indicators which are much lower than those found by Pickett and Wilkinson. Saunders does score a point here. Again and again, he states that social pathology is not only caused by inequality (or by the processes that cause inequality) but also by ‘cultural’ differences. He shows this for the USA by adding ‘southern states’ and ‘size of the black population’ as explaining variables to the regression equations. This improves the fit of the regressions. Aside from ‘maths and literacy scores’ (which are not influenced by the relative size of the black population in a state) all other social pathology variables are decisively influenced by the size of the (generally: poor) black population. To me, these findings are about the very definition of inequality! Saunders in fact does an excellent job to show that there is more to inequality than income – it is deeply engrained in the fabric of a society. But does he also prove that the statistical relation between income inequality and social pathology is bogus? No, he doesn’t, as he as well as Pickett and Wilkinson use the wrong method to (dis)prove this.
4. Inequality kills, in the Netherlands. What kinds of method should have been used to prove this internationally?
October 11th, the Dutch Centraal Bureau voor de Statistiek (CBS) published three studies on differences in income, wealth and health (Brakel en Knoops, 2010; Bruggink, 2010; Wingen, Berger-Van Sijl, Kunst and Otten, 2010). These studies are based on ‘agent based statistics’ on income and (perceived) pathology. The ‘individual tax number’ of Dutch citizens makes it possible to match different sets of individual data on incomes, wealth and health, which in turn makes it possible to compare ‘standardized household’ incomes with health and life expectancy. This is as good as it gets. Brakel and Knoops state that (my translation): “Internationally, no other studies comparing healthy life expectancy between people with a lower income and people with a higher income (household income, M.K.) are known”. The results of these studies are clear: inequality kills, inequality is sickening. People with an income below the poverty threshold (10.020 Euro) have a lower life expectancy (6 years) and a lower ‘healthy life expectancy’ (16 years, that is: sixteen years.). In the Netherlands, a reasonably equal country with a well organized health care system and excellent pre- and post natal care, which boasts the tallest people of the world (a sensitive indicator of health during childhood) people who are ‘not poor’ (i.e. with an income higher than 10.020,–) can expect sixteen healthy years more than poor people. And the differences get larger when we look at rich people: wealth, as well as health, as well as the combination of health and wealth have independent positive relationships with health and life expectancy. High income, high wealth people have much longer and very much healthier lives than low income, low wealth people. Methodological improvement no. 1: it’s not enough to look at income inequality, wealth inequality counts too, as well as the combination of wealth and income (to be sure: Saunders does mention that wealth inequality in the Scandinavian states is higher than in the U.K., which to me was a real surprise). Methodological improvement no 2: we do not only have to look at averages per country but also at differences within countries: are differences in, for instance, life expectancy or teenage births between the lowest and the highest quintile higher in the (unequal) USA or U.K. than in the (more equal) Netherlands or Japan? The reason to do this is of course the well known ‘omitted explaining variable’ problem. If there are omitted variables which influence the average level of life expectancy it is better to look at the differences between income groups (though there may also been omitted variables which influence differences). Just looking at averages might hide the real differences.
Where does this leave us?
The results of all these studies (including Saunders and Gordon) are clear. Inequality increased during the last decades – but surely not everywhere. (France is, according to Gordon, a clear exception). Inequality is, beyond any doubt, statistically connected with social pathologies like large differences in individual health, though we do have to take demographic, economic, geographical and cultural differences into account when we want to understand and explain these differences (nothing new here, in fact). Even in rather equal and well organized countries, there are large differences in life expectancy and even larger differences in health between income groups, not only when we compare the poor with the rich but even when we compare the poor with the ‘not poor’. Being poor is bad for you. But is it also, as Pickett and Wilkinson imply, also bad for the others? I think we indeed do need more precise data while we do have to take complexity into account. But Saunders clearly proofs that there are societies which can manage equality as well as low levels of social pathology and a high life expectancy. And can he mention one example of the opposite? The USA? Russia? Mexico? Israel? Portugal (lots of homicides, over there)? Also: has somebody already calculated the costs of sixteen more unhealthy years?
P.S. when preparing this blog I stumbled upon what is in all probability a Tinbergen article (CBS, 1946). In this article, income inequality in Denmark, the U.K, the USA, France and the Netherlands between (about) 1915 and (about) 1942 is measured with Pareto’s alpha (The ‘smell test’: yes, Tinbergen, who during World War II worked at the CBS). It’s not a new discussion. The results: Inequality decreased everywhere after WWI and again after about 1930. The 1920’s increase was specific to the USA. Underscoring Saunders ‘cultural’ explanation of inequality: between 1921 and 1939 Denmark was by far the most equal country of the lot. And now that I think of it: the ‘Saez approach’ to economics has a very distinct Tinbergen-smell to it.
Brakel, M. en Knoops, K., ‘Gezonde levensverwachting korter bij de lage inkomens’ in: CBS (2010) Bevolkingstrends, 2010-III 29-35.
Bruggink, J.W., ‘De verschillende dimensies van levensverwachting zonder lichamelijke beperkingen’ in: CBS (2010) Bevolkingstrends, 2010-III 36-42.
CBS (1946), ‘Pareto’s alpha en de inkomensverdeling in Denemarken, Frankrijk, Groot-Britannie, Nederland en de Verenigde Staten’, Statistische en Econmetrische onderzoekingen 2 . 55-67. Den Haag.
Gordon, R. J. (2009), Misperceptions about the maginitude and timing of changes in American income inequality, NBER working Paper No 15351.
Piketty, T., and E. Saez (2007), ‘Income inequality in the United States, 1913-2002’ in: Atkinson, A.B. and T. Pikkety, Top Incomes over the Twentieth Century. A Contrast between European and English-Speaking Countries (Oxford, 2007). See the Emmanuel Saez website for the 2003 article and the 2010 update of data.
Pickett. K. and R. Wilkinson (2009), The spirit level. Why equal societies almost always do better. Allen Lane.
Saunders, P. (2010), Beware false prophets. Equality, the Good Society and the Spirit Level. Policy exchange, London.
Wingen, M., M. Berger-Van Sijl, A. Kunst and F. Otten, ‘Inkomen en vermogen als indicatoren van gezondheidsverschillen’ in: CBS (2010) Bevolkingstrends 2010-III 43-49.