Look at the changes, not at the levels

A few people have asked me recently for my view on The Spirit Level by Richard Wilkinson and Kate Pickett, which is apparently having some impact in policy circles. John Kay’s view in the FT comes closest to my own:

a larger source of irritation is the authors’ apparent belief that the application of regression methods to economic and social statistics is as novel to social science as it apparently is to medicine. The evidence presented in the book is mostly a series of scatter diagrams, with a regression line drawn through them. No data is provided on the estimated equations, or on relevant statistical tests. If you remove the bold lines from the diagram, the pattern of points mostly looks random, and the data dominated by a few outliers.

The United States, the most unequal of the countries considered, scores poorly on virtually all the social indicators used. Japan, rated one of the most equal, has long life expectancy, a small prison population and low levels of violence. Within Europe the Scandinavian countries are generally distinguished by high levels of both equality and social performance. These observations probably account for most of Wilkinson and Pickett’s findings.

I’m about as anti-inequality an economist as you’ll find. But my own empirical work on the issue has convinced me that when you look at within-country changes, the picture that emerges is very different to what you see when you look at a snapshot across countries over time. For example, it’s certainly true that in unequal countries, lifespans are shorter and infant mortality is higher. But here’s what you get if you compare changes in inequality with changes in mortality (from a paper with Tim Smeeding and Christopher Jencks).

clip_image002[5]

Note: Changes in infant mortality are scaled in reverse, to allow comparability with life expectancy measures. LE is life expectancy at birth (years). IM is infant mortality (deaths per 1000). All changes are expressed on a ‘per decade’ basis (ie. annualized and multiplied by 10). We exclude the three poorest countries in the OECD (Mexico and Poland; and Turkey for lack of data), and the richest (Luxembourg). Countries and years covered are Australia (AUS) 1981-2001, Belgium (BEL) 1985-2000, Canada (CAN) 1981-2000, France (FRA) 1981-2000, Germany (DEU) 1981-2000, Italy (ITA) 1986-2000, Netherlands (NLD) 1983-99, Norway (NOR) 1979-2000, Spain (ESP) 1980-2000, Sweden (SWE) 1981-2000, Switzerland (CHE) 1982-2000, the United Kingdom (GBR) 1979-99, and the United States (USA) 1979-2000.

Yup, the graph slopes up. In other words, countries that experienced big increases in inequality saw bigger improvements in health than those where inequality stayed stable or fell. In most cases, the effect isn’t significant, but the data certainly don’t support the hypothesis that rising inequality harms population health.

From a policy standpoint, specifications based on changes must surely be more compelling than specifications based on cross-country snapshots at a point in time. Australia can never literally become the Netherlands, but we can see what happens when our level of inequality rises and theirs falls.

After working on inequality and mortality, I have had similar experiences in looking at data on inequality and savings with Alberto Posso (we find no relationship), and in looking at inequality and growth with Dan Andrews and Christopher Jencks (we find that inequality has no impact on growth over the period 1905-2000, and conclude that inequality is good for growth over the period 1960-2000). In both cases, I had begun the project secretly hoping to find that inequality was bad, and wound up reluctantly reporting no such thing.

All this has made me think that the ‘instrumental’ reasons for worrying about inequality tend to be pretty flimsy, and that the best reason to care about inequality is the declining marginal utility of income (a dollar brings less happiness to a millionaire than a pauper). Inequality matters, but let’s not overegg the pudding.

5 thoughts on “Look at the changes, not at the levels”

  1. “The United States, the most unequal of the countries considered, scores poorly on virtually all the social indicators used. Japan, rated one of the most equal, has long life expectancy, a small prison population and low levels of violence. Within Europe the Scandinavian countries are generally distinguished by high levels of both equality and social performance. These observations probably account for most of Wilkinson and Pickett’s findings.”

    Well, not exactly. The US is not the only anglo country to have high inequality and poor performanc on a range of social indicators. The UK performs pretty poorly as well.

    Can I surmise then that you think that the relationship in levels (all those countries can’t be outliers – together they make up nearly half of the OECD) is spurious in that there are likely other omitted variables that explain both the better social outcomes that are displayed, on average in more equal countries?

    That might be true, but you have to be careful in relying too much on the changes data. It could also be that the process of distributional changes itself has consequences that are bad for social outcomes.  For example, Banerjee and Dufflo show that non-linearities are very important and that large changes in inequality in either direction are bad for growth. That suggests that complicated political economy issues are at play and I can’t see why the same couldn’t hold for social indicators, in addition to growth.

    Like

  2. Thanks LO – important points.
    “you have to be careful in relying too much on the changes data. It could also be that the process of distributional changes itself has consequences that are bad for social outcomes.”
    But isn’t this policy-relevant? If pro-equality policies don’t actually deliver the goods, surely we want to know about it. Put another way, Sweden may have better indicators than Australia. But if the evidence suggests that an attempt to Swedify Australia would do more harm than good, then this is the policy punchline, right?
    “Banerjee and Dufflo show that non-linearities are very important and that large changes in inequality in either direction are bad for growth.”
    Andrews, Jencks and I don’t find support for this in our top incomes & growth paper, but that could be an artefact of the data. However, I find BD puzzling from theoretical grounds too, as it implies that fluctuations in inequality will eventually cause growth to grind to a halt.

    Like

  3. I have often thought that the reason for this effect (level correlation, no change correlation) might be that the inequality of one’s society has the most effect in infant and formative years, since it influences early diet, education, stress, stress on the parents, etc.  If that was the case, then a change in inequality would not show unambiguously positive effects until enough years had passed for kids born in the more equal era to be impacting the life expectancy, crime, unemployment, smoking, etc statistics. A 15-30 year time lag. Has this idea been tested anywhere?

    Like

Comments are closed.