Sunday 16 May 2010

US states

[This is a discussion of the issues mentioned on page 15 of The Spirit Level Delusion]

The Spirit Level Delusion focuses on differences between nations. It does not look at differences between US states. There are several reasons for this. 

Firstly, since—as I show—inequality does not influence outcomes on an international level, it is highly improbable that it would influence outcomes between regions within a country. Although the USA is a diverse place, there are greater cultural differences between, say, Israel and Japan than there are between Texas and California. Perhaps this explains why, even by the standards of The Spirit Level, the correlations shown for US states are weak. It is a stretch to say that the graphs they show on, for example, pages 59 and 141 are indicative of a genuine causal relationship.

Secondly, a discussion of US states would make the book twice as long and, for the reasons above, this would be an unnecessary strain on both reader and author.

Thirdly, and most importantly, the health and social problems highlighted in the USA are more plausibly associated with absolute income than with inequality. Wilkinson and Pickett deny this, but their own evidence contradicts them. On page 22 of The Spirit Level, they show two graphs. The first shows their index of health and social problems against inequality, the second shows the same index against 'national income' (this must be a mistake—they surely mean 'state income'). These data are shown below.

Wilkinson and Pickett state that the second of these graphs shows no association between income and health and social problems, but even the casual reader can see that this is not true. Clearly, the worst afflicted states are the poorest. Wilkinson and Pickett divert the reader's attention from this fact by not showing the linear regression line. Had they done so, the graph would have looked like this...

There is a reasonably strong correlation here. The poorest states do worse and none of the richer states do badly. It is, however, worth asking why the correlation is not even stronger. The answer lies in the dozen states in the bottom left corner. These states do not show a close fit, but that won't concern the people who live there. All these states are doing better than would be expected from their income. This is no great surprise. These are sparsely populated, predominantly rural, ethnically homogenous Northern states like North Dakota, Wyoming and Montana. No one pretends that wealth is the only factor that makes a state or country successful, but as this graph shows, it certainly helps. Look at the top right-hand corner. Those are the states that have higher state incomes but perform badly. Don't see any? Exactly. 

Connecticut, Massachusetts and New York hold the key to understanding this data. They are outliers in Wilkinson and Pickett's inequality graph (above) and consistently fail to follow the pattern in The Spirit Level's other graphs. How three very unequal states do so well is an important question which Wilkinson and Pickett never address, but the answer is simple. They do well because of their wealth—inequality is irrelevant. Massachusetts performs so well because it is the third richest state. Connecticut and New York are outliers on the inequality graph but are a perfect fit on the income graph. 

Mississippi and Louisiana perform very poorly under most criteria. They have the lowest life expectancies and the highest rate of infant mortality of any US states. But they are also the poorest and third poorest states respectively. This, more than inequality, must be considered the real driver of outcomes in these states. 

We cannot rule out some indirect effect from inequality, but any effect is far from proven. The apparent correlation between inequality and outcomes is not, in itself, reason enough to cry causation. Other factors, such as the proportion of African-Americans in the population (shown below), show a stronger correlation but here, too, it would be foolish to insist on a direct causal relationship. 

Many other such graphs could be created showing strong correlations—some valid, some spurious—and that is only to be expected. Outcomes in societies are dependent on countless cultural, political, economic and demographic factors that interweave over many years. This is an obvious point to make, but in the light of The Spirit Level's monocausal 'theory of everything', it is one that needs to be reasserted.

No comments: