Wednesday, 19 April 2017

The Spirit Level: A Rebuttal - Part IV

This is part IV of IV of a rebuttal/review of 'The Spirit Level' by Richard Wilkinson and Kate Pickett. If you haven't already read them here are Parts I, II, and III.

Further Thoughts

On Statistics & Correlations

It really should not need to be pointed out, but it does – because it seems constantly to be forgotten – that correlation does not equal causation.  The most famous example of this is the fact that there is a strong positive correlation between ice-cream consumption and drowning accidents.  As explained excellently by Fallacy Man at The Logic of Science:

“Both of them increase together and decrease together, yet it would clearly be absurd to claim that ice cream sales are causing people to drown.  The reality is that both ice cream sales and drowning accidents are caused by a third variable: time of year.  People consume more ice cream in the summer than in the winter and people spend more time in the water in the summer than in the winter (which leads to more drowning accidents).  Thus, these factors are correlated, but not causally related.”

This is why the correlation between increased income inequality and increased ‘bad thing’ is not necessarily evidence that income inequality causes said ‘bad thing’ (or vice-versa).

Wilkinson and Pickett seem to have lost sight of this and spend most of the book citing example after example of correlations between income inequality and ‘bad thing’ x, y and z.  Yet, they never actually describe a plausible causal mechanism which is driving these relationships, or rule out that the causality runs the other way, or rule out explanations based on various other factors which could be driving both income inequality and various health and social outcomes, or rule out their findings being the result of statistical quirks (despite insisting that they have, they present no evidence of such in their 300+ page book).

On The Spirit Level Delusion Blog, Christopher Snowdon plots a couple of graphs, the first showing belief in God (people responding “God is very important in my life”) versus inequality and the second depicting cinema attendance versus inequality in developed countries.

The point of this is that both of these graphs show strong correlations, in fact, as Snowdon observes, the correlations he finds here are stronger than many of the correlations cited by Wilkinson and Pickett in ‘The Spirit Level’ in support of their thesis that inequality causes all manner of ills.

As Snowdon points out with these examples, it would be extremely hasty to jump from these apparent correlations to the conclusion that religiosity or going to the cinema causes inequality or vice-versa.

As Snowdon explains in his own words:

“...even if you exclude chance as a possibility (and that would be very hasty), explaining them in terms of income inequality requires a vivid imagination and a near-obsession with wealth redistribution.

As ever, our willingness to accept statistical associations depends on our susceptibility to the underlying message.  Perhaps a socialist atheist would find these associations compelling.  Or maybe a Christian film buff could use them as an argument in favour of capitalism.  The rest of us might shrug our shoulders and say 'so what?'”

Economist Tim Harford, writing on his blog, recounts a cautionary tale in statistics from econometrician Sir David Hendry:

“In 1980, the econometrician David Hendry investigated a key economic question: what causes inflation?  Hendry looked to the data for insight.  He speculated that a particular variable, X, was largely responsible.  He assembled data on variable X, performed a few deft mathematical tweaks and compared his transformed X with the path of consumer prices in the UK.  Graphing the result showed an astonishingly close fit.

The only snag: X was cumulative rainfall.  Since consumer prices and cumulative rainfall both rise over time, Hendry had an excellent platform for finding his spurious correlation.  Statistical sleight of hand did the rest.

Hendry wanted to demonstrate just how easy it was to produce plausible nonsense by misusing the tools of statistics.  ‘It is meaningless to talk about “confirming” theories when spurious results are so easily obtained,’ he wrote.”

On Normative Sociology

Joseph Heath, professor of philosophy at the University of Toronto, discusses the problem of what he calls ‘normative sociology’ on the ‘In Due Course’ blog.

Professor Heath defines ‘normative sociology’ as the pernicious habit to study what we would like the causes of a problem to be to the neglect of the actual causes, with reference to an offhand joke by Robert Nozick in ‘Anarchy, State and Utopia’.  Here’s Heath’s explanation:

“When this goes uncorrected, you can get the phenomenon of ‘politically correct’ explanations for various social problems – where there’s no hard evidence that A actually causes B, but where people, for one reason or another, think that A ought to be the explanation for B.  This can lead to a situation in which denying that A is the cause of B becomes morally stigmatized, and so people affirm the connection primarily because they feel obliged to, not because they’ve been persuaded by any evidence.”

This is essentially the converse of David Hume’s is-ought fallacy.  Here people are reasoning what is the cause from what they think ought to be the cause.

Heath uses racism as an example of this ‘politically correct’ behaviour:

“I routinely hear extraordinary causal powers being ascribed to ‘racism’ – claims that far outstrip available evidence.  Some of these claims may well be true, but there is a clear moral stigma associated with questioning the causal connection being posited – which is perverse, since the question of what causes what should be a purely empirical one.  Questioning the connection, however, is likely to attract charges of seeking to ‘minimize racism.’  (Indeed, many people, just reading the previous two sentences, will already be thinking to themselves ‘Oh my God, this guy is seeking to minimize racism.’)  There also seems to be a sense that, because racism is an incredibly bad thing, it must also cause a lot of other bad things.  But what is at work here is basically an intuition about how the moral order is organized, not one about the causal order.  It’s always possible for something to be extremely bad (intrinsically, as it were), or extremely common, and yet causally not all that significant.”

He goes on to further break this down into four main variants:
  1. Wanting a policy lever
  2. Worrying about “blaming the victim”
  3. Picking one side of a correlation
  4. Metaphysical views

I won’t get into the details of the first two variants, but the latter two are of particular relevance here.  On ‘picking one side of a correlation’, Heath observes:

“Statistical analysis often reveals a correlation between two things, but as we all know, correlation does not imply causation.  If A tends to go hand-in-hand with B, it could be that 1) A causes B, or 2) B causes A, or 3) A and B are mutually reinforcing, or 4) there is some third thing, C, that causes both A and B.  It is, however, very very common for statistical correlations to be reported as causal ones.  Because this sort of sloppy thinking happens all the time, it’s not so difficult for people who would like to believe that A causes B to respond to evidence of correlation between the two as confirmation of their view.”
Quite.  I would add that there is a fifth explanation of any apparent correlation between A and B: chance.  If you look at enough permutations of enough variables, you’re bound to find some that appear to be correlated as a result of pure chance.  I think Heath implicitly acknowledges this with his use of the phrase ‘...tends to go hand-in hand…’, suggesting that chance can be ruled out if the same correlation appears repeatedly in different settings or across long periods of time.

On the ‘metaphysical views’ point, Professor Heath claims [emphasis added]:

“...often there is a sense that the moral awfulness of some action or episode requires that it have enormous consequences.  This can easily lead to the view that anyone who denies the causal effects is in some way minimizing or downplaying the moral awfulness.”

“A good example of this… involves attitudes towards rising inequality.  Many people think this is very bad.  And yet, there is also a desire to believe that, if it is very bad, then it must also cause a lot of other bad things.  (Richard Wilkinson and Kate Pickett’s book The Spirit Level is an example of this tendency, as is Joseph Stiglitz’s The Price of Inequality.)”

“And yet, anyone who denies that inequality has these effects is liable to stand accused of seeking to makes excuses for it...”

On Inequality & Absolute Wellbeing

Writing for the Adam Smith Institute, Ben Southwood suggests that we shouldn’t care about income inequality at all.

Referring to an earlier post by Sam Bowman, Southwood notes the fact that in studies, people have been shown to have extremely poor knowledge about the actual levels and extent of income inequality in their societies.  Given a choice of five possible income distributions, survey respondents barely do better than random chance at selecting the one which most accurately depicts the society in which they live.  There is an enormous disconnect between reality and the public's perceptions when it comes to income distributions and inequality of incomes.

In an earlier blog post, Southwood considers the roles of luck and education on success.  For anyone familiar with the literature on this topic it should come as no surprise that investment in education, once you control for selection effects, does not pay the dividends that many seem to think it will.  This should serve as a stark reminder that luck plays a role, causality is rarely linear or even obvious and that the desire for an effective policy lever does not mean that one actually exists.

Warren Meyer (AKA Coyote), discussing inequality and absolute well-being on his blog, compares Denmark and Sweden (countries which are often pointed to by Democrats as ones to be emulated in terms of their redistributive policies) to the US in terms of disposable household income by income decile, using data from the LIS Cross-national Data Center and finds that:

“...all the way down to at least the 10th percentile poorest people, the poor in the US are as well or better off than the poor in Denmark and Sweden.  And everyone else, including those at the 20th and 25th percentile we would still likely call ‘poor’, are way better off in the US.”

Or, as he summarises it:

“...while the middle class in the US is richer than the middle class in Denmark, and the rich in the US are richer than the rich in Denmark, the poor in the US are not poorer than those in Denmark.

And isn't this what we really care about?  The absolute well-being of the poor?”

He concludes that :

All this talk about reducing income inequality by emulating Denmark is thus not about making the poor better off, but just about cutting the rich and middle class down to size.”

The Spirit Level Fallacy

The final chapter of ‘The Spirit Level Delusion’ Christopher Snowdon titles ‘The Spirit Level Fallacy’.  In it, he argues that “...The Spirit Level uses the statistical methods of epidemiology...”, many of which, he points out, “...were developed by the botanist and statistician Ronald Aylmer Fisher in the 1920s for the purpose of studying plants.”.

Snowdon points out that these methods were originally applied to randomised, controlled experiments, where a single variable (such as the application of fertiliser) could be altered between two otherwise identical groups of samples.  Things inevitably become more complicated when trying to study groups of people, which are more complex, and differ in many more ways that cannot be controlled by the experimenter.  This is taken to the extreme when comparing data for entire nations:

“Comparing aggregate data from whole countries is as far from the randomised controlled double-blind trial as it is possible to get under the mantle of science.  Neither random nor controlled, ecological studies (ie. comparing whole nations) allow unlimited scope for interpretation.  The researcher can focus on any variable over any period of time amongst infinitely varied populations.  The pitfalls of using international data to make broad assumptions is so well known that it even has its own term—the ‘ecological fallacy’.”

Snowdon sums it up as:

“The Spirit Level relies on the conceit that wealthy countries are basically the same, with inequality being the main difference between them.  Consequently any disparity in social problems must be the result of inequality.  This is faulty logic and a classic ecological fallacy.”

He continues:

“It takes a peculiarly blinkered view of the world to portray Sweden, a country which has not fought a war since 1814, as being fundamentally the same as Israel, except in its distribution of wealth.  Or to equate Spain and Portugal, both fascist dictatorships until the 1970s, with Belgium.  Or to imagine that a culturally homogenous, traditional Asian country like Japan can only be distinguished from the United States by reference to the gap between the richest and poorest 20% of the population.”

“Hong Kong and Singapore could certainly be cited as proof that economic freedom leads to greater inequality.  What cannot be shown is that this inequality has led to ill-health or more social problems.  If anything, the reverse is true.
The success of Hong Kong and Singapore in no way proves that inequality makes things better.  That would be to replace one fallacious argument with another.”


In the final chapter of ‘The Spirit Level Delusion’, under the subtitle of ‘Only Connect’, Christopher Snowdon observes that there is a strong correlation between latitude of a country’s capital city and the education attainment of its 15 year olds.  He wryly remarks “Meaningless though this correlation is, it cannot be explained by chance.”

Snowdon sums the whole thing up better than I could, so I shall borrow some of his words here:

“In truth, one can find associations anywhere.”

“...these statistical associations—strong though they are—tell us no more about the world we live in than the graphs that could be produced that show that foreign aid ‘causes’ theft or that recycling ‘causes’ suicide...”

“The single-minded focus on income inequality requires us to ignore countless differences between countries comprising hundreds of millions of people.  If we can overcome that hurdle of logic, we must take another leap of faith and believe that the widening gap between rich and poor is the single most important development of the last forty years.  Seismic changes in society—the decline of religion, women's liberation, the sexual revolution, multiculturalism, individualism—must be ignored in favour of a theory of everything that rests on the populations psychological response to inequality.”

“People are unequal in many ways, of which income is just one.  The inequality theorists never tell us what is so special about income that makes its uneven distribution so uniquely dangerous that society needs remaking in order to afford protection.”

“If we must compete for status, there are worse ways to do so than by comparing the fruits of our labour.  For most of human history, status has been earned through violence, promiscuity and oppression.”

“Even if, through a miracle of socialism, you are able to prevent people from using wealth as a barometer of success, they will find other ways—perhaps more harmful ways—of asserting their position in society.”

“Driven by a hunger for improved circumstance, society has reached unparalleled heights of happiness, health and prosperity.  Pessimists and misanthropes insist that this long road of progress has come to an end.  They always have.  But given the choice between chasing happiness on and endless road or having our desires curbed and our horizons lowered, most of us will tolerate a little status anxiety for the freedom to keep on moving.”

No comments:

Post a Comment

Comments are moderated. Abusive and/or hateful comments will not be permitted.

The blog author accepts no responsibility for the content of reader's comments.