Sunday, January 2, 2011

Nicholas Kristof makes a fool of the New York Times

Nicholas Kristof has given a raving review of the book "The Spirit Level" in the NYT. He is clearly unaware of the debate in Europe, where the "Spirit Level" has largely been debunked due to poor scientific quality.

I have written about this, and would like to remind you of 3 facts about the book "The Spirit Level" (links at the end):

1. The book is scientifically dishonest by never telling the readers that it represents a minority opinion within social sciences. Higher ranked researchers in better schools using more rigorous methods have come to the opposite conclusion about income inequality causing lower life expectancy. Yet the reader of the book, or readers of the New York Times, would never even know the other research existed.

Wilkinson and Picket give their trusting readers the impression they are reading novel research results. In fact, they are reading claims that were debunked years ago.

Let me refer to scientific authority, because that's what Kristof does. Princeton Professor Angus Deaton is currently the 57th highest ranked economist in the world. He is an expert in health economics, statistical methods and one of the foremost expert in determining causality. He is in often mentioned as in the running's for the Nobel prize. Deaton is a typical center-left academic, not some right-winger.

In 2004, Angus Deaton reviewed the research on income inequality and life expectancy for the Journal of Economic Literature, the most authorities such journal in economics. Deaton in particular looked at the empirical evidence for the claims of Wilkinson. He concluded:

"it is not true that income inequality itself is a major determinant of public health. There is no robust relationship between life expectancy and income inequality among the rich countries, and the correlation across the states and cities of the United States is almost certainly the result of something that is correlated with income inequality, but is not income inequality itself"."

Kristof repeats the claim in the New York Times that income inequality causes poor health outcomes in American states and cities. This particular hypothesis was looked at by Deaton and Lubotsky in 2003. They analyzed the link between income inequality and mortality, controlling for demographic differences. They found (from abstract):

"Conditional on the fraction black, neither city nor state mortality rates are correlated with income inequality"

2. Those nice convincing graphs in Spirit Level graphs are to a large extent based on manipulating and cherry-picking.

For instance, I, some blogger in his pajamas, forced Wilkinson to admit publicly in the Wall Street Journal that he had established no statistically significant relationship between life expectancy and income inequality across countries.

Most of their other claims could not be replicated either (although I didn't test everything). For instance, Wilkinson and Picket assert that they have proven that income inequality cases lower innovativeness. I was suspicious of their claim that the United States was one of the least innovative countries in the world, at the same technological level as Portugal.

It turns out the "distinguished" scientists who convinced Kristof by citing "mountains of data to support their argument" picked the innovation data from amateur internet site Nationmaster! Their claims was not supported using patent data from the World Intellectual Property Organization.

3. Wilkinson and Picket confuse correlation with causality.

Sure, they pay lip-service to causality. But they never tell the readers that among all the papers they cite, there is not one single article that has shown any causal effect of income inequality on health or social problems.

Every single article Wilkinson and Picket cite (and they are lots of them), is some version of a correlation studies, sometimes with few and sometimes with more controls, but never with identified causal links.

You don't believe me? It's true.

Wilkinson and Picket rely on 2 meta-studies to create the impression of massive external scientific support for their view. The big one, with close to 200 papers, was written by no other than themselves (!). You can read it here:

In the paper Wilkinson and Picket explain their criteria for determining if a paper is evidence for their hypothesis (inequality causes poor health).

“We classified them as wholly supportive if they reported only statistically significant associations between greater income inequality and poorer population health”

Did you catch the key word? "association". In terms of causality, "association" is the equivalent of correlation (although association doesn't have to be linear). Association is what researchers write when they have failed to establish causality.

This is how Wikipedia explains the term:

“In quantitative research, the term "association" is often used to emphasize that a relationship being discussed is not necessarily causal”

Wilkinson and Picket gather a bunch of articles (87 have a statistically significant relationship) where there is some form of correlation between inequality and health. But of course, poor health can cause inequality, and third factors (say, drug use or unemployment) can cause both inequality and poor health. Establishing causality here is not trivial, such as from smoking and mortality, its central to the debate.

In fact, it is hard to think about any social problem that doesn't depress income for those afflicted (and thus create income inequality) as a side effect. So typically what will happen is that there is a weak and not very robust correlation between inequality and poor social and health outcomes, which vanishes when researchers do a more careful analysis with more controls.

So proving a causal link between inequality and health/social outcomes is not a detail, it's essential to accepting their story.

But Wilkinson and Picket never establish causality. They never even claim to in the scientific version of their paper. There, facing a less gullible audience, they admit that they are not showing causal effects, just statistically significant "associations".

But in the book, and on the internet, and to their audiences, and to trusting suckers such as Nicholas Kristof, Wilkinson and Picket give the dishonest impression that the papers they are citing contain causal evidence.

Wilkinson and Picket are relying on the lack of scientific sophistication among the general public. The public doesn't always understand the difference between correlation and causality, and can obviously not be aware of the studies in economics that already debunked the claims in the Spirit Level. But isn't it the job of the New York Times to ensure scientific standards for their readers?

Whenever pushed into a corner, Wilkinson and Picket rely on the number of citations in their book. However the number of studies happens to be irrelevant when the problem is reverse causality. 20 or 200 or 2000 correlation studies of the same spurious relationship will give the same answer. If you have failed to identify a causal link, all the 200 correlations are equality worthless.

The New York Times is allowing Nicholas Kristof to embarrass their newspaper. Don't simply trust everything Wilkinson and Picket claim because you dislike inequality (so do I) and are easily impressed by academics (I used to, but than I grew up). Academics sometimes fall to the temptation to abuse their authority and our trust to package their opinion as science or to sell books. Do your homework.

From the start I noticed that these two are dishonest people, which I don't like. I have thus gone after them aggressively. Since the book is such a house of cards, it was not hard.

No comments:

Post a Comment