Causation and correlation again

A recurring theme in this blog is that correlation is not causation. Yet it's easy, and tempting, to make the mental leap from correlation to causation without sufficient information. Today's example is this column from Lindsay Abrams in The Atlantic. Titled "When Trees Die, People Die" the column describes a study that appears in the American Journal of Preventive Medicine about the relationship between trees and human health.

Here's how Abrams describes the study:

Within four years of first becoming infested, the ash trees die -- over 100 million since the plague began. In some cases, their death has an immediate impact, as they fall on cars, houses, and people. In the long term, their disappearance means parks and neighborhoods, once tree-lined, are now bare.

Something else, less readily apparent, may have happened as well. When the U.S. Forest Service looked at mortality rates in counties affected by the emerald ash borer, they found increased mortality rates. Specifically, more people were dying of cardiovascular and lower respiratory tract illness -- the first and third most common causes of death in the U.S. As the infestation took over in each of these places, the connection to poor health strengthened.

Note the use of the word "may" in the first sentence of the second paragraph, and the conclusion in the last sentence. But the post overstates the conclusions of the study, which is available here. The authors, having surveyed the literature, conclude that their study strengthens the correlation, but reiterate that they can conclude nothing about causation.

It's the kind of thought-provoking study that should inspire more studies. Let's hope it does.

High fructose corn syrup, diet, and studies

There's been a lot of news coverage in the past couple of days about a study, "High fructose corn syrup and diabetes prevalence: A global perspective" by Michael I. Goran and others published this month in the journal Global Public Health. See, for example, here and here. (The NY Times has posted a .pdf of the article.) The paper itself is worth reading for several reasons.

• It's really interesting. Across the globe, the number of people with diabetes is increasing, from 153 million in 1980 to 347 million in 2008. Most of the increase is coming as "Western-style" diets, ie those with lots of processed foods, carbohydrates, and especially sugar, become more widely consumed. And, as the report says:

  A growing body of evidence supports the hypothesis that in addition to overall sugar intake, fructose is especially detrimental to metabolic health and risk for type 2 diabetes. This is of particular concern given the global changes that are occurring in the use of high fructose corn syrup (HFCS) in food and beverage production . . . (citations omitted)

  The data table shows that, while not all the higher-diabetes countries consume HFCS, all the countries that consume HFCS have a diabetes prevalence that is higher - 20% higher - than countries that do not use HFCS. Note that this is an ecological study, looking at populations, not at individuals, and does not infer causation.
  
• It suggests a reason why increased consumption of HFCS can contribute to an increase in rates of diabetes, even though HFCS does not depend on insulin. Instead, HFCS is metabolized by the liver, and does not generate leptin (which makes you feel full). In addition, there is some evidence that fructose helps generate fat, particularly the bad fat around your waste. If you're heavier, you're more likely to develop diabetes. In fact, diabetes-management sites like this one recommend keeping an eye on how much HFCS you consume.
• Nonetheless, it's important to remember that this study describes correlation, not causation. The New York Times was a little incautious when it quoted Marion Nestle as saying that the study's conclusion was "a stretch." She seems to think so too, and this discussion on her blog is worth reading. Some highlights:

As with all correlational studies, something else could be going on that causes HFCS, sugars of all types, and diabetes to increase.

 And, later on:

Yes, HFCS is sugar(s)—glucose and fructose.  So is table sugar (sucrose).

But the bottom line goes for both: Everyone would be better off eating less sugar(s).

Correlation is not causation

It's kind of in the category of 'Man Bites Dog' but yet again a New York Times headline writer has confused correlation and causation. The headline reads: "Want Your Union to Last? Marry in New Jersey". Kind of grabs your attention. Except for this problem: people who are older when they marry and have more education are less likely to divorce. More of those people live in New Jersey and other northeastern states, hence the lower divorce rates. 

The story itself gets the issue right. The article is illustrated by a nice graphic - that's the screen shot above.