The New York Times health section is making it too easy today. Let's start with the final paragraph of this article to lay out the conclusion:
“They show quite convincingly, yet again, that overweight and, in particular, obesity, raise the risk of mortality,’’ Dr. Stampfer said. “It really should be the final word on this issue that’s arisen as to whether overweight is actually bad for you or not.”
Strong words, eh? Too bad he can't back them up. Here's the main claim being made regarding the modestly overweight:
The researchers said the more telling analysis arose when they focused on 186,000 healthy men and women who had never smoked. Among men and women, being overweight raised the risk of death 20 percent to 40 percent compared with normal-weight people, the researchers said.
Sounds scary. Let's examine some of the methodological issues so that we can aspire to the correct level of scared-ness. First, the population is drawn from those in the AARP. As the article acknowledges:
Because all the participants were recruited from AARP, they are not exactly representative of the population as a whole, and the participants reported their weight themselves, which contributes to some uncertainty in the results.
Kudos for the honesty.
Dr. Leitzmann said neither factor would probably skew the results considerably because “it’s a very large sample of people.”
Statistically the word skew has a specific meaning and the more correct term would be bias. Also, the good doctor is full of it. A large, non-representative sample is still non-representative. One possible advantage is that with a large enough sample you could select a random subsample that is representative of whatever population you're trying to capture (in this case the whole US). The article does not say whether they did this or not.
Secondly, the size of the sample reduces uncertainty but it does not reduce bias. In this case it means that peoples self-reported weight has been measured very accurately. It does not mean that their actual weight has been measured accurately at all.
Errors in reporting weights — people sometimes say they weigh less than they actually do — would also not produce a large effect, he added.
At least he acknowledges the potential bias, but, as I said above, it is not possible to know if the results are truely accurate or not.
Additionally, the mechanism for categorizing people as overweight is the body-mass index (BMI). The BMI was invented over 150 years ago. Its purpose is to categorize those with average body composition. Anyone with abnormal body composition, will not be categorized correctly. As an example, I am just below the overweight cut-off with my BMI, even though I am in excellent health. The disparity is due to my lifting weights and the subsequently higher muscular composition of my body.
“No single study is able to solve a controversy of this magnitude,” Dr. Leitzmann said, but he recommended that anyone overweight “should be looking to lose weight.”
I have to respect his honesty in his conclusion here. He may be reaching a bit, but he at least acknowledges that his study is not the end-all be-all of all research. Unlike Dr. Stampfer, who was not involved in the study, but throws out the concluding quote found above. This study suggests the possibility that being slighly overweight increases mortality, but it does not even come close to being the "final word" on the subject.
The study is an example of an observational study, in which a sample of existing data is analyzed for correlations. These types of studies are far inferior to randomly selecting the sample and assigning people to one category or the other. Obviously this is not possible to examine this problem; it is not realistic to randomly assign certain people to be overweight or normal weight. That means that there will be no single study, no matter how large, that will definitively answer the question.