In Woody Allen’s cult classic Sleeper, Allen’s character Miles Monroe is cryogenically frozen, then defrosted 200 years later. The doctors of the future who examine Miles are tickled by his quaint, 20th-century superstitions about diet and health:
Dr. Melik: This morning for breakfast he requested something called “wheat germ, organic honey, and tiger’s milk.”
Dr. Aragon: [chuckling] Oh, yes. Those are the charmed substances that some years ago were thought to contain life-preserving properties.
Dr. Melik: You mean there was no deep fat? No steak or cream pies or … hot fudge?
Dr. Aragon: Those were thought to be unhealthy … precisely the opposite of what we now know to be true.
Dr. Melik: Incredible
It’s been more than 40 years since Allen’s movie premiered, but his satire of public health research sadly still resonates. A widely circulated New York Times blog post reported this week on a study purporting to show that people who run at least five minutes a day live around three years longer than those who don’t. This finding was determined to be true after “adjusting for” various characteristics of study subjects—their gender, whether they smoked, any family history of heart disease, and so forth.
The problem with this study—and the many related observational studies on what does or doesn’t make us live longer—is that healthy people are different in all sorts of ways from unhealthy ones. Some of the differences between runners and nonrunners can be accounted for, albeit somewhat imperfectly, by considering observable attributes like height, weight, age, etc. But inevitably there are differences that get left out of the analyses: runners might sleep longer; they might eat more almonds or blueberries; maybe they have less stressful work lives, which in turn facilitate a few minutes of exercise each day; maybe they are less depressed because they have shorter commutes; maybe their commutes involve more walking than driving. The list of other attributes and habits, any of which might contribute to a longer observed life span, extends to infinity. Is it running that accounts for the difference in longevity between runners and nonrunners? Or one of these other, unobserved differences? Who knows?
Social scientists take these concerns much more seriously than they did even just a few years ago. We live, for better or worse, by the mantra that correlation does not equal causation. As a result, we increasingly run randomized trials, the type of study that has been the gold standard in biomedical research for many decades. What difference does it make for a child to be in a class of 22 kids versus 15? Randomly assign one group of third graders in Tennessee to be in small classes and another group to larger ones. By definition, in such a study, there are no unobserved differences between the two groups, since assignment to one or the other was purely a matter of random chance. And if you can’t make assignments randomly yourself, you can search the world for a source of random variation that does the job for you. In a famous study on class size and student achievement, economists Joshua Angrist and Victor Lavy took advantage of the so-called “Maimonides Rule,” which limited Israeli classrooms to 40 students. By law, the arrival of the 41st fifth grader automatically triggered a cut in class size, by half. If you assume that schools with 40 fifth graders are basically the same as schools that have 41 (thus causing a split), you have yourself a “natural experiment” in which you can observe whether cutting class size in half has an effect. (It does, at least for fifth graders.)
Unfortunately, social scientists who study health care still lag well behind their counterparts in medicine in their use of randomized studies. For example, according to Amy Finkelstein, an MIT economist who advocates for the increased use of randomized trials in this area, less than 20 percent of U.S. health care delivery studies (studies that look at, for example, efforts to encourage patients to take their meds or to reduce re-admissions to hospitals) are randomized. By contrast, about 80 percent of studies of U.S. medical interventions (like trials of drug and medical devices) are randomized.
It’s true that there’s never been a decades-long shortage of almonds or blueberries, which might have helped us study what happens when our consumption changes as a result of an externally imposed condition. And it’s hard to force people to exercise or to force a control group to be sedentary so you can run proper experiments on whether running improves longevity. But it’s not impossible—future researchers could randomly provide economic incentives to exercise and use this “encouragement design” to study the impact of running.
So where does this leave us? It’s surely not a bad idea to spend a few minutes a day on the treadmill: There is information in correlations like the one reported in the Times. While you’re at it, go ahead and drink some red wine and eat a bit of chocolate. It probably won’t hurt you, and maybe it’ll help you live a bit longer. Maybe.
But the best advice on a balanced life might come from another cult classic. At the end of Monty Python’s The Meaning of Life, the narrator is handed an envelope containing the following directive: “Try and be nice to people, avoid eating fat, read a good book every now and then, get some walking in, and try and live together in peace and harmony with people of all creeds and nations.” That’s sage counsel, at least until we read in next week’s Times that fatty steaks and cream pies are, in fact, the key to immortality.