Running from the proof: correlation does not mean causation
A couple of days ago Gary Taubes, who was visiting family in Los Angeles, drove up to Santa Barbara, and he, MD and I got together for a long lunch. We talked about all the things we always discuss, most of which have nothing to do with nutrition or nutritional science. But, as always when we get together, talk did turn to science and the sorry state of nutritional science in the world today.
We discussed a Stanford study that was recently published in the Archives of Internal Medicine demonstrating that those who are runners live longer and have less disability. The paper proves absolutely nothing, yet an enormous number of people, many of whom should know better, profess that this study is the smoking gun that ties exercise to longevity and health.
The researchers sent questionnaires to 538 members of the 50+ Runners Association and healthy controls from the Stanford area who were 50 years old or older in 1984. By 2005, 284 runners and 156 controls had completed the 21-year follow-up. After crunching the data, the researchers found that more than double the number of controls had died as compared to the runners and that the total disability score for the controls was much higher (i.e., they had a greater degree of disability) than for the runners. The conclusion as stated by the authors:
Vigorous exercise (running) at middle and older ages is associated with reduced disability in later life and a notable survival advantage. [My italics]
The conclusion of this study, which is absolutely accurate, is that running is associated with longevity and reduced disability. The two are associated or correlated. The study does not prove that running increases longevity or decreases disability. Correlation is not causation. But you wouldn’t know that from the press coverage of this study.
Let’s explore this study a little because it provides a good example of what these kinds of studies show and what they don’t. And why we should never rely on association or correlation studies as proof of anything.
First, in the USA (and in other countries as well) it is a given that exercise is good for one. We believe that exercising will make us live longer. And a lot of us believe that exercise will make us fitter and have less disability as we age. When we see a study that seems to prove this, we accept it without batting an eye. Even though the study shows no such thing. We are culturally primed to accept scanty evidence to prove what we already believe about exercise.
Let’s assume that our cultural dogma was that exercise was unhealthy and that those who did vigorous exercise were killing themselves. We used to kind of think that in this country. I’ve often heard it said that someone (who was employed in a physically demanding job) had worked himself to an early death. So let’s suppose that were our cultural default, and we are looking at the same data as presented in this study. We would ‘know’ that exercise isn’t good for us, and, in fact, is probably bad, so we would look for different differences between the two groups. We might find that the group of runners drank way more water than the non-runners. And let’s assume for argument’s sake that our culture believes that water consumption is good and healthful. This same study data could then have been presented in terms of increased water consumption negated the negative effects of running, and even allowed runners to live longer than non-runners who drank less water. What we look for in these studies is culturally driven.
Were we in France – where I’ve yet to see anyone jogging – we might get a totally different take. We might have studies showing that men who have mistresses (and keep them well occupied) live longer than men who don’t. So therefore having a mistress leads to longevity…? (Probably not in my case, especially if it were discovered.)
Studies showing associations or correlations between between activity A and result B do not mean that A causes B. Yet it is extremely easy to be beguiled into believing that A causes B, especially if it makes sense based on our belief system. It doesn’t make logical or scientific sense to jump to that conclusion, although far too many people do.
I suspect that there are a number of people reading this post right now who firmly believe – despite all I have written above – that the study in question really does prove that running improves longevity because it seems to make so much sense. But let’s explore a little further.
We don’t know and can’t possibly determine from this study whether it is the running that increases longevity or whether there is some facet of personality or physiology that drives one to run that increases longevity. Maybe people who are destined to live longer take to running or other forms or aerobic exercise. People who are depressed typically have shorter lives, and people who are depressed tend not to join groups or exercise. Perhaps a number of the people in the control group are depressed, leading to an increase in early deaths in that group. It could be that the people who have the time to join a running group and spend the time running are more financially stable and are happier. Both of those conditions are correlated with longevity. There are far too many factors separating the two groups to dissect out the one and attach the benefit to it. But our culture firmly believes that exercise promotes longevity, so this experiment seems to bear that out, and most people accept it without looking to deeply.
One of the commenters on this blog gave a good example of this in a comment today. Let’s say we have a study showing that red cars are involved in more accidents than white cars (which may be true, for all I know.) Does this prove that red cars are more dangerous? Or does it prove that drivers who have the mindset that motivates them to choose red cars are less careful drivers? Or both? Or does it prove the opposite about white cars? You can’t tell.
If what I say is true, you make ask, why even do these kinds of studies if they have no value? The answer is that they do have value. They allow scientists to derive hypotheses that can then be tested. But often the rigid testing required for proof is impossible, so people who should know better fall back on these kinds of associative studies as proof since real proof can’t be obtained. Why not? Let’s take a look.
If we wanted to firmly establish that running increased longevity, we would have to do the following study. We would have to select a large group of subjects who are all of the same age and level of health. We would then have to randomize them into two groups: one a running group and the other the sedentary group. We would have to force the people in the running group to run and force the people in the sedentary group to be sedentary. Then we would have to follow these two groups for 21 years. If the runners then substantially outlived the non-runners, we might have a case that running promotes longevity. But until we do this, we can’t state that we have proof. You can see the immense difficulty involved in performing such a study. You can’t simply randomize someone who isn’t really driven to run into the group of runners and expect that person to run daily for 21 years. Yet that’s pretty much what you have to do to establish causality.
The moral of this story is to view these studies for what they are. Interesting but not proof.