Skip to content

Does racquetball save lives?

Asher Meir points to this news report and writes:

8e5 people in study, about half reported exercising, about half not. About 10% died overall. So overall death rate difference of 28% is pretty remarkable. It means about 3500 deaths instead of 4500 for a similar sample size.

But when you compare the rate of heart disease risk specifically (about 2% died of heart disease, or around 1000 in each sample), for runners vs. racket sports specifically (less than 10% each) you are really shooting in the dark. Say around 5000 people engaged in each kind of activity and around 100 died of heart disease in each group, sounds like normal variation.

Also they eliminated people who had heart disease at the beginning of the study, not sure why they would do this.

I guess the biggest issue is not controlling for endogeneity of activity, as the people who are frail and sickly are probably not engaging in much sports activity.

Not sure how much is author hype vs. journalist hype.

My reply: This reminds me of the “what does not kill me makes me stronger” principle. The elimination of people with risk at beginning of study, that’s interesting. I can see why it makes sense to do this and I can also see how it can cause bias. I guess the right way to do this is to express results conditional on initial risk.


Afterwards I realized the biggest weakness: They do not control for income. If you examine the results you see that the people who engage in the most expensive forms of exercise (e.g. racquetball) have the lowest morbidity. Could be just a proxy for income.

Anyway, a flawed study is better than no study, the next time they can try to control for income.

Yup. Gotta start somewhere. Make all your data and code available, don’t hype your claims, and we can go from there.

P.S. Meir also pointed me to this book, “The Lion in the Living Room: How House Cats Tamed Us and Took Over the World,” by Abigail Tucker. I haven’t read it—really I should spend less time online and more time reading books!—but it has a pretty good title, I’ll grant it that.


  1. Michael Chaiton says:

    It’s pretty standard practice to restrict the sample to those without disease at the beginning of a cohort study so that only people at risk of getting the disease are included. This study does it to control for temporality so that we can have some assurance that heart disease isn’t leading to physical activity reduction.

    Also, the study does control for education, which is arguably a better proxy than income for the SES and raquetball effect suggested.

  2. Bill Harris says:

    Just to set the record straight, isn’t that 8e4, not 8e5?

  3. Anoneuoid says:

    Dr. Haitham Ahmed, a preventive cardiologist at the Cleveland Clinic, said the most important finding of the study was that all exercise is good exercise.

    “If you played any sport, you had a 28% risk reduction of dying of any cause,” said Ahmed, who was not involved in the study. “Which means that exercise is good, no matter what you do and no matter which way you look at it.”

    It’s the same errors over and over and over…

    • I think a big part of the confusion here comes from taking a functional input-output approach where time is ignored. Clearly, people’s health and life-duration are not just numbers coming out of a static function, with playing a sport producing different numbers than not playing a sport.

      People grow through time. If they play a sport, it’s probably healthy for them for the most part, this healthy behavior leads to taking a certain trajectory, if they don’t play a sport, but are healthy for some other reason, they might have a similar trajectory, but if they have a health issue, then this issue leads them… to not play sports, and also to take a downward trajectory in health… Without a time-evolution view of what is going on, you get this confusion “does playing sports cause health, or does health cause people to be more likely to play sports?” Collapsing the whole history of this person’s life and activities down to a summary with 2 numbers. Since x(t) causes y(t+dt) and y(t) causes x(t+dt) the right answer to the summary question is both, neither, all of the above, maybe, sometimes, it depends.

      IF we tease these things apart using an appropriate time-series view, it’s easy to see how fitness comes from physical activity, one kind of which is playing sports… and also fitness, which also comes from age and genetics too, makes it easier to play a sport today. Injury is going to affect activity level, which affects fitness in a few weeks or months, and then that affects whether you can be playing sports in a year or two… and a good exercise regimen now enforced well, can increase fitness, leading to feeling better in a few months, leading to it being easier to imagine taking up a sport… it’s just insanity that an *essentially time-evolving thing* such as “health” is treated like a single number in leads to a single number out…

  4. Chris says:

    I didn’t know that racquetball was expensive.

Leave a Reply