Tuesday, February 10, 2015

Bad science journalism 101: ignoring confidence limits

You might have seen the news articles -- JOGGING CAN KILL! Well, I agree, but this article provides only sketchy evidence at best.


The basic idea is that they asked people already enrolled in a different study of heart disease incidence if and how much they jogged. Then they did a Cox proportional hazards model to look at the effects of jogging. The conclusion that was reported was that a little jogging is good (prevents deaths), but alot of jogging is bad.

The first thing that made me suspicious was breaking a continuous variable (hours of jogging per week) into categories. Then I went and looked at the results.

OK, I can see that joggers are less likely to die than sedentary non-joggers, but look at the confidence limits on the strenuous jogger group. No effect is included! The categorization has created a monster category, with relatively few people in it, and as a result the confidence limits have blown out. There is no way that they should be claiming a "U-shaped effect" of jogging on all-cause mortality. The journalists have grabbed onto something fictitious, admittedly with the avid participation of the scientists involved.

I am a non-jogger for religious reasons. But I hope my current exercise involving dog-walking, weightlifting, and yoga doesn't put me in the "sedentary non-jogger category"!

No comments:

Post a Comment