Wednesday, July 30, 2014

Dosage Matters. If It Doesn't, Your Study is LIkely Wrong

A warning sign of a biased estimate of a treatment effect: no relationship between dosage and the outcome.
In this study highlighted in the New York Times, the authors conclude that five minutes of regular running has the same effect on health as 150 minutes.
This statistical red flag signals that runners, as a group, are different from non-runners in ways that have their own effect on health. In stats parlance: omitted variables bias.