« January 2009 | Main | November 2009 »

August 10, 2009

Confounding

A student asked why direction of an association would reverse when you add a potential confounder. I returned this illustration.

View image

adjustment.GIF
In this hypothetical example, for both men and women eating more low calorie foods is associated with a lower BMI. However, if men eat more low cal food and have higher BMI then the combined association reverses.