The difference between confounding and effect modification
In: 1
(Statistics is hard to explain to a five year old. I’ll explain like you’re a fourteen year old.)
These are both examples of a **third** variable affecting a statistical association.
If you stratify your data by the confounding variable, the association disappears (there actually was **no** association in the first place!).
If you split up your data by a modifying variable, the effect is strengthened or weakened (there **is** an association in the first place, but you can change the strength of the association by accounting for a third modifying variable).
**I’ll give you an example!**
*People who drink their coffee black are more likely to get lung cancer*. You notice in your data that smokers seem to have disproportionately higher rates of lung cancer. If you pull out the smokers and only compare coffee drinkers who smoke, there is no difference in rates of lung cancer. Smoking **confounds** the results; there was actually no association between black coffee and lung cancer in the first place.
*Heavier people have more upper body strength*. This time you notice that men are disproportionately stronger than women. If you pull out men, you’ll see that the effect of weight is *greater* in men (who tend to have more lean mass). But heavier women are still stronger than lighter women–the effect is just less than in men. There truly is an effect of body weight on strength, but that effect is **modified** by the sex of the participant.
Latest Answers