Chap 12: Causal networks

“Correlation is not causation”

We say:

“Correlation is not causation.”

This is shorthand for a more meaningful statement,

“Data alone cannot demonstrate causation. Doing that takes knowledge about how the system you’re studying works.”

Correlation is causation

More precisely: When we see a genuine correlation between to variables, there must exist a causal path connecting them. That path can be any of the following

Confounding

How to deal with confounding: Block the path through C.

But don’t block mediators.