Notes on "The Book of Why"

Confounder and Mediator (Chapter 4)

Control for a mediator (e.g., add to a regression model) could block the effect between the treatment and the outcome.

You should only control for $Z$ if you have a "strong prior reason" ... This "strong prior reason" is nothing more or less than a causal assumption.

Simpson's Paradox and Lord's Paradox (Chapter 6)

Change in the causal assumption should affect the way we formulate the effect between the treatment and the outcome.

Reference

• Pearl, Judea and Dana Mackenzie. The Book of Why: The New Science of Cause and Effect. Basic Books, 2018.