As requested, slower graphs! Also added a graph on collider bias, the webpage explanation helps there.
These graphs are intended to show what standard causal inference methods actually *do* to data, and how they work.
This is what controlling for a binary variable looks like:
These graphs are intended to show what standard causal inference methods actually *do* to data, and how they work.
This is what controlling for a binary variable looks like:
One version of matching:
Instrumental Variables
Fixed effects:
Difference-in-Difference
(a simplified) Regression Discontinuity
Inventing a correlation out of nowhere by controlling for a collider:
I'll probably end up pinning this version of the thread. Also, these are compiled on a website with more detailed explanations of what's going on, and causal diagrams: nickchk.com/causalgraphs.h…
Code is on GitHub at github.com/NickCH-K/causa…
