The Intuition behind Instrumental Variables

This video provides an intuitive idea about Instrumental Variables (IV) regression.

The slides are freely available from my github page:

The video starts by critically analyzing the standard approach for presenting instrumental variables, using regression equations. Standard equations do not show the direction of causality, and therefore can lead to confusion.

Causal graphs (i.e. DAGs, Directed Acyclic Graphs) show the intuition behind IV in a much clearer way. Causal graphs show why you need to bring your own causal knowledge when applying IV. Bringing 2 pieces of causal knowledge (IV first stage, IV exclusion restriction) allows us to obtain a third piece of causal knowledge.

The video explains the limitations of a common definition of IV: "z is correlated with x but uncorrelated with u." The video also explains how RCTs (Randomized Controlled Trials) and RDD (Regression Discontinuity Design) can be understood in terms of an IV setup.