10 Identification and the causal hierarchy

 

This chapter covers

  • Motivating examples for identification
  • Using y0 for identification and deriving estimands
  • How to derive counterfactual graphs in y0
  • Deriving SWIGs for graph-based counterfactual identification

The practice of advancing machine learning often relies on a blind confidence that more data and the right architecture can solve any task. For tasks with causal elements, causal identification can make that less of a matter of faith and more of a science. It can tell us when more data won’t help, and what types of inductive biases are needed for the algorithm to work.

10.1 The causal hierarchy

10.1.1 Where questions and queries fall on the hierarchy

10.1.2 Where models and assumptions fall on the hierarchy

10.1.3 Where data falls on the hierarchy

10.1.4 The causal hierarchy theorem

10.2 Identification and the causal inference workflow

10.2.1 Defining identification

10.2.2 The causal inference workflow

10.2.3 Separating identification and estimation

10.3 Identification with backdoor adjustment

10.3.1 The backdoor adjustment formula

10.3.2 Demystifying the back door

10.4 Graphical identification with the do-calculus

10.4.1 Demystifying the do-calculus

10.5 Graphical identification algorithms

10.5.2 Demystifying the front door