Skip to main content

Causal Inference Course

7 Instrumental variables

7.1 Experimental settings

Oct 17. Slides.

An instrumental variable (IV) identification strategy applies when a treatment effect of \(A\) on \(Y\) is confounded by unobserved variables (\(U\)), but an instrument \(Z\) creates random unconfounded variation in \(A\).

A very clean setting for IV is randomized experiments with non-compliance: an experimenter randomizes the assigned treatment (\(Z\)) but the actual treatment (\(A\)) may be unequal to \(Z\) because some units do not follow their assignment. Our first class will discuss this setting.

7.2 Lab: Instrumental Variable analysis in Code

In this lab we will implement instrumental variables estimators in R.

October 18 Slides. Download the R Markdown file here. Update: Solutions to the coding exercise are here.

7.3 Observational settings

Oct 19 Slides. After class, read Hernán and Robins 2020 Chapter 16.

On Thursday, we move on to IV analysis in observational settings. Here we focus on the casual assumptions required for IV. These assumptions often hold by design in experiments with non-compliance. In observational settings, they can be more doubtful.