7 · Instrumental variables
Identification when a regressor is endogenous
The final topic. When a regressor is correlated with the error term, no amount of controlling will fix it — you need outside variation. This topic introduces instrumental variables and two-stage least squares through the classic returns-to-schooling problem.
Slides
| Deck | Covers |
|---|---|
| 🖥 Instrumental variable | Endogeneity, instrument validity, two-stage least squares, returns to schooling |
Exercises
| Exercise | What you practise |
|---|---|
| 📋 AE12 Instrumental variable | Two-stage least squares from first principles on Angrist & Krueger’s data |
📦 Download the bundle: ae12-iv — includes the ak91.dta quarter-of-birth dataset and both sets of worked solutions.
WarningSolutions are in the bundle, not on the web
ae12-iv-solutions.Rmd and ae12-iv-solutions-2.Rmd are included in the downloadable bundle but have no rendered page here: their modelsummary call uses a coef_omit pattern that current versions of the package reject. Open them in RStudio to work through them.